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 (swq_has_sleeper(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         return kvm_get_vcpu_by_id(kvm, id);
 396 }
 397
 398 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 399 {
 400         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 401         vpa->yield_count = cpu_to_be32(1);
 402 }
 403
 404 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 405                    unsigned long addr, unsigned long len)
 406 {
 407         /* check address is cacheline aligned */
 408         if (addr & (L1_CACHE_BYTES - 1))
 409                 return -EINVAL;
 410         spin_lock(&vcpu->arch.vpa_update_lock);
 411         if (v->next_gpa != addr || v->len != len) {
 412                 v->next_gpa = addr;
 413                 v->len = addr ? len : 0;
 414                 v->update_pending = 1;
 415         }
 416         spin_unlock(&vcpu->arch.vpa_update_lock);
 417         return 0;
 418 }
 419
 420 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 421 struct reg_vpa {
 422         u32 dummy;
 423         union {
 424                 __be16 hword;
 425                 __be32 word;
 426         } length;
 427 };
 428
 429 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 430 {
 431         if (vpap->update_pending)
 432                 return vpap->next_gpa != 0;
 433         return vpap->pinned_addr != NULL;
 434 }
 435
 436 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 437                                        unsigned long flags,
 438                                        unsigned long vcpuid, unsigned long vpa)
 439 {
 440         struct kvm *kvm = vcpu->kvm;
 441         unsigned long len, nb;
 442         void *va;
 443         struct kvm_vcpu *tvcpu;
 444         int err;
 445         int subfunc;
 446         struct kvmppc_vpa *vpap;
 447
 448         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 449         if (!tvcpu)
 450                 return H_PARAMETER;
 451
 452         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 453         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 454             subfunc == H_VPA_REG_SLB) {
 455                 /* Registering new area - address must be cache-line aligned */
 456                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 457                         return H_PARAMETER;
 458
 459                 /* convert logical addr to kernel addr and read length */
 460                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 461                 if (va == NULL)
 462                         return H_PARAMETER;
 463                 if (subfunc == H_VPA_REG_VPA)
 464                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 465                 else
 466                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 467                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 468
 469                 /* Check length */
 470                 if (len > nb || len < sizeof(struct reg_vpa))
 471                         return H_PARAMETER;
 472         } else {
 473                 vpa = 0;
 474                 len = 0;
 475         }
 476
 477         err = H_PARAMETER;
 478         vpap = NULL;
 479         spin_lock(&tvcpu->arch.vpa_update_lock);
 480
 481         switch (subfunc) {
 482         case H_VPA_REG_VPA:             /* register VPA */
 483                 /*
 484                  * The size of our lppaca is 1kB because of the way we align
 485                  * it for the guest to avoid crossing a 4kB boundary. We only
 486                  * use 640 bytes of the structure though, so we should accept
 487                  * clients that set a size of 640.
 488                  */
 489                 if (len < 640)
 490                         break;
 491                 vpap = &tvcpu->arch.vpa;
 492                 err = 0;
 493                 break;
 494
 495         case H_VPA_REG_DTL:             /* register DTL */
 496                 if (len < sizeof(struct dtl_entry))
 497                         break;
 498                 len -= len % sizeof(struct dtl_entry);
 499
 500                 /* Check that they have previously registered a VPA */
 501                 err = H_RESOURCE;
 502                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 503                         break;
 504
 505                 vpap = &tvcpu->arch.dtl;
 506                 err = 0;
 507                 break;
 508
 509         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 510                 /* Check that they have previously registered a VPA */
 511                 err = H_RESOURCE;
 512                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 513                         break;
 514
 515                 vpap = &tvcpu->arch.slb_shadow;
 516                 err = 0;
 517                 break;
 518
 519         case H_VPA_DEREG_VPA:           /* deregister VPA */
 520                 /* Check they don't still have a DTL or SLB buf registered */
 521                 err = H_RESOURCE;
 522                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 523                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 524                         break;
 525
 526                 vpap = &tvcpu->arch.vpa;
 527                 err = 0;
 528                 break;
 529
 530         case H_VPA_DEREG_DTL:           /* deregister DTL */
 531                 vpap = &tvcpu->arch.dtl;
 532                 err = 0;
 533                 break;
 534
 535         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 536                 vpap = &tvcpu->arch.slb_shadow;
 537                 err = 0;
 538                 break;
 539         }
 540
 541         if (vpap) {
 542                 vpap->next_gpa = vpa;
 543                 vpap->len = len;
 544                 vpap->update_pending = 1;
 545         }
 546
 547         spin_unlock(&tvcpu->arch.vpa_update_lock);
 548
 549         return err;
 550 }
 551
 552 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 553 {
 554         struct kvm *kvm = vcpu->kvm;
 555         void *va;
 556         unsigned long nb;
 557         unsigned long gpa;
 558
 559         /*
 560          * We need to pin the page pointed to by vpap->next_gpa,
 561          * but we can't call kvmppc_pin_guest_page under the lock
 562          * as it does get_user_pages() and down_read().  So we
 563          * have to drop the lock, pin the page, then get the lock
 564          * again and check that a new area didn't get registered
 565          * in the meantime.
 566          */
 567         for (;;) {
 568                 gpa = vpap->next_gpa;
 569                 spin_unlock(&vcpu->arch.vpa_update_lock);
 570                 va = NULL;
 571                 nb = 0;
 572                 if (gpa)
 573                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 574                 spin_lock(&vcpu->arch.vpa_update_lock);
 575                 if (gpa == vpap->next_gpa)
 576                         break;
 577                 /* sigh... unpin that one and try again */
 578                 if (va)
 579                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 580         }
 581
 582         vpap->update_pending = 0;
 583         if (va && nb < vpap->len) {
 584                 /*
 585                  * If it's now too short, it must be that userspace
 586                  * has changed the mappings underlying guest memory,
 587                  * so unregister the region.
 588                  */
 589                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 590                 va = NULL;
 591         }
 592         if (vpap->pinned_addr)
 593                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 594                                         vpap->dirty);
 595         vpap->gpa = gpa;
 596         vpap->pinned_addr = va;
 597         vpap->dirty = false;
 598         if (va)
 599                 vpap->pinned_end = va + vpap->len;
 600 }
 601
 602 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 603 {
 604         if (!(vcpu->arch.vpa.update_pending ||
 605               vcpu->arch.slb_shadow.update_pending ||
 606               vcpu->arch.dtl.update_pending))
 607                 return;
 608
 609         spin_lock(&vcpu->arch.vpa_update_lock);
 610         if (vcpu->arch.vpa.update_pending) {
 611                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 612                 if (vcpu->arch.vpa.pinned_addr)
 613                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 614         }
 615         if (vcpu->arch.dtl.update_pending) {
 616                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 617                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 618                 vcpu->arch.dtl_index = 0;
 619         }
 620         if (vcpu->arch.slb_shadow.update_pending)
 621                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 622         spin_unlock(&vcpu->arch.vpa_update_lock);
 623 }
 624
 625 /*
 626  * Return the accumulated stolen time for the vcore up until `now'.
 627  * The caller should hold the vcore lock.
 628  */
 629 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 630 {
 631         u64 p;
 632         unsigned long flags;
 633
 634         spin_lock_irqsave(&vc->stoltb_lock, flags);
 635         p = vc->stolen_tb;
 636         if (vc->vcore_state != VCORE_INACTIVE &&
 637             vc->preempt_tb != TB_NIL)
 638                 p += now - vc->preempt_tb;
 639         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 640         return p;
 641 }
 642
 643 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 644                                     struct kvmppc_vcore *vc)
 645 {
 646         struct dtl_entry *dt;
 647         struct lppaca *vpa;
 648         unsigned long stolen;
 649         unsigned long core_stolen;
 650         u64 now;
 651         unsigned long flags;
 652
 653         dt = vcpu->arch.dtl_ptr;
 654         vpa = vcpu->arch.vpa.pinned_addr;
 655         now = mftb();
 656         core_stolen = vcore_stolen_time(vc, now);
 657         stolen = core_stolen - vcpu->arch.stolen_logged;
 658         vcpu->arch.stolen_logged = core_stolen;
 659         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 660         stolen += vcpu->arch.busy_stolen;
 661         vcpu->arch.busy_stolen = 0;
 662         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 663         if (!dt || !vpa)
 664                 return;
 665         memset(dt, 0, sizeof(struct dtl_entry));
 666         dt->dispatch_reason = 7;
 667         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 668         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 669         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 670         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 671         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 672         ++dt;
 673         if (dt == vcpu->arch.dtl.pinned_end)
 674                 dt = vcpu->arch.dtl.pinned_addr;
 675         vcpu->arch.dtl_ptr = dt;
 676         /* order writing *dt vs. writing vpa->dtl_idx */
 677         smp_wmb();
 678         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 679         vcpu->arch.dtl.dirty = true;
 680 }
 681
 682 /* See if there is a doorbell interrupt pending for a vcpu */
 683 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
 684 {
 685         int thr;
 686         struct kvmppc_vcore *vc;
 687
 688         if (vcpu->arch.doorbell_request)
 689                 return true;
 690         /*
 691          * Ensure that the read of vcore->dpdes comes after the read
 692          * of vcpu->doorbell_request.  This barrier matches the
 693          * lwsync in book3s_hv_rmhandlers.S just before the
 694          * fast_guest_return label.
 695          */
 696         smp_rmb();
 697         vc = vcpu->arch.vcore;
 698         thr = vcpu->vcpu_id - vc->first_vcpuid;
 699         return !!(vc->dpdes & (1 << thr));
 700 }
 701
 702 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 703 {
 704         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 705                 return true;
 706         if ((!vcpu->arch.vcore->arch_compat) &&
 707             cpu_has_feature(CPU_FTR_ARCH_207S))
 708                 return true;
 709         return false;
 710 }
 711
 712 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 713                              unsigned long resource, unsigned long value1,
 714                              unsigned long value2)
 715 {
 716         switch (resource) {
 717         case H_SET_MODE_RESOURCE_SET_CIABR:
 718                 if (!kvmppc_power8_compatible(vcpu))
 719                         return H_P2;
 720                 if (value2)
 721                         return H_P4;
 722                 if (mflags)
 723                         return H_UNSUPPORTED_FLAG_START;
 724                 /* Guests can't breakpoint the hypervisor */
 725                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 726                         return H_P3;
 727                 vcpu->arch.ciabr  = value1;
 728                 return H_SUCCESS;
 729         case H_SET_MODE_RESOURCE_SET_DAWR:
 730                 if (!kvmppc_power8_compatible(vcpu))
 731                         return H_P2;
 732                 if (mflags)
 733                         return H_UNSUPPORTED_FLAG_START;
 734                 if (value2 & DABRX_HYP)
 735                         return H_P4;
 736                 vcpu->arch.dawr  = value1;
 737                 vcpu->arch.dawrx = value2;
 738                 return H_SUCCESS;
 739         default:
 740                 return H_TOO_HARD;
 741         }
 742 }
 743
 744 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 745 {
 746         struct kvmppc_vcore *vcore = target->arch.vcore;
 747
 748         /*
 749          * We expect to have been called by the real mode handler
 750          * (kvmppc_rm_h_confer()) which would have directly returned
 751          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 752          * have useful work to do and should not confer) so we don't
 753          * recheck that here.
 754          */
 755
 756         spin_lock(&vcore->lock);
 757         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 758             vcore->vcore_state != VCORE_INACTIVE &&
 759             vcore->runner)
 760                 target = vcore->runner;
 761         spin_unlock(&vcore->lock);
 762
 763         return kvm_vcpu_yield_to(target);
 764 }
 765
 766 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 767 {
 768         int yield_count = 0;
 769         struct lppaca *lppaca;
 770
 771         spin_lock(&vcpu->arch.vpa_update_lock);
 772         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 773         if (lppaca)
 774                 yield_count = be32_to_cpu(lppaca->yield_count);
 775         spin_unlock(&vcpu->arch.vpa_update_lock);
 776         return yield_count;
 777 }
 778
 779 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 780 {
 781         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 782         unsigned long target, ret = H_SUCCESS;
 783         int yield_count;
 784         struct kvm_vcpu *tvcpu;
 785         int idx, rc;
 786
 787         if (req <= MAX_HCALL_OPCODE &&
 788             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 789                 return RESUME_HOST;
 790
 791         switch (req) {
 792         case H_CEDE:
 793                 break;
 794         case H_PROD:
 795                 target = kvmppc_get_gpr(vcpu, 4);
 796                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 797                 if (!tvcpu) {
 798                         ret = H_PARAMETER;
 799                         break;
 800                 }
 801                 tvcpu->arch.prodded = 1;
 802                 smp_mb();
 803                 if (tvcpu->arch.ceded)
 804                         kvmppc_fast_vcpu_kick_hv(tvcpu);
 805                 break;
 806         case H_CONFER:
 807                 target = kvmppc_get_gpr(vcpu, 4);
 808                 if (target == -1)
 809                         break;
 810                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 811                 if (!tvcpu) {
 812                         ret = H_PARAMETER;
 813                         break;
 814                 }
 815                 yield_count = kvmppc_get_gpr(vcpu, 5);
 816                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 817                         break;
 818                 kvm_arch_vcpu_yield_to(tvcpu);
 819                 break;
 820         case H_REGISTER_VPA:
 821                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 822                                         kvmppc_get_gpr(vcpu, 5),
 823                                         kvmppc_get_gpr(vcpu, 6));
 824                 break;
 825         case H_RTAS:
 826                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 827                         return RESUME_HOST;
 828
 829                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 830                 rc = kvmppc_rtas_hcall(vcpu);
 831                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 832
 833                 if (rc == -ENOENT)
 834                         return RESUME_HOST;
 835                 else if (rc == 0)
 836                         break;
 837
 838                 /* Send the error out to userspace via KVM_RUN */
 839                 return rc;
 840         case H_LOGICAL_CI_LOAD:
 841                 ret = kvmppc_h_logical_ci_load(vcpu);
 842                 if (ret == H_TOO_HARD)
 843                         return RESUME_HOST;
 844                 break;
 845         case H_LOGICAL_CI_STORE:
 846                 ret = kvmppc_h_logical_ci_store(vcpu);
 847                 if (ret == H_TOO_HARD)
 848                         return RESUME_HOST;
 849                 break;
 850         case H_SET_MODE:
 851                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 852                                         kvmppc_get_gpr(vcpu, 5),
 853                                         kvmppc_get_gpr(vcpu, 6),
 854                                         kvmppc_get_gpr(vcpu, 7));
 855                 if (ret == H_TOO_HARD)
 856                         return RESUME_HOST;
 857                 break;
 858         case H_XIRR:
 859         case H_CPPR:
 860         case H_EOI:
 861         case H_IPI:
 862         case H_IPOLL:
 863         case H_XIRR_X:
 864                 if (kvmppc_xics_enabled(vcpu)) {
 865                         if (xive_enabled()) {
 866                                 ret = H_NOT_AVAILABLE;
 867                                 return RESUME_GUEST;
 868                         }
 869                         ret = kvmppc_xics_hcall(vcpu, req);
 870                         break;
 871                 }
 872                 return RESUME_HOST;
 873         case H_PUT_TCE:
 874                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
 875                                                 kvmppc_get_gpr(vcpu, 5),
 876                                                 kvmppc_get_gpr(vcpu, 6));
 877                 if (ret == H_TOO_HARD)
 878                         return RESUME_HOST;
 879                 break;
 880         case H_PUT_TCE_INDIRECT:
 881                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
 882                                                 kvmppc_get_gpr(vcpu, 5),
 883                                                 kvmppc_get_gpr(vcpu, 6),
 884                                                 kvmppc_get_gpr(vcpu, 7));
 885                 if (ret == H_TOO_HARD)
 886                         return RESUME_HOST;
 887                 break;
 888         case H_STUFF_TCE:
 889                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
 890                                                 kvmppc_get_gpr(vcpu, 5),
 891                                                 kvmppc_get_gpr(vcpu, 6),
 892                                                 kvmppc_get_gpr(vcpu, 7));
 893                 if (ret == H_TOO_HARD)
 894                         return RESUME_HOST;
 895                 break;
 896         default:
 897                 return RESUME_HOST;
 898         }
 899         kvmppc_set_gpr(vcpu, 3, ret);
 900         vcpu->arch.hcall_needed = 0;
 901         return RESUME_GUEST;
 902 }
 903
 904 static int kvmppc_hcall_impl_hv(unsigned long cmd)
 905 {
 906         switch (cmd) {
 907         case H_CEDE:
 908         case H_PROD:
 909         case H_CONFER:
 910         case H_REGISTER_VPA:
 911         case H_SET_MODE:
 912         case H_LOGICAL_CI_LOAD:
 913         case H_LOGICAL_CI_STORE:
 914 #ifdef CONFIG_KVM_XICS
 915         case H_XIRR:
 916         case H_CPPR:
 917         case H_EOI:
 918         case H_IPI:
 919         case H_IPOLL:
 920         case H_XIRR_X:
 921 #endif
 922                 return 1;
 923         }
 924
 925         /* See if it's in the real-mode table */
 926         return kvmppc_hcall_impl_hv_realmode(cmd);
 927 }
 928
 929 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
 930                                         struct kvm_vcpu *vcpu)
 931 {
 932         u32 last_inst;
 933
 934         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
 935                                         EMULATE_DONE) {
 936                 /*
 937                  * Fetch failed, so return to guest and
 938                  * try executing it again.
 939                  */
 940                 return RESUME_GUEST;
 941         }
 942
 943         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
 944                 run->exit_reason = KVM_EXIT_DEBUG;
 945                 run->debug.arch.address = kvmppc_get_pc(vcpu);
 946                 return RESUME_HOST;
 947         } else {
 948                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 949                 return RESUME_GUEST;
 950         }
 951 }
 952
 953 static void do_nothing(void *x)
 954 {
 955 }
 956
 957 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
 958 {
 959         int thr, cpu, pcpu, nthreads;
 960         struct kvm_vcpu *v;
 961         unsigned long dpdes;
 962
 963         nthreads = vcpu->kvm->arch.emul_smt_mode;
 964         dpdes = 0;
 965         cpu = vcpu->vcpu_id & ~(nthreads - 1);
 966         for (thr = 0; thr < nthreads; ++thr, ++cpu) {
 967                 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
 968                 if (!v)
 969                         continue;
 970                 /*
 971                  * If the vcpu is currently running on a physical cpu thread,
 972                  * interrupt it in order to pull it out of the guest briefly,
 973                  * which will update its vcore->dpdes value.
 974                  */
 975                 pcpu = READ_ONCE(v->cpu);
 976                 if (pcpu >= 0)
 977                         smp_call_function_single(pcpu, do_nothing, NULL, 1);
 978                 if (kvmppc_doorbell_pending(v))
 979                         dpdes |= 1 << thr;
 980         }
 981         return dpdes;
 982 }
 983
 984 /*
 985  * On POWER9, emulate doorbell-related instructions in order to
 986  * give the guest the illusion of running on a multi-threaded core.
 987  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
 988  * and mfspr DPDES.
 989  */
 990 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
 991 {
 992         u32 inst, rb, thr;
 993         unsigned long arg;
 994         struct kvm *kvm = vcpu->kvm;
 995         struct kvm_vcpu *tvcpu;
 996
 997         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
 998                 return RESUME_GUEST;
 999         if (get_op(inst) != 31)
1000                 return EMULATE_FAIL;
1001         rb = get_rb(inst);
1002         thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1003         switch (get_xop(inst)) {
1004         case OP_31_XOP_MSGSNDP:
1005                 arg = kvmppc_get_gpr(vcpu, rb);
1006                 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1007                         break;
1008                 arg &= 0x3f;
1009                 if (arg >= kvm->arch.emul_smt_mode)
1010                         break;
1011                 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1012                 if (!tvcpu)
1013                         break;
1014                 if (!tvcpu->arch.doorbell_request) {
1015                         tvcpu->arch.doorbell_request = 1;
1016                         kvmppc_fast_vcpu_kick_hv(tvcpu);
1017                 }
1018                 break;
1019         case OP_31_XOP_MSGCLRP:
1020                 arg = kvmppc_get_gpr(vcpu, rb);
1021                 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1022                         break;
1023                 vcpu->arch.vcore->dpdes = 0;
1024                 vcpu->arch.doorbell_request = 0;
1025                 break;
1026         case OP_31_XOP_MFSPR:
1027                 switch (get_sprn(inst)) {
1028                 case SPRN_TIR:
1029                         arg = thr;
1030                         break;
1031                 case SPRN_DPDES:
1032                         arg = kvmppc_read_dpdes(vcpu);
1033                         break;
1034                 default:
1035                         return EMULATE_FAIL;
1036                 }
1037                 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1038                 break;
1039         default:
1040                 return EMULATE_FAIL;
1041         }
1042         kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1043         return RESUME_GUEST;
1044 }
1045
1046 /* Called with vcpu->arch.vcore->lock held */
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                         /* Need vcore unlocked to call kvmppc_get_last_inst */
1167                         spin_unlock(&vcpu->arch.vcore->lock);
1168                         r = kvmppc_emulate_debug_inst(run, vcpu);
1169                         spin_lock(&vcpu->arch.vcore->lock);
1170                 } else {
1171                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1172                         r = RESUME_GUEST;
1173                 }
1174                 break;
1175         /*
1176          * This occurs if the guest (kernel or userspace), does something that
1177          * is prohibited by HFSCR.
1178          * On POWER9, this could be a doorbell instruction that we need
1179          * to emulate.
1180          * Otherwise, we just generate a program interrupt to the guest.
1181          */
1182         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1183                 r = EMULATE_FAIL;
1184                 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1185                     cpu_has_feature(CPU_FTR_ARCH_300)) {
1186                         /* Need vcore unlocked to call kvmppc_get_last_inst */
1187                         spin_unlock(&vcpu->arch.vcore->lock);
1188                         r = kvmppc_emulate_doorbell_instr(vcpu);
1189                         spin_lock(&vcpu->arch.vcore->lock);
1190                 }
1191                 if (r == EMULATE_FAIL) {
1192                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1193                         r = RESUME_GUEST;
1194                 }
1195                 break;
1196         case BOOK3S_INTERRUPT_HV_RM_HARD:
1197                 r = RESUME_PASSTHROUGH;
1198                 break;
1199         default:
1200                 kvmppc_dump_regs(vcpu);
1201                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1202                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1203                         vcpu->arch.shregs.msr);
1204                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1205                 r = RESUME_HOST;
1206                 break;
1207         }
1208
1209         return r;
1210 }
1211
1212 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1213                                             struct kvm_sregs *sregs)
1214 {
1215         int i;
1216
1217         memset(sregs, 0, sizeof(struct kvm_sregs));
1218         sregs->pvr = vcpu->arch.pvr;
1219         for (i = 0; i < vcpu->arch.slb_max; i++) {
1220                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1221                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1222         }
1223
1224         return 0;
1225 }
1226
1227 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1228                                             struct kvm_sregs *sregs)
1229 {
1230         int i, j;
1231
1232         /* Only accept the same PVR as the host's, since we can't spoof it */
1233         if (sregs->pvr != vcpu->arch.pvr)
1234                 return -EINVAL;
1235
1236         j = 0;
1237         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1238                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1239                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1240                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1241                         ++j;
1242                 }
1243         }
1244         vcpu->arch.slb_max = j;
1245
1246         return 0;
1247 }
1248
1249 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1250                 bool preserve_top32)
1251 {
1252         struct kvm *kvm = vcpu->kvm;
1253         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1254         u64 mask;
1255
1256         spin_lock(&vc->lock);
1257         /*
1258          * If ILE (interrupt little-endian) has changed, update the
1259          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1260          */
1261         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1262                 struct kvm_vcpu *vcpu;
1263                 int i;
1264
1265                 kvm_for_each_vcpu(i, vcpu, kvm) {
1266                         if (vcpu->arch.vcore != vc)
1267                                 continue;
1268                         if (new_lpcr & LPCR_ILE)
1269                                 vcpu->arch.intr_msr |= MSR_LE;
1270                         else
1271                                 vcpu->arch.intr_msr &= ~MSR_LE;
1272                 }
1273         }
1274
1275         /*
1276          * Userspace can only modify DPFD (default prefetch depth),
1277          * ILE (interrupt little-endian) and TC (translation control).
1278          * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1279          */
1280         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1281         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1282                 mask |= LPCR_AIL;
1283         /*
1284          * On POWER9, allow userspace to enable large decrementer for the
1285          * guest, whether or not the host has it enabled.
1286          */
1287         if (cpu_has_feature(CPU_FTR_ARCH_300))
1288                 mask |= LPCR_LD;
1289
1290         /* Broken 32-bit version of LPCR must not clear top bits */
1291         if (preserve_top32)
1292                 mask &= 0xFFFFFFFF;
1293         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1294         spin_unlock(&vc->lock);
1295 }
1296
1297 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1298                                  union kvmppc_one_reg *val)
1299 {
1300         int r = 0;
1301         long int i;
1302
1303         switch (id) {
1304         case KVM_REG_PPC_DEBUG_INST:
1305                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1306                 break;
1307         case KVM_REG_PPC_HIOR:
1308                 *val = get_reg_val(id, 0);
1309                 break;
1310         case KVM_REG_PPC_DABR:
1311                 *val = get_reg_val(id, vcpu->arch.dabr);
1312                 break;
1313         case KVM_REG_PPC_DABRX:
1314                 *val = get_reg_val(id, vcpu->arch.dabrx);
1315                 break;
1316         case KVM_REG_PPC_DSCR:
1317                 *val = get_reg_val(id, vcpu->arch.dscr);
1318                 break;
1319         case KVM_REG_PPC_PURR:
1320                 *val = get_reg_val(id, vcpu->arch.purr);
1321                 break;
1322         case KVM_REG_PPC_SPURR:
1323                 *val = get_reg_val(id, vcpu->arch.spurr);
1324                 break;
1325         case KVM_REG_PPC_AMR:
1326                 *val = get_reg_val(id, vcpu->arch.amr);
1327                 break;
1328         case KVM_REG_PPC_UAMOR:
1329                 *val = get_reg_val(id, vcpu->arch.uamor);
1330                 break;
1331         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1332                 i = id - KVM_REG_PPC_MMCR0;
1333                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1334                 break;
1335         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1336                 i = id - KVM_REG_PPC_PMC1;
1337                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1338                 break;
1339         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1340                 i = id - KVM_REG_PPC_SPMC1;
1341                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1342                 break;
1343         case KVM_REG_PPC_SIAR:
1344                 *val = get_reg_val(id, vcpu->arch.siar);
1345                 break;
1346         case KVM_REG_PPC_SDAR:
1347                 *val = get_reg_val(id, vcpu->arch.sdar);
1348                 break;
1349         case KVM_REG_PPC_SIER:
1350                 *val = get_reg_val(id, vcpu->arch.sier);
1351                 break;
1352         case KVM_REG_PPC_IAMR:
1353                 *val = get_reg_val(id, vcpu->arch.iamr);
1354                 break;
1355         case KVM_REG_PPC_PSPB:
1356                 *val = get_reg_val(id, vcpu->arch.pspb);
1357                 break;
1358         case KVM_REG_PPC_DPDES:
1359                 /*
1360                  * On POWER9, where we are emulating msgsndp etc.,
1361                  * we return 1 bit for each vcpu, which can come from
1362                  * either vcore->dpdes or doorbell_request.
1363                  * On POWER8, doorbell_request is 0.
1364                  */
1365                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes |
1366                                    vcpu->arch.doorbell_request);
1367                 break;
1368         case KVM_REG_PPC_VTB:
1369                 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1370                 break;
1371         case KVM_REG_PPC_DAWR:
1372                 *val = get_reg_val(id, vcpu->arch.dawr);
1373                 break;
1374         case KVM_REG_PPC_DAWRX:
1375                 *val = get_reg_val(id, vcpu->arch.dawrx);
1376                 break;
1377         case KVM_REG_PPC_CIABR:
1378                 *val = get_reg_val(id, vcpu->arch.ciabr);
1379                 break;
1380         case KVM_REG_PPC_CSIGR:
1381                 *val = get_reg_val(id, vcpu->arch.csigr);
1382                 break;
1383         case KVM_REG_PPC_TACR:
1384                 *val = get_reg_val(id, vcpu->arch.tacr);
1385                 break;
1386         case KVM_REG_PPC_TCSCR:
1387                 *val = get_reg_val(id, vcpu->arch.tcscr);
1388                 break;
1389         case KVM_REG_PPC_PID:
1390                 *val = get_reg_val(id, vcpu->arch.pid);
1391                 break;
1392         case KVM_REG_PPC_ACOP:
1393                 *val = get_reg_val(id, vcpu->arch.acop);
1394                 break;
1395         case KVM_REG_PPC_WORT:
1396                 *val = get_reg_val(id, vcpu->arch.wort);
1397                 break;
1398         case KVM_REG_PPC_TIDR:
1399                 *val = get_reg_val(id, vcpu->arch.tid);
1400                 break;
1401         case KVM_REG_PPC_PSSCR:
1402                 *val = get_reg_val(id, vcpu->arch.psscr);
1403                 break;
1404         case KVM_REG_PPC_VPA_ADDR:
1405                 spin_lock(&vcpu->arch.vpa_update_lock);
1406                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1407                 spin_unlock(&vcpu->arch.vpa_update_lock);
1408                 break;
1409         case KVM_REG_PPC_VPA_SLB:
1410                 spin_lock(&vcpu->arch.vpa_update_lock);
1411                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1412                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1413                 spin_unlock(&vcpu->arch.vpa_update_lock);
1414                 break;
1415         case KVM_REG_PPC_VPA_DTL:
1416                 spin_lock(&vcpu->arch.vpa_update_lock);
1417                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1418                 val->vpaval.length = vcpu->arch.dtl.len;
1419                 spin_unlock(&vcpu->arch.vpa_update_lock);
1420                 break;
1421         case KVM_REG_PPC_TB_OFFSET:
1422                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1423                 break;
1424         case KVM_REG_PPC_LPCR:
1425         case KVM_REG_PPC_LPCR_64:
1426                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1427                 break;
1428         case KVM_REG_PPC_PPR:
1429                 *val = get_reg_val(id, vcpu->arch.ppr);
1430                 break;
1431 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1432         case KVM_REG_PPC_TFHAR:
1433                 *val = get_reg_val(id, vcpu->arch.tfhar);
1434                 break;
1435         case KVM_REG_PPC_TFIAR:
1436                 *val = get_reg_val(id, vcpu->arch.tfiar);
1437                 break;
1438         case KVM_REG_PPC_TEXASR:
1439                 *val = get_reg_val(id, vcpu->arch.texasr);
1440                 break;
1441         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1442                 i = id - KVM_REG_PPC_TM_GPR0;
1443                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1444                 break;
1445         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1446         {
1447                 int j;
1448                 i = id - KVM_REG_PPC_TM_VSR0;
1449                 if (i < 32)
1450                         for (j = 0; j < TS_FPRWIDTH; j++)
1451                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1452                 else {
1453                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1454                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1455                         else
1456                                 r = -ENXIO;
1457                 }
1458                 break;
1459         }
1460         case KVM_REG_PPC_TM_CR:
1461                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1462                 break;
1463         case KVM_REG_PPC_TM_XER:
1464                 *val = get_reg_val(id, vcpu->arch.xer_tm);
1465                 break;
1466         case KVM_REG_PPC_TM_LR:
1467                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1468                 break;
1469         case KVM_REG_PPC_TM_CTR:
1470                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1471                 break;
1472         case KVM_REG_PPC_TM_FPSCR:
1473                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1474                 break;
1475         case KVM_REG_PPC_TM_AMR:
1476                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1477                 break;
1478         case KVM_REG_PPC_TM_PPR:
1479                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1480                 break;
1481         case KVM_REG_PPC_TM_VRSAVE:
1482                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1483                 break;
1484         case KVM_REG_PPC_TM_VSCR:
1485                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1486                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1487                 else
1488                         r = -ENXIO;
1489                 break;
1490         case KVM_REG_PPC_TM_DSCR:
1491                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1492                 break;
1493         case KVM_REG_PPC_TM_TAR:
1494                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1495                 break;
1496 #endif
1497         case KVM_REG_PPC_ARCH_COMPAT:
1498                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1499                 break;
1500         case KVM_REG_PPC_DEC_EXPIRY:
1501                 *val = get_reg_val(id, vcpu->arch.dec_expires +
1502                                    vcpu->arch.vcore->tb_offset);
1503                 break;
1504         default:
1505                 r = -EINVAL;
1506                 break;
1507         }
1508
1509         return r;
1510 }
1511
1512 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1513                                  union kvmppc_one_reg *val)
1514 {
1515         int r = 0;
1516         long int i;
1517         unsigned long addr, len;
1518
1519         switch (id) {
1520         case KVM_REG_PPC_HIOR:
1521                 /* Only allow this to be set to zero */
1522                 if (set_reg_val(id, *val))
1523                         r = -EINVAL;
1524                 break;
1525         case KVM_REG_PPC_DABR:
1526                 vcpu->arch.dabr = set_reg_val(id, *val);
1527                 break;
1528         case KVM_REG_PPC_DABRX:
1529                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1530                 break;
1531         case KVM_REG_PPC_DSCR:
1532                 vcpu->arch.dscr = set_reg_val(id, *val);
1533                 break;
1534         case KVM_REG_PPC_PURR:
1535                 vcpu->arch.purr = set_reg_val(id, *val);
1536                 break;
1537         case KVM_REG_PPC_SPURR:
1538                 vcpu->arch.spurr = set_reg_val(id, *val);
1539                 break;
1540         case KVM_REG_PPC_AMR:
1541                 vcpu->arch.amr = set_reg_val(id, *val);
1542                 break;
1543         case KVM_REG_PPC_UAMOR:
1544                 vcpu->arch.uamor = set_reg_val(id, *val);
1545                 break;
1546         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1547                 i = id - KVM_REG_PPC_MMCR0;
1548                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1549                 break;
1550         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1551                 i = id - KVM_REG_PPC_PMC1;
1552                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1553                 break;
1554         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1555                 i = id - KVM_REG_PPC_SPMC1;
1556                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1557                 break;
1558         case KVM_REG_PPC_SIAR:
1559                 vcpu->arch.siar = set_reg_val(id, *val);
1560                 break;
1561         case KVM_REG_PPC_SDAR:
1562                 vcpu->arch.sdar = set_reg_val(id, *val);
1563                 break;
1564         case KVM_REG_PPC_SIER:
1565                 vcpu->arch.sier = set_reg_val(id, *val);
1566                 break;
1567         case KVM_REG_PPC_IAMR:
1568                 vcpu->arch.iamr = set_reg_val(id, *val);
1569                 break;
1570         case KVM_REG_PPC_PSPB:
1571                 vcpu->arch.pspb = set_reg_val(id, *val);
1572                 break;
1573         case KVM_REG_PPC_DPDES:
1574                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1575                 break;
1576         case KVM_REG_PPC_VTB:
1577                 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1578                 break;
1579         case KVM_REG_PPC_DAWR:
1580                 vcpu->arch.dawr = set_reg_val(id, *val);
1581                 break;
1582         case KVM_REG_PPC_DAWRX:
1583                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1584                 break;
1585         case KVM_REG_PPC_CIABR:
1586                 vcpu->arch.ciabr = set_reg_val(id, *val);
1587                 /* Don't allow setting breakpoints in hypervisor code */
1588                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1589                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1590                 break;
1591         case KVM_REG_PPC_CSIGR:
1592                 vcpu->arch.csigr = set_reg_val(id, *val);
1593                 break;
1594         case KVM_REG_PPC_TACR:
1595                 vcpu->arch.tacr = set_reg_val(id, *val);
1596                 break;
1597         case KVM_REG_PPC_TCSCR:
1598                 vcpu->arch.tcscr = set_reg_val(id, *val);
1599                 break;
1600         case KVM_REG_PPC_PID:
1601                 vcpu->arch.pid = set_reg_val(id, *val);
1602                 break;
1603         case KVM_REG_PPC_ACOP:
1604                 vcpu->arch.acop = set_reg_val(id, *val);
1605                 break;
1606         case KVM_REG_PPC_WORT:
1607                 vcpu->arch.wort = set_reg_val(id, *val);
1608                 break;
1609         case KVM_REG_PPC_TIDR:
1610                 vcpu->arch.tid = set_reg_val(id, *val);
1611                 break;
1612         case KVM_REG_PPC_PSSCR:
1613                 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1614                 break;
1615         case KVM_REG_PPC_VPA_ADDR:
1616                 addr = set_reg_val(id, *val);
1617                 r = -EINVAL;
1618                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1619                               vcpu->arch.dtl.next_gpa))
1620                         break;
1621                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1622                 break;
1623         case KVM_REG_PPC_VPA_SLB:
1624                 addr = val->vpaval.addr;
1625                 len = val->vpaval.length;
1626                 r = -EINVAL;
1627                 if (addr && !vcpu->arch.vpa.next_gpa)
1628                         break;
1629                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1630                 break;
1631         case KVM_REG_PPC_VPA_DTL:
1632                 addr = val->vpaval.addr;
1633                 len = val->vpaval.length;
1634                 r = -EINVAL;
1635                 if (addr && (len < sizeof(struct dtl_entry) ||
1636                              !vcpu->arch.vpa.next_gpa))
1637                         break;
1638                 len -= len % sizeof(struct dtl_entry);
1639                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1640                 break;
1641         case KVM_REG_PPC_TB_OFFSET:
1642                 /*
1643                  * POWER9 DD1 has an erratum where writing TBU40 causes
1644                  * the timebase to lose ticks.  So we don't let the
1645                  * timebase offset be changed on P9 DD1.  (It is
1646                  * initialized to zero.)
1647                  */
1648                 if (cpu_has_feature(CPU_FTR_POWER9_DD1))
1649                         break;
1650                 /* round up to multiple of 2^24 */
1651                 vcpu->arch.vcore->tb_offset =
1652                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1653                 break;
1654         case KVM_REG_PPC_LPCR:
1655                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1656                 break;
1657         case KVM_REG_PPC_LPCR_64:
1658                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1659                 break;
1660         case KVM_REG_PPC_PPR:
1661                 vcpu->arch.ppr = set_reg_val(id, *val);
1662                 break;
1663 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1664         case KVM_REG_PPC_TFHAR:
1665                 vcpu->arch.tfhar = set_reg_val(id, *val);
1666                 break;
1667         case KVM_REG_PPC_TFIAR:
1668                 vcpu->arch.tfiar = set_reg_val(id, *val);
1669                 break;
1670         case KVM_REG_PPC_TEXASR:
1671                 vcpu->arch.texasr = set_reg_val(id, *val);
1672                 break;
1673         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1674                 i = id - KVM_REG_PPC_TM_GPR0;
1675                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1676                 break;
1677         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1678         {
1679                 int j;
1680                 i = id - KVM_REG_PPC_TM_VSR0;
1681                 if (i < 32)
1682                         for (j = 0; j < TS_FPRWIDTH; j++)
1683                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1684                 else
1685                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1686                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1687                         else
1688                                 r = -ENXIO;
1689                 break;
1690         }
1691         case KVM_REG_PPC_TM_CR:
1692                 vcpu->arch.cr_tm = set_reg_val(id, *val);
1693                 break;
1694         case KVM_REG_PPC_TM_XER:
1695                 vcpu->arch.xer_tm = set_reg_val(id, *val);
1696                 break;
1697         case KVM_REG_PPC_TM_LR:
1698                 vcpu->arch.lr_tm = set_reg_val(id, *val);
1699                 break;
1700         case KVM_REG_PPC_TM_CTR:
1701                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1702                 break;
1703         case KVM_REG_PPC_TM_FPSCR:
1704                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1705                 break;
1706         case KVM_REG_PPC_TM_AMR:
1707                 vcpu->arch.amr_tm = set_reg_val(id, *val);
1708                 break;
1709         case KVM_REG_PPC_TM_PPR:
1710                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1711                 break;
1712         case KVM_REG_PPC_TM_VRSAVE:
1713                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1714                 break;
1715         case KVM_REG_PPC_TM_VSCR:
1716                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1717                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1718                 else
1719                         r = - ENXIO;
1720                 break;
1721         case KVM_REG_PPC_TM_DSCR:
1722                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1723                 break;
1724         case KVM_REG_PPC_TM_TAR:
1725                 vcpu->arch.tar_tm = set_reg_val(id, *val);
1726                 break;
1727 #endif
1728         case KVM_REG_PPC_ARCH_COMPAT:
1729                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1730                 break;
1731         case KVM_REG_PPC_DEC_EXPIRY:
1732                 vcpu->arch.dec_expires = set_reg_val(id, *val) -
1733                         vcpu->arch.vcore->tb_offset;
1734                 break;
1735         default:
1736                 r = -EINVAL;
1737                 break;
1738         }
1739
1740         return r;
1741 }
1742
1743 /*
1744  * On POWER9, threads are independent and can be in different partitions.
1745  * Therefore we consider each thread to be a subcore.
1746  * There is a restriction that all threads have to be in the same
1747  * MMU mode (radix or HPT), unfortunately, but since we only support
1748  * HPT guests on a HPT host so far, that isn't an impediment yet.
1749  */
1750 static int threads_per_vcore(void)
1751 {
1752         if (cpu_has_feature(CPU_FTR_ARCH_300))
1753                 return 1;
1754         return threads_per_subcore;
1755 }
1756
1757 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1758 {
1759         struct kvmppc_vcore *vcore;
1760
1761         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1762
1763         if (vcore == NULL)
1764                 return NULL;
1765
1766         spin_lock_init(&vcore->lock);
1767         spin_lock_init(&vcore->stoltb_lock);
1768         init_swait_queue_head(&vcore->wq);
1769         vcore->preempt_tb = TB_NIL;
1770         vcore->lpcr = kvm->arch.lpcr;
1771         vcore->first_vcpuid = core * kvm->arch.smt_mode;
1772         vcore->kvm = kvm;
1773         INIT_LIST_HEAD(&vcore->preempt_list);
1774
1775         return vcore;
1776 }
1777
1778 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1779 static struct debugfs_timings_element {
1780         const char *name;
1781         size_t offset;
1782 } timings[] = {
1783         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
1784         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
1785         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
1786         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
1787         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
1788 };
1789
1790 #define N_TIMINGS       (sizeof(timings) / sizeof(timings[0]))
1791
1792 struct debugfs_timings_state {
1793         struct kvm_vcpu *vcpu;
1794         unsigned int    buflen;
1795         char            buf[N_TIMINGS * 100];
1796 };
1797
1798 static int debugfs_timings_open(struct inode *inode, struct file *file)
1799 {
1800         struct kvm_vcpu *vcpu = inode->i_private;
1801         struct debugfs_timings_state *p;
1802
1803         p = kzalloc(sizeof(*p), GFP_KERNEL);
1804         if (!p)
1805                 return -ENOMEM;
1806
1807         kvm_get_kvm(vcpu->kvm);
1808         p->vcpu = vcpu;
1809         file->private_data = p;
1810
1811         return nonseekable_open(inode, file);
1812 }
1813
1814 static int debugfs_timings_release(struct inode *inode, struct file *file)
1815 {
1816         struct debugfs_timings_state *p = file->private_data;
1817
1818         kvm_put_kvm(p->vcpu->kvm);
1819         kfree(p);
1820         return 0;
1821 }
1822
1823 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1824                                     size_t len, loff_t *ppos)
1825 {
1826         struct debugfs_timings_state *p = file->private_data;
1827         struct kvm_vcpu *vcpu = p->vcpu;
1828         char *s, *buf_end;
1829         struct kvmhv_tb_accumulator tb;
1830         u64 count;
1831         loff_t pos;
1832         ssize_t n;
1833         int i, loops;
1834         bool ok;
1835
1836         if (!p->buflen) {
1837                 s = p->buf;
1838                 buf_end = s + sizeof(p->buf);
1839                 for (i = 0; i < N_TIMINGS; ++i) {
1840                         struct kvmhv_tb_accumulator *acc;
1841
1842                         acc = (struct kvmhv_tb_accumulator *)
1843                                 ((unsigned long)vcpu + timings[i].offset);
1844                         ok = false;
1845                         for (loops = 0; loops < 1000; ++loops) {
1846                                 count = acc->seqcount;
1847                                 if (!(count & 1)) {
1848                                         smp_rmb();
1849                                         tb = *acc;
1850                                         smp_rmb();
1851                                         if (count == acc->seqcount) {
1852                                                 ok = true;
1853                                                 break;
1854                                         }
1855                                 }
1856                                 udelay(1);
1857                         }
1858                         if (!ok)
1859                                 snprintf(s, buf_end - s, "%s: stuck\n",
1860                                         timings[i].name);
1861                         else
1862                                 snprintf(s, buf_end - s,
1863                                         "%s: %llu %llu %llu %llu\n",
1864                                         timings[i].name, count / 2,
1865                                         tb_to_ns(tb.tb_total),
1866                                         tb_to_ns(tb.tb_min),
1867                                         tb_to_ns(tb.tb_max));
1868                         s += strlen(s);
1869                 }
1870                 p->buflen = s - p->buf;
1871         }
1872
1873         pos = *ppos;
1874         if (pos >= p->buflen)
1875                 return 0;
1876         if (len > p->buflen - pos)
1877                 len = p->buflen - pos;
1878         n = copy_to_user(buf, p->buf + pos, len);
1879         if (n) {
1880                 if (n == len)
1881                         return -EFAULT;
1882                 len -= n;
1883         }
1884         *ppos = pos + len;
1885         return len;
1886 }
1887
1888 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1889                                      size_t len, loff_t *ppos)
1890 {
1891         return -EACCES;
1892 }
1893
1894 static const struct file_operations debugfs_timings_ops = {
1895         .owner   = THIS_MODULE,
1896         .open    = debugfs_timings_open,
1897         .release = debugfs_timings_release,
1898         .read    = debugfs_timings_read,
1899         .write   = debugfs_timings_write,
1900         .llseek  = generic_file_llseek,
1901 };
1902
1903 /* Create a debugfs directory for the vcpu */
1904 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1905 {
1906         char buf[16];
1907         struct kvm *kvm = vcpu->kvm;
1908
1909         snprintf(buf, sizeof(buf), "vcpu%u", id);
1910         if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1911                 return;
1912         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1913         if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1914                 return;
1915         vcpu->arch.debugfs_timings =
1916                 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1917                                     vcpu, &debugfs_timings_ops);
1918 }
1919
1920 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1921 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1922 {
1923 }
1924 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1925
1926 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1927                                                    unsigned int id)
1928 {
1929         struct kvm_vcpu *vcpu;
1930         int err;
1931         int core;
1932         struct kvmppc_vcore *vcore;
1933
1934         err = -ENOMEM;
1935         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1936         if (!vcpu)
1937                 goto out;
1938
1939         err = kvm_vcpu_init(vcpu, kvm, id);
1940         if (err)
1941                 goto free_vcpu;
1942
1943         vcpu->arch.shared = &vcpu->arch.shregs;
1944 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1945         /*
1946          * The shared struct is never shared on HV,
1947          * so we can always use host endianness
1948          */
1949 #ifdef __BIG_ENDIAN__
1950         vcpu->arch.shared_big_endian = true;
1951 #else
1952         vcpu->arch.shared_big_endian = false;
1953 #endif
1954 #endif
1955         vcpu->arch.mmcr[0] = MMCR0_FC;
1956         vcpu->arch.ctrl = CTRL_RUNLATCH;
1957         /* default to host PVR, since we can't spoof it */
1958         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1959         spin_lock_init(&vcpu->arch.vpa_update_lock);
1960         spin_lock_init(&vcpu->arch.tbacct_lock);
1961         vcpu->arch.busy_preempt = TB_NIL;
1962         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1963
1964         /*
1965          * Set the default HFSCR for the guest from the host value.
1966          * This value is only used on POWER9.
1967          * On POWER9 DD1, TM doesn't work, so we make sure to
1968          * prevent the guest from using it.
1969          * On POWER9, we want to virtualize the doorbell facility, so we
1970          * turn off the HFSCR bit, which causes those instructions to trap.
1971          */
1972         vcpu->arch.hfscr = mfspr(SPRN_HFSCR);
1973         if (!cpu_has_feature(CPU_FTR_TM))
1974                 vcpu->arch.hfscr &= ~HFSCR_TM;
1975         if (cpu_has_feature(CPU_FTR_ARCH_300))
1976                 vcpu->arch.hfscr &= ~HFSCR_MSGP;
1977
1978         kvmppc_mmu_book3s_hv_init(vcpu);
1979
1980         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1981
1982         init_waitqueue_head(&vcpu->arch.cpu_run);
1983
1984         mutex_lock(&kvm->lock);
1985         vcore = NULL;
1986         err = -EINVAL;
1987         core = id / kvm->arch.smt_mode;
1988         if (core < KVM_MAX_VCORES) {
1989                 vcore = kvm->arch.vcores[core];
1990                 if (!vcore) {
1991                         err = -ENOMEM;
1992                         vcore = kvmppc_vcore_create(kvm, core);
1993                         kvm->arch.vcores[core] = vcore;
1994                         kvm->arch.online_vcores++;
1995                 }
1996         }
1997         mutex_unlock(&kvm->lock);
1998
1999         if (!vcore)
2000                 goto uninit_vcpu;
2001
2002         spin_lock(&vcore->lock);
2003         ++vcore->num_threads;
2004         spin_unlock(&vcore->lock);
2005         vcpu->arch.vcore = vcore;
2006         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2007         vcpu->arch.thread_cpu = -1;
2008         vcpu->arch.prev_cpu = -1;
2009
2010         vcpu->arch.cpu_type = KVM_CPU_3S_64;
2011         kvmppc_sanity_check(vcpu);
2012
2013         debugfs_vcpu_init(vcpu, id);
2014
2015         return vcpu;
2016
2017 uninit_vcpu:
2018         kvm_vcpu_uninit(vcpu);
2019 free_vcpu:
2020         kmem_cache_free(kvm_vcpu_cache, vcpu);
2021 out:
2022         return ERR_PTR(err);
2023 }
2024
2025 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2026                               unsigned long flags)
2027 {
2028         int err;
2029         int esmt = 0;
2030
2031         if (flags)
2032                 return -EINVAL;
2033         if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2034                 return -EINVAL;
2035         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2036                 /*
2037                  * On POWER8 (or POWER7), the threading mode is "strict",
2038                  * so we pack smt_mode vcpus per vcore.
2039                  */
2040                 if (smt_mode > threads_per_subcore)
2041                         return -EINVAL;
2042         } else {
2043                 /*
2044                  * On POWER9, the threading mode is "loose",
2045                  * so each vcpu gets its own vcore.
2046                  */
2047                 esmt = smt_mode;
2048                 smt_mode = 1;
2049         }
2050         mutex_lock(&kvm->lock);
2051         err = -EBUSY;
2052         if (!kvm->arch.online_vcores) {
2053                 kvm->arch.smt_mode = smt_mode;
2054                 kvm->arch.emul_smt_mode = esmt;
2055                 err = 0;
2056         }
2057         mutex_unlock(&kvm->lock);
2058
2059         return err;
2060 }
2061
2062 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2063 {
2064         if (vpa->pinned_addr)
2065                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2066                                         vpa->dirty);
2067 }
2068
2069 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2070 {
2071         spin_lock(&vcpu->arch.vpa_update_lock);
2072         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2073         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2074         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2075         spin_unlock(&vcpu->arch.vpa_update_lock);
2076         kvm_vcpu_uninit(vcpu);
2077         kmem_cache_free(kvm_vcpu_cache, vcpu);
2078 }
2079
2080 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2081 {
2082         /* Indicate we want to get back into the guest */
2083         return 1;
2084 }
2085
2086 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2087 {
2088         unsigned long dec_nsec, now;
2089
2090         now = get_tb();
2091         if (now > vcpu->arch.dec_expires) {
2092                 /* decrementer has already gone negative */
2093                 kvmppc_core_queue_dec(vcpu);
2094                 kvmppc_core_prepare_to_enter(vcpu);
2095                 return;
2096         }
2097         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
2098                    / tb_ticks_per_sec;
2099         hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2100         vcpu->arch.timer_running = 1;
2101 }
2102
2103 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
2104 {
2105         vcpu->arch.ceded = 0;
2106         if (vcpu->arch.timer_running) {
2107                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2108                 vcpu->arch.timer_running = 0;
2109         }
2110 }
2111
2112 extern int __kvmppc_vcore_entry(void);
2113
2114 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2115                                    struct kvm_vcpu *vcpu)
2116 {
2117         u64 now;
2118
2119         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2120                 return;
2121         spin_lock_irq(&vcpu->arch.tbacct_lock);
2122         now = mftb();
2123         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2124                 vcpu->arch.stolen_logged;
2125         vcpu->arch.busy_preempt = now;
2126         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2127         spin_unlock_irq(&vcpu->arch.tbacct_lock);
2128         --vc->n_runnable;
2129         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2130 }
2131
2132 static int kvmppc_grab_hwthread(int cpu)
2133 {
2134         struct paca_struct *tpaca;
2135         long timeout = 10000;
2136
2137         /*
2138          * ISA v3.0 idle routines do not set hwthread_state or test
2139          * hwthread_req, so they can not grab idle threads.
2140          */
2141         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2142                 WARN(1, "KVM: can not control sibling threads\n");
2143                 return -EBUSY;
2144         }
2145
2146         tpaca = &paca[cpu];
2147
2148         /* Ensure the thread won't go into the kernel if it wakes */
2149         tpaca->kvm_hstate.kvm_vcpu = NULL;
2150         tpaca->kvm_hstate.kvm_vcore = NULL;
2151         tpaca->kvm_hstate.napping = 0;
2152         smp_wmb();
2153         tpaca->kvm_hstate.hwthread_req = 1;
2154
2155         /*
2156          * If the thread is already executing in the kernel (e.g. handling
2157          * a stray interrupt), wait for it to get back to nap mode.
2158          * The smp_mb() is to ensure that our setting of hwthread_req
2159          * is visible before we look at hwthread_state, so if this
2160          * races with the code at system_reset_pSeries and the thread
2161          * misses our setting of hwthread_req, we are sure to see its
2162          * setting of hwthread_state, and vice versa.
2163          */
2164         smp_mb();
2165         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2166                 if (--timeout <= 0) {
2167                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
2168                         return -EBUSY;
2169                 }
2170                 udelay(1);
2171         }
2172         return 0;
2173 }
2174
2175 static void kvmppc_release_hwthread(int cpu)
2176 {
2177         struct paca_struct *tpaca;
2178
2179         tpaca = &paca[cpu];
2180         tpaca->kvm_hstate.kvm_vcpu = NULL;
2181         tpaca->kvm_hstate.kvm_vcore = NULL;
2182         tpaca->kvm_hstate.kvm_split_mode = NULL;
2183         if (!cpu_has_feature(CPU_FTR_ARCH_300))
2184                 tpaca->kvm_hstate.hwthread_req = 0;
2185
2186 }
2187
2188 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2189 {
2190         int i;
2191
2192         cpu = cpu_first_thread_sibling(cpu);
2193         cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2194         /*
2195          * Make sure setting of bit in need_tlb_flush precedes
2196          * testing of cpu_in_guest bits.  The matching barrier on
2197          * the other side is the first smp_mb() in kvmppc_run_core().
2198          */
2199         smp_mb();
2200         for (i = 0; i < threads_per_core; ++i)
2201                 if (cpumask_test_cpu(cpu + i, &kvm->arch.cpu_in_guest))
2202                         smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2203 }
2204
2205 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2206 {
2207         struct kvm *kvm = vcpu->kvm;
2208
2209         /*
2210          * With radix, the guest can do TLB invalidations itself,
2211          * and it could choose to use the local form (tlbiel) if
2212          * it is invalidating a translation that has only ever been
2213          * used on one vcpu.  However, that doesn't mean it has
2214          * only ever been used on one physical cpu, since vcpus
2215          * can move around between pcpus.  To cope with this, when
2216          * a vcpu moves from one pcpu to another, we need to tell
2217          * any vcpus running on the same core as this vcpu previously
2218          * ran to flush the TLB.  The TLB is shared between threads,
2219          * so we use a single bit in .need_tlb_flush for all 4 threads.
2220          */
2221         if (vcpu->arch.prev_cpu != pcpu) {
2222                 if (vcpu->arch.prev_cpu >= 0 &&
2223                     cpu_first_thread_sibling(vcpu->arch.prev_cpu) !=
2224                     cpu_first_thread_sibling(pcpu))
2225                         radix_flush_cpu(kvm, vcpu->arch.prev_cpu, vcpu);
2226                 vcpu->arch.prev_cpu = pcpu;
2227         }
2228 }
2229
2230 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2231 {
2232         int cpu;
2233         struct paca_struct *tpaca;
2234         struct kvm *kvm = vc->kvm;
2235
2236         cpu = vc->pcpu;
2237         if (vcpu) {
2238                 if (vcpu->arch.timer_running) {
2239                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2240                         vcpu->arch.timer_running = 0;
2241                 }
2242                 cpu += vcpu->arch.ptid;
2243                 vcpu->cpu = vc->pcpu;
2244                 vcpu->arch.thread_cpu = cpu;
2245                 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2246         }
2247         tpaca = &paca[cpu];
2248         tpaca->kvm_hstate.kvm_vcpu = vcpu;
2249         tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2250         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2251         smp_wmb();
2252         tpaca->kvm_hstate.kvm_vcore = vc;
2253         if (cpu != smp_processor_id())
2254                 kvmppc_ipi_thread(cpu);
2255 }
2256
2257 static void kvmppc_wait_for_nap(void)
2258 {
2259         int cpu = smp_processor_id();
2260         int i, loops;
2261         int n_threads = threads_per_vcore();
2262
2263         if (n_threads <= 1)
2264                 return;
2265         for (loops = 0; loops < 1000000; ++loops) {
2266                 /*
2267                  * Check if all threads are finished.
2268                  * We set the vcore pointer when starting a thread
2269                  * and the thread clears it when finished, so we look
2270                  * for any threads that still have a non-NULL vcore ptr.
2271                  */
2272                 for (i = 1; i < n_threads; ++i)
2273                         if (paca[cpu + i].kvm_hstate.kvm_vcore)
2274                                 break;
2275                 if (i == n_threads) {
2276                         HMT_medium();
2277                         return;
2278                 }
2279                 HMT_low();
2280         }
2281         HMT_medium();
2282         for (i = 1; i < n_threads; ++i)
2283                 if (paca[cpu + i].kvm_hstate.kvm_vcore)
2284                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2285 }
2286
2287 /*
2288  * Check that we are on thread 0 and that any other threads in
2289  * this core are off-line.  Then grab the threads so they can't
2290  * enter the kernel.
2291  */
2292 static int on_primary_thread(void)
2293 {
2294         int cpu = smp_processor_id();
2295         int thr;
2296
2297         /* Are we on a primary subcore? */
2298         if (cpu_thread_in_subcore(cpu))
2299                 return 0;
2300
2301         thr = 0;
2302         while (++thr < threads_per_subcore)
2303                 if (cpu_online(cpu + thr))
2304                         return 0;
2305
2306         /* Grab all hw threads so they can't go into the kernel */
2307         for (thr = 1; thr < threads_per_subcore; ++thr) {
2308                 if (kvmppc_grab_hwthread(cpu + thr)) {
2309                         /* Couldn't grab one; let the others go */
2310                         do {
2311                                 kvmppc_release_hwthread(cpu + thr);
2312                         } while (--thr > 0);
2313                         return 0;
2314                 }
2315         }
2316         return 1;
2317 }
2318
2319 /*
2320  * A list of virtual cores for each physical CPU.
2321  * These are vcores that could run but their runner VCPU tasks are
2322  * (or may be) preempted.
2323  */
2324 struct preempted_vcore_list {
2325         struct list_head        list;
2326         spinlock_t              lock;
2327 };
2328
2329 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2330
2331 static void init_vcore_lists(void)
2332 {
2333         int cpu;
2334
2335         for_each_possible_cpu(cpu) {
2336                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2337                 spin_lock_init(&lp->lock);
2338                 INIT_LIST_HEAD(&lp->list);
2339         }
2340 }
2341
2342 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2343 {
2344         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2345
2346         vc->vcore_state = VCORE_PREEMPT;
2347         vc->pcpu = smp_processor_id();
2348         if (vc->num_threads < threads_per_vcore()) {
2349                 spin_lock(&lp->lock);
2350                 list_add_tail(&vc->preempt_list, &lp->list);
2351                 spin_unlock(&lp->lock);
2352         }
2353
2354         /* Start accumulating stolen time */
2355         kvmppc_core_start_stolen(vc);
2356 }
2357
2358 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2359 {
2360         struct preempted_vcore_list *lp;
2361
2362         kvmppc_core_end_stolen(vc);
2363         if (!list_empty(&vc->preempt_list)) {
2364                 lp = &per_cpu(preempted_vcores, vc->pcpu);
2365                 spin_lock(&lp->lock);
2366                 list_del_init(&vc->preempt_list);
2367                 spin_unlock(&lp->lock);
2368         }
2369         vc->vcore_state = VCORE_INACTIVE;
2370 }
2371
2372 /*
2373  * This stores information about the virtual cores currently
2374  * assigned to a physical core.
2375  */
2376 struct core_info {
2377         int             n_subcores;
2378         int             max_subcore_threads;
2379         int             total_threads;
2380         int             subcore_threads[MAX_SUBCORES];
2381         struct kvmppc_vcore *vc[MAX_SUBCORES];
2382 };
2383
2384 /*
2385  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2386  * respectively in 2-way micro-threading (split-core) mode.
2387  */
2388 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2389
2390 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2391 {
2392         memset(cip, 0, sizeof(*cip));
2393         cip->n_subcores = 1;
2394         cip->max_subcore_threads = vc->num_threads;
2395         cip->total_threads = vc->num_threads;
2396         cip->subcore_threads[0] = vc->num_threads;
2397         cip->vc[0] = vc;
2398 }
2399
2400 static bool subcore_config_ok(int n_subcores, int n_threads)
2401 {
2402         /* Can only dynamically split if unsplit to begin with */
2403         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2404                 return false;
2405         if (n_subcores > MAX_SUBCORES)
2406                 return false;
2407         if (n_subcores > 1) {
2408                 if (!(dynamic_mt_modes & 2))
2409                         n_subcores = 4;
2410                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2411                         return false;
2412         }
2413
2414         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2415 }
2416
2417 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2418 {
2419         vc->entry_exit_map = 0;
2420         vc->in_guest = 0;
2421         vc->napping_threads = 0;
2422         vc->conferring_threads = 0;
2423 }
2424
2425 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2426 {
2427         int n_threads = vc->num_threads;
2428         int sub;
2429
2430         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2431                 return false;
2432
2433         if (n_threads < cip->max_subcore_threads)
2434                 n_threads = cip->max_subcore_threads;
2435         if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2436                 return false;
2437         cip->max_subcore_threads = n_threads;
2438
2439         sub = cip->n_subcores;
2440         ++cip->n_subcores;
2441         cip->total_threads += vc->num_threads;
2442         cip->subcore_threads[sub] = vc->num_threads;
2443         cip->vc[sub] = vc;
2444         init_vcore_to_run(vc);
2445         list_del_init(&vc->preempt_list);
2446
2447         return true;
2448 }
2449
2450 /*
2451  * Work out whether it is possible to piggyback the execution of
2452  * vcore *pvc onto the execution of the other vcores described in *cip.
2453  */
2454 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2455                           int target_threads)
2456 {
2457         if (cip->total_threads + pvc->num_threads > target_threads)
2458                 return false;
2459
2460         return can_dynamic_split(pvc, cip);
2461 }
2462
2463 static void prepare_threads(struct kvmppc_vcore *vc)
2464 {
2465         int i;
2466         struct kvm_vcpu *vcpu;
2467
2468         for_each_runnable_thread(i, vcpu, vc) {
2469                 if (signal_pending(vcpu->arch.run_task))
2470                         vcpu->arch.ret = -EINTR;
2471                 else if (vcpu->arch.vpa.update_pending ||
2472                          vcpu->arch.slb_shadow.update_pending ||
2473                          vcpu->arch.dtl.update_pending)
2474                         vcpu->arch.ret = RESUME_GUEST;
2475                 else
2476                         continue;
2477                 kvmppc_remove_runnable(vc, vcpu);
2478                 wake_up(&vcpu->arch.cpu_run);
2479         }
2480 }
2481
2482 static void collect_piggybacks(struct core_info *cip, int target_threads)
2483 {
2484         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2485         struct kvmppc_vcore *pvc, *vcnext;
2486
2487         spin_lock(&lp->lock);
2488         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2489                 if (!spin_trylock(&pvc->lock))
2490                         continue;
2491                 prepare_threads(pvc);
2492                 if (!pvc->n_runnable) {
2493                         list_del_init(&pvc->preempt_list);
2494                         if (pvc->runner == NULL) {
2495                                 pvc->vcore_state = VCORE_INACTIVE;
2496                                 kvmppc_core_end_stolen(pvc);
2497                         }
2498                         spin_unlock(&pvc->lock);
2499                         continue;
2500                 }
2501                 if (!can_piggyback(pvc, cip, target_threads)) {
2502                         spin_unlock(&pvc->lock);
2503                         continue;
2504                 }
2505                 kvmppc_core_end_stolen(pvc);
2506                 pvc->vcore_state = VCORE_PIGGYBACK;
2507                 if (cip->total_threads >= target_threads)
2508                         break;
2509         }
2510         spin_unlock(&lp->lock);
2511 }
2512
2513 static bool recheck_signals(struct core_info *cip)
2514 {
2515         int sub, i;
2516         struct kvm_vcpu *vcpu;
2517
2518         for (sub = 0; sub < cip->n_subcores; ++sub)
2519                 for_each_runnable_thread(i, vcpu, cip->vc[sub])
2520                         if (signal_pending(vcpu->arch.run_task))
2521                                 return true;
2522         return false;
2523 }
2524
2525 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2526 {
2527         int still_running = 0, i;
2528         u64 now;
2529         long ret;
2530         struct kvm_vcpu *vcpu;
2531
2532         spin_lock(&vc->lock);
2533         now = get_tb();
2534         for_each_runnable_thread(i, vcpu, vc) {
2535                 /* cancel pending dec exception if dec is positive */
2536                 if (now < vcpu->arch.dec_expires &&
2537                     kvmppc_core_pending_dec(vcpu))
2538                         kvmppc_core_dequeue_dec(vcpu);
2539
2540                 trace_kvm_guest_exit(vcpu);
2541
2542                 ret = RESUME_GUEST;
2543                 if (vcpu->arch.trap)
2544                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2545                                                     vcpu->arch.run_task);
2546
2547                 vcpu->arch.ret = ret;
2548                 vcpu->arch.trap = 0;
2549
2550                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2551                         if (vcpu->arch.pending_exceptions)
2552                                 kvmppc_core_prepare_to_enter(vcpu);
2553                         if (vcpu->arch.ceded)
2554                                 kvmppc_set_timer(vcpu);
2555                         else
2556                                 ++still_running;
2557                 } else {
2558                         kvmppc_remove_runnable(vc, vcpu);
2559                         wake_up(&vcpu->arch.cpu_run);
2560                 }
2561         }
2562         if (!is_master) {
2563                 if (still_running > 0) {
2564                         kvmppc_vcore_preempt(vc);
2565                 } else if (vc->runner) {
2566                         vc->vcore_state = VCORE_PREEMPT;
2567                         kvmppc_core_start_stolen(vc);
2568                 } else {
2569                         vc->vcore_state = VCORE_INACTIVE;
2570                 }
2571                 if (vc->n_runnable > 0 && vc->runner == NULL) {
2572                         /* make sure there's a candidate runner awake */
2573                         i = -1;
2574                         vcpu = next_runnable_thread(vc, &i);
2575                         wake_up(&vcpu->arch.cpu_run);
2576                 }
2577         }
2578         spin_unlock(&vc->lock);
2579 }
2580
2581 /*
2582  * Clear core from the list of active host cores as we are about to
2583  * enter the guest. Only do this if it is the primary thread of the
2584  * core (not if a subcore) that is entering the guest.
2585  */
2586 static inline int kvmppc_clear_host_core(unsigned int cpu)
2587 {
2588         int core;
2589
2590         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2591                 return 0;
2592         /*
2593          * Memory barrier can be omitted here as we will do a smp_wmb()
2594          * later in kvmppc_start_thread and we need ensure that state is
2595          * visible to other CPUs only after we enter guest.
2596          */
2597         core = cpu >> threads_shift;
2598         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2599         return 0;
2600 }
2601
2602 /*
2603  * Advertise this core as an active host core since we exited the guest
2604  * Only need to do this if it is the primary thread of the core that is
2605  * exiting.
2606  */
2607 static inline int kvmppc_set_host_core(unsigned int cpu)
2608 {
2609         int core;
2610
2611         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2612                 return 0;
2613
2614         /*
2615          * Memory barrier can be omitted here because we do a spin_unlock
2616          * immediately after this which provides the memory barrier.
2617          */
2618         core = cpu >> threads_shift;
2619         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2620         return 0;
2621 }
2622
2623 static void set_irq_happened(int trap)
2624 {
2625         switch (trap) {
2626         case BOOK3S_INTERRUPT_EXTERNAL:
2627                 local_paca->irq_happened |= PACA_IRQ_EE;
2628                 break;
2629         case BOOK3S_INTERRUPT_H_DOORBELL:
2630                 local_paca->irq_happened |= PACA_IRQ_DBELL;
2631                 break;
2632         case BOOK3S_INTERRUPT_HMI:
2633                 local_paca->irq_happened |= PACA_IRQ_HMI;
2634                 break;
2635         }
2636 }
2637
2638 /*
2639  * Run a set of guest threads on a physical core.
2640  * Called with vc->lock held.
2641  */
2642 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2643 {
2644         struct kvm_vcpu *vcpu;
2645         int i;
2646         int srcu_idx;
2647         struct core_info core_info;
2648         struct kvmppc_vcore *pvc;
2649         struct kvm_split_mode split_info, *sip;
2650         int split, subcore_size, active;
2651         int sub;
2652         bool thr0_done;
2653         unsigned long cmd_bit, stat_bit;
2654         int pcpu, thr;
2655         int target_threads;
2656         int controlled_threads;
2657         int trap;
2658
2659         /*
2660          * Remove from the list any threads that have a signal pending
2661          * or need a VPA update done
2662          */
2663         prepare_threads(vc);
2664
2665         /* if the runner is no longer runnable, let the caller pick a new one */
2666         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2667                 return;
2668
2669         /*
2670          * Initialize *vc.
2671          */
2672         init_vcore_to_run(vc);
2673         vc->preempt_tb = TB_NIL;
2674
2675         /*
2676          * Number of threads that we will be controlling: the same as
2677          * the number of threads per subcore, except on POWER9,
2678          * where it's 1 because the threads are (mostly) independent.
2679          */
2680         controlled_threads = threads_per_vcore();
2681
2682         /*
2683          * Make sure we are running on primary threads, and that secondary
2684          * threads are offline.  Also check if the number of threads in this
2685          * guest are greater than the current system threads per guest.
2686          */
2687         if ((controlled_threads > 1) &&
2688             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2689                 for_each_runnable_thread(i, vcpu, vc) {
2690                         vcpu->arch.ret = -EBUSY;
2691                         kvmppc_remove_runnable(vc, vcpu);
2692                         wake_up(&vcpu->arch.cpu_run);
2693                 }
2694                 goto out;
2695         }
2696
2697         /*
2698          * See if we could run any other vcores on the physical core
2699          * along with this one.
2700          */
2701         init_core_info(&core_info, vc);
2702         pcpu = smp_processor_id();
2703         target_threads = controlled_threads;
2704         if (target_smt_mode && target_smt_mode < target_threads)
2705                 target_threads = target_smt_mode;
2706         if (vc->num_threads < target_threads)
2707                 collect_piggybacks(&core_info, target_threads);
2708
2709         /*
2710          * On radix, arrange for TLB flushing if necessary.
2711          * This has to be done before disabling interrupts since
2712          * it uses smp_call_function().
2713          */
2714         pcpu = smp_processor_id();
2715         if (kvm_is_radix(vc->kvm)) {
2716                 for (sub = 0; sub < core_info.n_subcores; ++sub)
2717                         for_each_runnable_thread(i, vcpu, core_info.vc[sub])
2718                                 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
2719         }
2720
2721         /*
2722          * Hard-disable interrupts, and check resched flag and signals.
2723          * If we need to reschedule or deliver a signal, clean up
2724          * and return without going into the guest(s).
2725          * If the hpte_setup_done flag has been cleared, don't go into the
2726          * guest because that means a HPT resize operation is in progress.
2727          */
2728         local_irq_disable();
2729         hard_irq_disable();
2730         if (lazy_irq_pending() || need_resched() ||
2731             recheck_signals(&core_info) ||
2732             (!kvm_is_radix(vc->kvm) && !vc->kvm->arch.hpte_setup_done)) {
2733                 local_irq_enable();
2734                 vc->vcore_state = VCORE_INACTIVE;
2735                 /* Unlock all except the primary vcore */
2736                 for (sub = 1; sub < core_info.n_subcores; ++sub) {
2737                         pvc = core_info.vc[sub];
2738                         /* Put back on to the preempted vcores list */
2739                         kvmppc_vcore_preempt(pvc);
2740                         spin_unlock(&pvc->lock);
2741                 }
2742                 for (i = 0; i < controlled_threads; ++i)
2743                         kvmppc_release_hwthread(pcpu + i);
2744                 return;
2745         }
2746
2747         kvmppc_clear_host_core(pcpu);
2748
2749         /* Decide on micro-threading (split-core) mode */
2750         subcore_size = threads_per_subcore;
2751         cmd_bit = stat_bit = 0;
2752         split = core_info.n_subcores;
2753         sip = NULL;
2754         if (split > 1) {
2755                 /* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2756                 if (split == 2 && (dynamic_mt_modes & 2)) {
2757                         cmd_bit = HID0_POWER8_1TO2LPAR;
2758                         stat_bit = HID0_POWER8_2LPARMODE;
2759                 } else {
2760                         split = 4;
2761                         cmd_bit = HID0_POWER8_1TO4LPAR;
2762                         stat_bit = HID0_POWER8_4LPARMODE;
2763                 }
2764                 subcore_size = MAX_SMT_THREADS / split;
2765                 sip = &split_info;
2766                 memset(&split_info, 0, sizeof(split_info));
2767                 split_info.rpr = mfspr(SPRN_RPR);
2768                 split_info.pmmar = mfspr(SPRN_PMMAR);
2769                 split_info.ldbar = mfspr(SPRN_LDBAR);
2770                 split_info.subcore_size = subcore_size;
2771                 for (sub = 0; sub < core_info.n_subcores; ++sub)
2772                         split_info.vc[sub] = core_info.vc[sub];
2773                 /* order writes to split_info before kvm_split_mode pointer */
2774                 smp_wmb();
2775         }
2776         for (thr = 0; thr < controlled_threads; ++thr)
2777                 paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2778
2779         /* Initiate micro-threading (split-core) if required */
2780         if (cmd_bit) {
2781                 unsigned long hid0 = mfspr(SPRN_HID0);
2782
2783                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2784                 mb();
2785                 mtspr(SPRN_HID0, hid0);
2786                 isync();
2787                 for (;;) {
2788                         hid0 = mfspr(SPRN_HID0);
2789                         if (hid0 & stat_bit)
2790                                 break;
2791                         cpu_relax();
2792                 }
2793         }
2794
2795         /* Start all the threads */
2796         active = 0;
2797         for (sub = 0; sub < core_info.n_subcores; ++sub) {
2798                 thr = subcore_thread_map[sub];
2799                 thr0_done = false;
2800                 active |= 1 << thr;
2801                 pvc = core_info.vc[sub];
2802                 pvc->pcpu = pcpu + thr;
2803                 for_each_runnable_thread(i, vcpu, pvc) {
2804                         kvmppc_start_thread(vcpu, pvc);
2805                         kvmppc_create_dtl_entry(vcpu, pvc);
2806                         trace_kvm_guest_enter(vcpu);
2807                         if (!vcpu->arch.ptid)
2808                                 thr0_done = true;
2809                         active |= 1 << (thr + vcpu->arch.ptid);
2810                 }
2811                 /*
2812                  * We need to start the first thread of each subcore
2813                  * even if it doesn't have a vcpu.
2814                  */
2815                 if (!thr0_done)
2816                         kvmppc_start_thread(NULL, pvc);
2817                 thr += pvc->num_threads;
2818         }
2819
2820         /*
2821          * Ensure that split_info.do_nap is set after setting
2822          * the vcore pointer in the PACA of the secondaries.
2823          */
2824         smp_mb();
2825         if (cmd_bit)
2826                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
2827
2828         /*
2829          * When doing micro-threading, poke the inactive threads as well.
2830          * This gets them to the nap instruction after kvm_do_nap,
2831          * which reduces the time taken to unsplit later.
2832          */
2833         if (split > 1)
2834                 for (thr = 1; thr < threads_per_subcore; ++thr)
2835                         if (!(active & (1 << thr)))
2836                                 kvmppc_ipi_thread(pcpu + thr);
2837
2838         vc->vcore_state = VCORE_RUNNING;
2839         preempt_disable();
2840
2841         trace_kvmppc_run_core(vc, 0);
2842
2843         for (sub = 0; sub < core_info.n_subcores; ++sub)
2844                 spin_unlock(&core_info.vc[sub]->lock);
2845
2846         /*
2847          * Interrupts will be enabled once we get into the guest,
2848          * so tell lockdep that we're about to enable interrupts.
2849          */
2850         trace_hardirqs_on();
2851
2852         guest_enter_irqoff();
2853
2854         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2855
2856         trap = __kvmppc_vcore_entry();
2857
2858         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2859
2860         trace_hardirqs_off();
2861         set_irq_happened(trap);
2862
2863         spin_lock(&vc->lock);
2864         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2865         vc->vcore_state = VCORE_EXITING;
2866
2867         /* wait for secondary threads to finish writing their state to memory */
2868         kvmppc_wait_for_nap();
2869
2870         /* Return to whole-core mode if we split the core earlier */
2871         if (split > 1) {
2872                 unsigned long hid0 = mfspr(SPRN_HID0);
2873                 unsigned long loops = 0;
2874
2875                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2876                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2877                 mb();
2878                 mtspr(SPRN_HID0, hid0);
2879                 isync();
2880                 for (;;) {
2881                         hid0 = mfspr(SPRN_HID0);
2882                         if (!(hid0 & stat_bit))
2883                                 break;
2884                         cpu_relax();
2885                         ++loops;
2886                 }
2887                 split_info.do_nap = 0;
2888         }
2889
2890         kvmppc_set_host_core(pcpu);
2891
2892         local_irq_enable();
2893         guest_exit();
2894
2895         /* Let secondaries go back to the offline loop */
2896         for (i = 0; i < controlled_threads; ++i) {
2897                 kvmppc_release_hwthread(pcpu + i);
2898                 if (sip && sip->napped[i])
2899                         kvmppc_ipi_thread(pcpu + i);
2900                 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
2901         }
2902
2903         spin_unlock(&vc->lock);
2904
2905         /* make sure updates to secondary vcpu structs are visible now */
2906         smp_mb();
2907
2908         preempt_enable();
2909
2910         for (sub = 0; sub < core_info.n_subcores; ++sub) {
2911                 pvc = core_info.vc[sub];
2912                 post_guest_process(pvc, pvc == vc);
2913         }
2914
2915         spin_lock(&vc->lock);
2916
2917  out:
2918         vc->vcore_state = VCORE_INACTIVE;
2919         trace_kvmppc_run_core(vc, 1);
2920 }
2921
2922 /*
2923  * Wait for some other vcpu thread to execute us, and
2924  * wake us up when we need to handle something in the host.
2925  */
2926 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2927                                  struct kvm_vcpu *vcpu, int wait_state)
2928 {
2929         DEFINE_WAIT(wait);
2930
2931         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2932         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2933                 spin_unlock(&vc->lock);
2934                 schedule();
2935                 spin_lock(&vc->lock);
2936         }
2937         finish_wait(&vcpu->arch.cpu_run, &wait);
2938 }
2939
2940 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
2941 {
2942         /* 10us base */
2943         if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
2944                 vc->halt_poll_ns = 10000;
2945         else
2946                 vc->halt_poll_ns *= halt_poll_ns_grow;
2947 }
2948
2949 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
2950 {
2951         if (halt_poll_ns_shrink == 0)
2952                 vc->halt_poll_ns = 0;
2953         else
2954                 vc->halt_poll_ns /= halt_poll_ns_shrink;
2955 }
2956
2957 #ifdef CONFIG_KVM_XICS
2958 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
2959 {
2960         if (!xive_enabled())
2961                 return false;
2962         return vcpu->arch.xive_saved_state.pipr <
2963                 vcpu->arch.xive_saved_state.cppr;
2964 }
2965 #else
2966 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
2967 {
2968         return false;
2969 }
2970 #endif /* CONFIG_KVM_XICS */
2971
2972 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
2973 {
2974         if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
2975             kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
2976                 return true;
2977
2978         return false;
2979 }
2980
2981 /*
2982  * Check to see if any of the runnable vcpus on the vcore have pending
2983  * exceptions or are no longer ceded
2984  */
2985 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
2986 {
2987         struct kvm_vcpu *vcpu;
2988         int i;
2989
2990         for_each_runnable_thread(i, vcpu, vc) {
2991                 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
2992                         return 1;
2993         }
2994
2995         return 0;
2996 }
2997
2998 /*
2999  * All the vcpus in this vcore are idle, so wait for a decrementer
3000  * or external interrupt to one of the vcpus.  vc->lock is held.
3001  */
3002 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3003 {
3004         ktime_t cur, start_poll, start_wait;
3005         int do_sleep = 1;
3006         u64 block_ns;
3007         DECLARE_SWAITQUEUE(wait);
3008
3009         /* Poll for pending exceptions and ceded state */
3010         cur = start_poll = ktime_get();
3011         if (vc->halt_poll_ns) {
3012                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3013                 ++vc->runner->stat.halt_attempted_poll;
3014
3015                 vc->vcore_state = VCORE_POLLING;
3016                 spin_unlock(&vc->lock);
3017
3018                 do {
3019                         if (kvmppc_vcore_check_block(vc)) {
3020                                 do_sleep = 0;
3021                                 break;
3022                         }
3023                         cur = ktime_get();
3024                 } while (single_task_running() && ktime_before(cur, stop));
3025
3026                 spin_lock(&vc->lock);
3027                 vc->vcore_state = VCORE_INACTIVE;
3028
3029                 if (!do_sleep) {
3030                         ++vc->runner->stat.halt_successful_poll;
3031                         goto out;
3032                 }
3033         }
3034
3035         prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3036
3037         if (kvmppc_vcore_check_block(vc)) {
3038                 finish_swait(&vc->wq, &wait);
3039                 do_sleep = 0;
3040                 /* If we polled, count this as a successful poll */
3041                 if (vc->halt_poll_ns)
3042                         ++vc->runner->stat.halt_successful_poll;
3043                 goto out;
3044         }
3045
3046         start_wait = ktime_get();
3047
3048         vc->vcore_state = VCORE_SLEEPING;
3049         trace_kvmppc_vcore_blocked(vc, 0);
3050         spin_unlock(&vc->lock);
3051         schedule();
3052         finish_swait(&vc->wq, &wait);
3053         spin_lock(&vc->lock);
3054         vc->vcore_state = VCORE_INACTIVE;
3055         trace_kvmppc_vcore_blocked(vc, 1);
3056         ++vc->runner->stat.halt_successful_wait;
3057
3058         cur = ktime_get();
3059
3060 out:
3061         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3062
3063         /* Attribute wait time */
3064         if (do_sleep) {
3065                 vc->runner->stat.halt_wait_ns +=
3066                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
3067                 /* Attribute failed poll time */
3068                 if (vc->halt_poll_ns)
3069                         vc->runner->stat.halt_poll_fail_ns +=
3070                                 ktime_to_ns(start_wait) -
3071                                 ktime_to_ns(start_poll);
3072         } else {
3073                 /* Attribute successful poll time */
3074                 if (vc->halt_poll_ns)
3075                         vc->runner->stat.halt_poll_success_ns +=
3076                                 ktime_to_ns(cur) -
3077                                 ktime_to_ns(start_poll);
3078         }
3079
3080         /* Adjust poll time */
3081         if (halt_poll_ns) {
3082                 if (block_ns <= vc->halt_poll_ns)
3083                         ;
3084                 /* We slept and blocked for longer than the max halt time */
3085                 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3086                         shrink_halt_poll_ns(vc);
3087                 /* We slept and our poll time is too small */
3088                 else if (vc->halt_poll_ns < halt_poll_ns &&
3089                                 block_ns < halt_poll_ns)
3090                         grow_halt_poll_ns(vc);
3091                 if (vc->halt_poll_ns > halt_poll_ns)
3092                         vc->halt_poll_ns = halt_poll_ns;
3093         } else
3094                 vc->halt_poll_ns = 0;
3095
3096         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3097 }
3098
3099 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3100 {
3101         int n_ceded, i, r;
3102         struct kvmppc_vcore *vc;
3103         struct kvm_vcpu *v;
3104
3105         trace_kvmppc_run_vcpu_enter(vcpu);
3106
3107         kvm_run->exit_reason = 0;
3108         vcpu->arch.ret = RESUME_GUEST;
3109         vcpu->arch.trap = 0;
3110         kvmppc_update_vpas(vcpu);
3111
3112         /*
3113          * Synchronize with other threads in this virtual core
3114          */
3115         vc = vcpu->arch.vcore;
3116         spin_lock(&vc->lock);
3117         vcpu->arch.ceded = 0;
3118         vcpu->arch.run_task = current;
3119         vcpu->arch.kvm_run = kvm_run;
3120         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3121         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3122         vcpu->arch.busy_preempt = TB_NIL;
3123         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3124         ++vc->n_runnable;
3125
3126         /*
3127          * This happens the first time this is called for a vcpu.
3128          * If the vcore is already running, we may be able to start
3129          * this thread straight away and have it join in.
3130          */
3131         if (!signal_pending(current)) {
3132                 if (vc->vcore_state == VCORE_PIGGYBACK) {
3133                         if (spin_trylock(&vc->lock)) {
3134                                 if (vc->vcore_state == VCORE_RUNNING &&
3135                                     !VCORE_IS_EXITING(vc)) {
3136                                         kvmppc_create_dtl_entry(vcpu, vc);
3137                                         kvmppc_start_thread(vcpu, vc);
3138                                         trace_kvm_guest_enter(vcpu);
3139                                 }
3140                                 spin_unlock(&vc->lock);
3141                         }
3142                 } else if (vc->vcore_state == VCORE_RUNNING &&
3143                            !VCORE_IS_EXITING(vc)) {
3144                         kvmppc_create_dtl_entry(vcpu, vc);
3145                         kvmppc_start_thread(vcpu, vc);
3146                         trace_kvm_guest_enter(vcpu);
3147                 } else if (vc->vcore_state == VCORE_SLEEPING) {
3148                         swake_up(&vc->wq);
3149                 }
3150
3151         }
3152
3153         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3154                !signal_pending(current)) {
3155                 /* See if the HPT and VRMA are ready to go */
3156                 if (!kvm_is_radix(vcpu->kvm) &&
3157                     !vcpu->kvm->arch.hpte_setup_done) {
3158                         spin_unlock(&vc->lock);
3159                         r = kvmppc_hv_setup_htab_rma(vcpu);
3160                         spin_lock(&vc->lock);
3161                         if (r) {
3162                                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3163                                 kvm_run->fail_entry.hardware_entry_failure_reason = 0;
3164                                 vcpu->arch.ret = r;
3165                                 break;
3166                         }
3167                 }
3168
3169                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3170                         kvmppc_vcore_end_preempt(vc);
3171
3172                 if (vc->vcore_state != VCORE_INACTIVE) {
3173                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3174                         continue;
3175                 }
3176                 for_each_runnable_thread(i, v, vc) {
3177                         kvmppc_core_prepare_to_enter(v);
3178                         if (signal_pending(v->arch.run_task)) {
3179                                 kvmppc_remove_runnable(vc, v);
3180                                 v->stat.signal_exits++;
3181                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3182                                 v->arch.ret = -EINTR;
3183                                 wake_up(&v->arch.cpu_run);
3184                         }
3185                 }
3186                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3187                         break;
3188                 n_ceded = 0;
3189                 for_each_runnable_thread(i, v, vc) {
3190                         if (!kvmppc_vcpu_woken(v))
3191                                 n_ceded += v->arch.ceded;
3192                         else
3193                                 v->arch.ceded = 0;
3194                 }
3195                 vc->runner = vcpu;
3196                 if (n_ceded == vc->n_runnable) {
3197                         kvmppc_vcore_blocked(vc);
3198                 } else if (need_resched()) {
3199                         kvmppc_vcore_preempt(vc);
3200                         /* Let something else run */
3201                         cond_resched_lock(&vc->lock);
3202                         if (vc->vcore_state == VCORE_PREEMPT)
3203                                 kvmppc_vcore_end_preempt(vc);
3204                 } else {
3205                         kvmppc_run_core(vc);
3206                 }
3207                 vc->runner = NULL;
3208         }
3209
3210         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3211                (vc->vcore_state == VCORE_RUNNING ||
3212                 vc->vcore_state == VCORE_EXITING ||
3213                 vc->vcore_state == VCORE_PIGGYBACK))
3214                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
3215
3216         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3217                 kvmppc_vcore_end_preempt(vc);
3218
3219         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3220                 kvmppc_remove_runnable(vc, vcpu);
3221                 vcpu->stat.signal_exits++;
3222                 kvm_run->exit_reason = KVM_EXIT_INTR;
3223                 vcpu->arch.ret = -EINTR;
3224         }
3225
3226         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
3227                 /* Wake up some vcpu to run the core */
3228                 i = -1;
3229                 v = next_runnable_thread(vc, &i);
3230                 wake_up(&v->arch.cpu_run);
3231         }
3232
3233         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
3234         spin_unlock(&vc->lock);
3235         return vcpu->arch.ret;
3236 }
3237
3238 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
3239 {
3240         int r;
3241         int srcu_idx;
3242         unsigned long ebb_regs[3] = {}; /* shut up GCC */
3243         unsigned long user_tar = 0;
3244         unsigned int user_vrsave;
3245
3246         if (!vcpu->arch.sane) {
3247                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3248                 return -EINVAL;
3249         }
3250
3251         /*
3252          * Don't allow entry with a suspended transaction, because
3253          * the guest entry/exit code will lose it.
3254          * If the guest has TM enabled, save away their TM-related SPRs
3255          * (they will get restored by the TM unavailable interrupt).
3256          */
3257 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3258         if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
3259             (current->thread.regs->msr & MSR_TM)) {
3260                 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
3261                         run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3262                         run->fail_entry.hardware_entry_failure_reason = 0;
3263                         return -EINVAL;
3264                 }
3265                 /* Enable TM so we can read the TM SPRs */
3266                 mtmsr(mfmsr() | MSR_TM);
3267                 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
3268                 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
3269                 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
3270                 current->thread.regs->msr &= ~MSR_TM;
3271         }
3272 #endif
3273
3274         kvmppc_core_prepare_to_enter(vcpu);
3275
3276         /* No need to go into the guest when all we'll do is come back out */
3277         if (signal_pending(current)) {
3278                 run->exit_reason = KVM_EXIT_INTR;
3279                 return -EINTR;
3280         }
3281
3282         atomic_inc(&vcpu->kvm->arch.vcpus_running);
3283         /* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
3284         smp_mb();
3285
3286         flush_all_to_thread(current);
3287
3288         /* Save userspace EBB and other register values */
3289         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3290                 ebb_regs[0] = mfspr(SPRN_EBBHR);
3291                 ebb_regs[1] = mfspr(SPRN_EBBRR);
3292                 ebb_regs[2] = mfspr(SPRN_BESCR);
3293                 user_tar = mfspr(SPRN_TAR);
3294         }
3295         user_vrsave = mfspr(SPRN_VRSAVE);
3296
3297         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
3298         vcpu->arch.pgdir = current->mm->pgd;
3299         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
3300
3301         do {
3302                 r = kvmppc_run_vcpu(run, vcpu);
3303
3304                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
3305                     !(vcpu->arch.shregs.msr & MSR_PR)) {
3306                         trace_kvm_hcall_enter(vcpu);
3307                         r = kvmppc_pseries_do_hcall(vcpu);
3308                         trace_kvm_hcall_exit(vcpu, r);
3309                         kvmppc_core_prepare_to_enter(vcpu);
3310                 } else if (r == RESUME_PAGE_FAULT) {
3311                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3312                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
3313                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
3314                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
3315                 } else if (r == RESUME_PASSTHROUGH) {
3316                         if (WARN_ON(xive_enabled()))
3317                                 r = H_SUCCESS;
3318                         else
3319                                 r = kvmppc_xics_rm_complete(vcpu, 0);
3320                 }
3321         } while (is_kvmppc_resume_guest(r));
3322
3323         /* Restore userspace EBB and other register values */
3324         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
3325                 mtspr(SPRN_EBBHR, ebb_regs[0]);
3326                 mtspr(SPRN_EBBRR, ebb_regs[1]);
3327                 mtspr(SPRN_BESCR, ebb_regs[2]);
3328                 mtspr(SPRN_TAR, user_tar);
3329                 mtspr(SPRN_FSCR, current->thread.fscr);
3330         }
3331         mtspr(SPRN_VRSAVE, user_vrsave);
3332
3333         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
3334         atomic_dec(&vcpu->kvm->arch.vcpus_running);
3335         return r;
3336 }
3337
3338 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
3339                                      int linux_psize)
3340 {
3341         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
3342
3343         if (!def->shift)
3344                 return;
3345         (*sps)->page_shift = def->shift;
3346         (*sps)->slb_enc = def->sllp;
3347         (*sps)->enc[0].page_shift = def->shift;
3348         (*sps)->enc[0].pte_enc = def->penc[linux_psize];
3349         /*
3350          * Add 16MB MPSS support if host supports it
3351          */
3352         if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
3353                 (*sps)->enc[1].page_shift = 24;
3354                 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
3355         }
3356         (*sps)++;
3357 }
3358
3359 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
3360                                          struct kvm_ppc_smmu_info *info)
3361 {
3362         struct kvm_ppc_one_seg_page_size *sps;
3363
3364         /*
3365          * Since we don't yet support HPT guests on a radix host,
3366          * return an error if the host uses radix.
3367          */
3368         if (radix_enabled())
3369                 return -EINVAL;
3370
3371         /*
3372          * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
3373          * POWER7 doesn't support keys for instruction accesses,
3374          * POWER8 and POWER9 do.
3375          */
3376         info->data_keys = 32;
3377         info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
3378
3379         info->flags = KVM_PPC_PAGE_SIZES_REAL;
3380         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
3381                 info->flags |= KVM_PPC_1T_SEGMENTS;
3382         info->slb_size = mmu_slb_size;
3383
3384         /* We only support these sizes for now, and no muti-size segments */
3385         sps = &info->sps[0];
3386         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
3387         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
3388         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
3389
3390         return 0;
3391 }
3392
3393 /*
3394  * Get (and clear) the dirty memory log for a memory slot.
3395  */
3396 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
3397                                          struct kvm_dirty_log *log)
3398 {
3399         struct kvm_memslots *slots;
3400         struct kvm_memory_slot *memslot;
3401         int i, r;
3402         unsigned long n;
3403         unsigned long *buf;
3404         struct kvm_vcpu *vcpu;
3405
3406         mutex_lock(&kvm->slots_lock);
3407
3408         r = -EINVAL;
3409         if (log->slot >= KVM_USER_MEM_SLOTS)
3410                 goto out;
3411
3412         slots = kvm_memslots(kvm);
3413         memslot = id_to_memslot(slots, log->slot);
3414         r = -ENOENT;
3415         if (!memslot->dirty_bitmap)
3416                 goto out;
3417
3418         /*
3419          * Use second half of bitmap area because radix accumulates
3420          * bits in the first half.
3421          */
3422         n = kvm_dirty_bitmap_bytes(memslot);
3423         buf = memslot->dirty_bitmap + n / sizeof(long);
3424         memset(buf, 0, n);
3425
3426         if (kvm_is_radix(kvm))
3427                 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
3428         else
3429                 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
3430         if (r)
3431                 goto out;
3432
3433         /* Harvest dirty bits from VPA and DTL updates */
3434         /* Note: we never modify the SLB shadow buffer areas */
3435         kvm_for_each_vcpu(i, vcpu, kvm) {
3436                 spin_lock(&vcpu->arch.vpa_update_lock);
3437                 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
3438                 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
3439                 spin_unlock(&vcpu->arch.vpa_update_lock);
3440         }
3441
3442         r = -EFAULT;
3443         if (copy_to_user(log->dirty_bitmap, buf, n))
3444                 goto out;
3445
3446         r = 0;
3447 out:
3448         mutex_unlock(&kvm->slots_lock);
3449         return r;
3450 }
3451
3452 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
3453                                         struct kvm_memory_slot *dont)
3454 {
3455         if (!dont || free->arch.rmap != dont->arch.rmap) {
3456                 vfree(free->arch.rmap);
3457                 free->arch.rmap = NULL;
3458         }
3459 }
3460
3461 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
3462                                          unsigned long npages)
3463 {
3464         /*
3465          * For now, if radix_enabled() then we only support radix guests,
3466          * and in that case we don't need the rmap array.
3467          */
3468         if (radix_enabled()) {
3469                 slot->arch.rmap = NULL;
3470                 return 0;
3471         }
3472
3473         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
3474         if (!slot->arch.rmap)
3475                 return -ENOMEM;
3476
3477         return 0;
3478 }
3479
3480 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
3481                                         struct kvm_memory_slot *memslot,
3482                                         const struct kvm_userspace_memory_region *mem)
3483 {
3484         return 0;
3485 }
3486
3487 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
3488                                 const struct kvm_userspace_memory_region *mem,
3489                                 const struct kvm_memory_slot *old,
3490                                 const struct kvm_memory_slot *new)
3491 {
3492         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
3493         struct kvm_memslots *slots;
3494         struct kvm_memory_slot *memslot;
3495
3496         /*
3497          * If we are making a new memslot, it might make
3498          * some address that was previously cached as emulated
3499          * MMIO be no longer emulated MMIO, so invalidate
3500          * all the caches of emulated MMIO translations.
3501          */
3502         if (npages)
3503                 atomic64_inc(&kvm->arch.mmio_update);
3504
3505         if (npages && old->npages && !kvm_is_radix(kvm)) {
3506                 /*
3507                  * If modifying a memslot, reset all the rmap dirty bits.
3508                  * If this is a new memslot, we don't need to do anything
3509                  * since the rmap array starts out as all zeroes,
3510                  * i.e. no pages are dirty.
3511                  */
3512                 slots = kvm_memslots(kvm);
3513                 memslot = id_to_memslot(slots, mem->slot);
3514                 kvmppc_hv_get_dirty_log_hpt(kvm, memslot, NULL);
3515         }
3516 }
3517
3518 /*
3519  * Update LPCR values in kvm->arch and in vcores.
3520  * Caller must hold kvm->lock.
3521  */
3522 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3523 {
3524         long int i;
3525         u32 cores_done = 0;
3526
3527         if ((kvm->arch.lpcr & mask) == lpcr)
3528                 return;
3529
3530         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3531
3532         for (i = 0; i < KVM_MAX_VCORES; ++i) {
3533                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3534                 if (!vc)
3535                         continue;
3536                 spin_lock(&vc->lock);
3537                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3538                 spin_unlock(&vc->lock);
3539                 if (++cores_done >= kvm->arch.online_vcores)
3540                         break;
3541         }
3542 }
3543
3544 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3545 {
3546         return;
3547 }
3548
3549 static void kvmppc_setup_partition_table(struct kvm *kvm)
3550 {
3551         unsigned long dw0, dw1;
3552
3553         if (!kvm_is_radix(kvm)) {
3554                 /* PS field - page size for VRMA */
3555                 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
3556                         ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
3557                 /* HTABSIZE and HTABORG fields */
3558                 dw0 |= kvm->arch.sdr1;
3559
3560                 /* Second dword as set by userspace */
3561                 dw1 = kvm->arch.process_table;
3562         } else {
3563                 dw0 = PATB_HR | radix__get_tree_size() |
3564                         __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
3565                 dw1 = PATB_GR | kvm->arch.process_table;
3566         }
3567
3568         mmu_partition_table_set_entry(kvm->arch.lpid, dw0, dw1);
3569 }
3570
3571 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3572 {
3573         int err = 0;
3574         struct kvm *kvm = vcpu->kvm;
3575         unsigned long hva;
3576         struct kvm_memory_slot *memslot;
3577         struct vm_area_struct *vma;
3578         unsigned long lpcr = 0, senc;
3579         unsigned long psize, porder;
3580         int srcu_idx;
3581
3582         mutex_lock(&kvm->lock);
3583         if (kvm->arch.hpte_setup_done)
3584                 goto out;       /* another vcpu beat us to it */
3585
3586         /* Allocate hashed page table (if not done already) and reset it */
3587         if (!kvm->arch.hpt.virt) {
3588                 int order = KVM_DEFAULT_HPT_ORDER;
3589                 struct kvm_hpt_info info;
3590
3591                 err = kvmppc_allocate_hpt(&info, order);
3592                 /* If we get here, it means userspace didn't specify a
3593                  * size explicitly.  So, try successively smaller
3594                  * sizes if the default failed. */
3595                 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
3596                         err  = kvmppc_allocate_hpt(&info, order);
3597
3598                 if (err < 0) {
3599                         pr_err("KVM: Couldn't alloc HPT\n");
3600                         goto out;
3601                 }
3602
3603                 kvmppc_set_hpt(kvm, &info);
3604         }
3605
3606         /* Look up the memslot for guest physical address 0 */
3607         srcu_idx = srcu_read_lock(&kvm->srcu);
3608         memslot = gfn_to_memslot(kvm, 0);
3609
3610         /* We must have some memory at 0 by now */
3611         err = -EINVAL;
3612         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3613                 goto out_srcu;
3614
3615         /* Look up the VMA for the start of this memory slot */
3616         hva = memslot->userspace_addr;
3617         down_read(&current->mm->mmap_sem);
3618         vma = find_vma(current->mm, hva);
3619         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3620                 goto up_out;
3621
3622         psize = vma_kernel_pagesize(vma);
3623
3624         up_read(&current->mm->mmap_sem);
3625
3626         /* We can handle 4k, 64k or 16M pages in the VRMA */
3627         if (psize >= 0x1000000)
3628                 psize = 0x1000000;
3629         else if (psize >= 0x10000)
3630                 psize = 0x10000;
3631         else
3632                 psize = 0x1000;
3633         porder = __ilog2(psize);
3634
3635         senc = slb_pgsize_encoding(psize);
3636         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3637                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3638         /* Create HPTEs in the hash page table for the VRMA */
3639         kvmppc_map_vrma(vcpu, memslot, porder);
3640
3641         /* Update VRMASD field in the LPCR */
3642         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3643                 /* the -4 is to account for senc values starting at 0x10 */
3644                 lpcr = senc << (LPCR_VRMASD_SH - 4);
3645                 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3646         } else {
3647                 kvmppc_setup_partition_table(kvm);
3648         }
3649
3650         /* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
3651         smp_wmb();
3652         kvm->arch.hpte_setup_done = 1;
3653         err = 0;
3654  out_srcu:
3655         srcu_read_unlock(&kvm->srcu, srcu_idx);
3656  out:
3657         mutex_unlock(&kvm->lock);
3658         return err;
3659
3660  up_out:
3661         up_read(&current->mm->mmap_sem);
3662         goto out_srcu;
3663 }
3664
3665 #ifdef CONFIG_KVM_XICS
3666 /*
3667  * Allocate a per-core structure for managing state about which cores are
3668  * running in the host versus the guest and for exchanging data between
3669  * real mode KVM and CPU running in the host.
3670  * This is only done for the first VM.
3671  * The allocated structure stays even if all VMs have stopped.
3672  * It is only freed when the kvm-hv module is unloaded.
3673  * It's OK for this routine to fail, we just don't support host
3674  * core operations like redirecting H_IPI wakeups.
3675  */
3676 void kvmppc_alloc_host_rm_ops(void)
3677 {
3678         struct kvmppc_host_rm_ops *ops;
3679         unsigned long l_ops;
3680         int cpu, core;
3681         int size;
3682
3683         /* Not the first time here ? */
3684         if (kvmppc_host_rm_ops_hv != NULL)
3685                 return;
3686
3687         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3688         if (!ops)
3689                 return;
3690
3691         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3692         ops->rm_core = kzalloc(size, GFP_KERNEL);
3693
3694         if (!ops->rm_core) {
3695                 kfree(ops);
3696                 return;
3697         }
3698
3699         cpus_read_lock();
3700
3701         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3702                 if (!cpu_online(cpu))
3703                         continue;
3704
3705                 core = cpu >> threads_shift;
3706                 ops->rm_core[core].rm_state.in_host = 1;
3707         }
3708
3709         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3710
3711         /*
3712          * Make the contents of the kvmppc_host_rm_ops structure visible
3713          * to other CPUs before we assign it to the global variable.
3714          * Do an atomic assignment (no locks used here), but if someone
3715          * beats us to it, just free our copy and return.
3716          */
3717         smp_wmb();
3718         l_ops = (unsigned long) ops;
3719
3720         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3721                 cpus_read_unlock();
3722                 kfree(ops->rm_core);
3723                 kfree(ops);
3724                 return;
3725         }
3726
3727         cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
3728                                              "ppc/kvm_book3s:prepare",
3729                                              kvmppc_set_host_core,
3730                                              kvmppc_clear_host_core);
3731         cpus_read_unlock();
3732 }
3733
3734 void kvmppc_free_host_rm_ops(void)
3735 {
3736         if (kvmppc_host_rm_ops_hv) {
3737                 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
3738                 kfree(kvmppc_host_rm_ops_hv->rm_core);
3739                 kfree(kvmppc_host_rm_ops_hv);
3740                 kvmppc_host_rm_ops_hv = NULL;
3741         }
3742 }
3743 #endif
3744
3745 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3746 {
3747         unsigned long lpcr, lpid;
3748         char buf[32];
3749         int ret;
3750
3751         /* Allocate the guest's logical partition ID */
3752
3753         lpid = kvmppc_alloc_lpid();
3754         if ((long)lpid < 0)
3755                 return -ENOMEM;
3756         kvm->arch.lpid = lpid;
3757
3758         kvmppc_alloc_host_rm_ops();
3759
3760         /*
3761          * Since we don't flush the TLB when tearing down a VM,
3762          * and this lpid might have previously been used,
3763          * make sure we flush on each core before running the new VM.
3764          * On POWER9, the tlbie in mmu_partition_table_set_entry()
3765          * does this flush for us.
3766          */
3767         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3768                 cpumask_setall(&kvm->arch.need_tlb_flush);
3769
3770         /* Start out with the default set of hcalls enabled */
3771         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3772                sizeof(kvm->arch.enabled_hcalls));
3773
3774         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3775                 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3776
3777         /* Init LPCR for virtual RMA mode */
3778         kvm->arch.host_lpid = mfspr(SPRN_LPID);
3779         kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3780         lpcr &= LPCR_PECE | LPCR_LPES;
3781         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3782                 LPCR_VPM0 | LPCR_VPM1;
3783         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3784                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3785         /* On POWER8 turn on online bit to enable PURR/SPURR */
3786         if (cpu_has_feature(CPU_FTR_ARCH_207S))
3787                 lpcr |= LPCR_ONL;
3788         /*
3789          * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
3790          * Set HVICE bit to enable hypervisor virtualization interrupts.
3791          * Set HEIC to prevent OS interrupts to go to hypervisor (should
3792          * be unnecessary but better safe than sorry in case we re-enable
3793          * EE in HV mode with this LPCR still set)
3794          */
3795         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3796                 lpcr &= ~LPCR_VPM0;
3797                 lpcr |= LPCR_HVICE | LPCR_HEIC;
3798
3799                 /*
3800                  * If xive is enabled, we route 0x500 interrupts directly
3801                  * to the guest.
3802                  */
3803                 if (xive_enabled())
3804                         lpcr |= LPCR_LPES;
3805         }
3806
3807         /*
3808          * For now, if the host uses radix, the guest must be radix.
3809          */
3810         if (radix_enabled()) {
3811                 kvm->arch.radix = 1;
3812                 lpcr &= ~LPCR_VPM1;
3813                 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
3814                 ret = kvmppc_init_vm_radix(kvm);
3815                 if (ret) {
3816                         kvmppc_free_lpid(kvm->arch.lpid);
3817                         return ret;
3818                 }
3819                 kvmppc_setup_partition_table(kvm);
3820         }
3821
3822         kvm->arch.lpcr = lpcr;
3823
3824         /* Initialization for future HPT resizes */
3825         kvm->arch.resize_hpt = NULL;
3826
3827         /*
3828          * Work out how many sets the TLB has, for the use of
3829          * the TLB invalidation loop in book3s_hv_rmhandlers.S.
3830          */
3831         if (kvm_is_radix(kvm))
3832                 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;     /* 128 */
3833         else if (cpu_has_feature(CPU_FTR_ARCH_300))
3834                 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;      /* 256 */
3835         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
3836                 kvm->arch.tlb_sets = POWER8_TLB_SETS;           /* 512 */
3837         else
3838                 kvm->arch.tlb_sets = POWER7_TLB_SETS;           /* 128 */
3839
3840         /*
3841          * Track that we now have a HV mode VM active. This blocks secondary
3842          * CPU threads from coming online.
3843          * On POWER9, we only need to do this for HPT guests on a radix
3844          * host, which is not yet supported.
3845          */
3846         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3847                 kvm_hv_vm_activated();
3848
3849         /*
3850          * Initialize smt_mode depending on processor.
3851          * POWER8 and earlier have to use "strict" threading, where
3852          * all vCPUs in a vcore have to run on the same (sub)core,
3853          * whereas on POWER9 the threads can each run a different
3854          * guest.
3855          */
3856         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3857                 kvm->arch.smt_mode = threads_per_subcore;
3858         else
3859                 kvm->arch.smt_mode = 1;
3860         kvm->arch.emul_smt_mode = 1;
3861
3862         /*
3863          * Create a debugfs directory for the VM
3864          */
3865         snprintf(buf, sizeof(buf), "vm%d", current->pid);
3866         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3867         if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3868                 kvmppc_mmu_debugfs_init(kvm);
3869
3870         return 0;
3871 }
3872
3873 static void kvmppc_free_vcores(struct kvm *kvm)
3874 {
3875         long int i;
3876
3877         for (i = 0; i < KVM_MAX_VCORES; ++i)
3878                 kfree(kvm->arch.vcores[i]);
3879         kvm->arch.online_vcores = 0;
3880 }
3881
3882 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3883 {
3884         debugfs_remove_recursive(kvm->arch.debugfs_dir);
3885
3886         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3887                 kvm_hv_vm_deactivated();
3888
3889         kvmppc_free_vcores(kvm);
3890
3891         kvmppc_free_lpid(kvm->arch.lpid);
3892
3893         if (kvm_is_radix(kvm))
3894                 kvmppc_free_radix(kvm);
3895         else
3896                 kvmppc_free_hpt(&kvm->arch.hpt);
3897
3898         kvmppc_free_pimap(kvm);
3899 }
3900
3901 /* We don't need to emulate any privileged instructions or dcbz */
3902 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3903                                      unsigned int inst, int *advance)
3904 {
3905         return EMULATE_FAIL;
3906 }
3907
3908 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3909                                         ulong spr_val)
3910 {
3911         return EMULATE_FAIL;
3912 }
3913
3914 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3915                                         ulong *spr_val)
3916 {
3917         return EMULATE_FAIL;
3918 }
3919
3920 static int kvmppc_core_check_processor_compat_hv(void)
3921 {
3922         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3923             !cpu_has_feature(CPU_FTR_ARCH_206))
3924                 return -EIO;
3925
3926         return 0;
3927 }
3928
3929 #ifdef CONFIG_KVM_XICS
3930
3931 void kvmppc_free_pimap(struct kvm *kvm)
3932 {
3933         kfree(kvm->arch.pimap);
3934 }
3935
3936 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
3937 {
3938         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
3939 }
3940
3941 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3942 {
3943         struct irq_desc *desc;
3944         struct kvmppc_irq_map *irq_map;
3945         struct kvmppc_passthru_irqmap *pimap;
3946         struct irq_chip *chip;
3947         int i, rc = 0;
3948
3949         if (!kvm_irq_bypass)
3950                 return 1;
3951
3952         desc = irq_to_desc(host_irq);
3953         if (!desc)
3954                 return -EIO;
3955
3956         mutex_lock(&kvm->lock);
3957
3958         pimap = kvm->arch.pimap;
3959         if (pimap == NULL) {
3960                 /* First call, allocate structure to hold IRQ map */
3961                 pimap = kvmppc_alloc_pimap();
3962                 if (pimap == NULL) {
3963                         mutex_unlock(&kvm->lock);
3964                         return -ENOMEM;
3965                 }
3966                 kvm->arch.pimap = pimap;
3967         }
3968
3969         /*
3970          * For now, we only support interrupts for which the EOI operation
3971          * is an OPAL call followed by a write to XIRR, since that's
3972          * what our real-mode EOI code does, or a XIVE interrupt
3973          */
3974         chip = irq_data_get_irq_chip(&desc->irq_data);
3975         if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
3976                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
3977                         host_irq, guest_gsi);
3978                 mutex_unlock(&kvm->lock);
3979                 return -ENOENT;
3980         }
3981
3982         /*
3983          * See if we already have an entry for this guest IRQ number.
3984          * If it's mapped to a hardware IRQ number, that's an error,
3985          * otherwise re-use this entry.
3986          */
3987         for (i = 0; i < pimap->n_mapped; i++) {
3988                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
3989                         if (pimap->mapped[i].r_hwirq) {
3990                                 mutex_unlock(&kvm->lock);
3991                                 return -EINVAL;
3992                         }
3993                         break;
3994                 }
3995         }
3996
3997         if (i == KVMPPC_PIRQ_MAPPED) {
3998                 mutex_unlock(&kvm->lock);
3999                 return -EAGAIN;         /* table is full */
4000         }
4001
4002         irq_map = &pimap->mapped[i];
4003
4004         irq_map->v_hwirq = guest_gsi;
4005         irq_map->desc = desc;
4006
4007         /*
4008          * Order the above two stores before the next to serialize with
4009          * the KVM real mode handler.
4010          */
4011         smp_wmb();
4012         irq_map->r_hwirq = desc->irq_data.hwirq;
4013
4014         if (i == pimap->n_mapped)
4015                 pimap->n_mapped++;
4016
4017         if (xive_enabled())
4018                 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
4019         else
4020                 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
4021         if (rc)
4022                 irq_map->r_hwirq = 0;
4023
4024         mutex_unlock(&kvm->lock);
4025
4026         return 0;
4027 }
4028
4029 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
4030 {
4031         struct irq_desc *desc;
4032         struct kvmppc_passthru_irqmap *pimap;
4033         int i, rc = 0;
4034
4035         if (!kvm_irq_bypass)
4036                 return 0;
4037
4038         desc = irq_to_desc(host_irq);
4039         if (!desc)
4040                 return -EIO;
4041
4042         mutex_lock(&kvm->lock);
4043         if (!kvm->arch.pimap)
4044                 goto unlock;
4045
4046         pimap = kvm->arch.pimap;
4047
4048         for (i = 0; i < pimap->n_mapped; i++) {
4049                 if (guest_gsi == pimap->mapped[i].v_hwirq)
4050                         break;
4051         }
4052
4053         if (i == pimap->n_mapped) {
4054                 mutex_unlock(&kvm->lock);
4055                 return -ENODEV;
4056         }
4057
4058         if (xive_enabled())
4059                 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
4060         else
4061                 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
4062
4063         /* invalidate the entry (what do do on error from the above ?) */
4064         pimap->mapped[i].r_hwirq = 0;
4065
4066         /*
4067          * We don't free this structure even when the count goes to
4068          * zero. The structure is freed when we destroy the VM.
4069          */
4070  unlock:
4071         mutex_unlock(&kvm->lock);
4072         return rc;
4073 }
4074
4075 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
4076                                              struct irq_bypass_producer *prod)
4077 {
4078         int ret = 0;
4079         struct kvm_kernel_irqfd *irqfd =
4080                 container_of(cons, struct kvm_kernel_irqfd, consumer);
4081
4082         irqfd->producer = prod;
4083
4084         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4085         if (ret)
4086                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
4087                         prod->irq, irqfd->gsi, ret);
4088
4089         return ret;
4090 }
4091
4092 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
4093                                               struct irq_bypass_producer *prod)
4094 {
4095         int ret;
4096         struct kvm_kernel_irqfd *irqfd =
4097                 container_of(cons, struct kvm_kernel_irqfd, consumer);
4098
4099         irqfd->producer = NULL;
4100
4101         /*
4102          * When producer of consumer is unregistered, we change back to
4103          * default external interrupt handling mode - KVM real mode
4104          * will switch back to host.
4105          */
4106         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
4107         if (ret)
4108                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
4109                         prod->irq, irqfd->gsi, ret);
4110 }
4111 #endif
4112
4113 static long kvm_arch_vm_ioctl_hv(struct file *filp,
4114                                  unsigned int ioctl, unsigned long arg)
4115 {
4116         struct kvm *kvm __maybe_unused = filp->private_data;
4117         void __user *argp = (void __user *)arg;
4118         long r;
4119
4120         switch (ioctl) {
4121
4122         case KVM_PPC_ALLOCATE_HTAB: {
4123                 u32 htab_order;
4124
4125                 r = -EFAULT;
4126                 if (get_user(htab_order, (u32 __user *)argp))
4127                         break;
4128                 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
4129                 if (r)
4130                         break;
4131                 r = 0;
4132                 break;
4133         }
4134
4135         case KVM_PPC_GET_HTAB_FD: {
4136                 struct kvm_get_htab_fd ghf;
4137
4138                 r = -EFAULT;
4139                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
4140                         break;
4141                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
4142                 break;
4143         }
4144
4145         case KVM_PPC_RESIZE_HPT_PREPARE: {
4146                 struct kvm_ppc_resize_hpt rhpt;
4147
4148                 r = -EFAULT;
4149                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4150                         break;
4151
4152                 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
4153                 break;
4154         }
4155
4156         case KVM_PPC_RESIZE_HPT_COMMIT: {
4157                 struct kvm_ppc_resize_hpt rhpt;
4158
4159                 r = -EFAULT;
4160                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
4161                         break;
4162
4163                 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
4164                 break;
4165         }
4166
4167         default:
4168                 r = -ENOTTY;
4169         }
4170
4171         return r;
4172 }
4173
4174 /*
4175  * List of hcall numbers to enable by default.
4176  * For compatibility with old userspace, we enable by default
4177  * all hcalls that were implemented before the hcall-enabling
4178  * facility was added.  Note this list should not include H_RTAS.
4179  */
4180 static unsigned int default_hcall_list[] = {
4181         H_REMOVE,
4182         H_ENTER,
4183         H_READ,
4184         H_PROTECT,
4185         H_BULK_REMOVE,
4186         H_GET_TCE,
4187         H_PUT_TCE,
4188         H_SET_DABR,
4189         H_SET_XDABR,
4190         H_CEDE,
4191         H_PROD,
4192         H_CONFER,
4193         H_REGISTER_VPA,
4194 #ifdef CONFIG_KVM_XICS
4195         H_EOI,
4196         H_CPPR,
4197         H_IPI,
4198         H_IPOLL,
4199         H_XIRR,
4200         H_XIRR_X,
4201 #endif
4202         0
4203 };
4204
4205 static void init_default_hcalls(void)
4206 {
4207         int i;
4208         unsigned int hcall;
4209
4210         for (i = 0; default_hcall_list[i]; ++i) {
4211                 hcall = default_hcall_list[i];
4212                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
4213                 __set_bit(hcall / 4, default_enabled_hcalls);
4214         }
4215 }
4216
4217 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
4218 {
4219         unsigned long lpcr;
4220         int radix;
4221
4222         /* If not on a POWER9, reject it */
4223         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4224                 return -ENODEV;
4225
4226         /* If any unknown flags set, reject it */
4227         if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
4228                 return -EINVAL;
4229
4230         /* We can't change a guest to/from radix yet */
4231         radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
4232         if (radix != kvm_is_radix(kvm))
4233                 return -EINVAL;
4234
4235         /* GR (guest radix) bit in process_table field must match */
4236         if (!!(cfg->process_table & PATB_GR) != radix)
4237                 return -EINVAL;
4238
4239         /* Process table size field must be reasonable, i.e. <= 24 */
4240         if ((cfg->process_table & PRTS_MASK) > 24)
4241                 return -EINVAL;
4242
4243         mutex_lock(&kvm->lock);
4244         kvm->arch.process_table = cfg->process_table;
4245         kvmppc_setup_partition_table(kvm);
4246
4247         lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
4248         kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
4249         mutex_unlock(&kvm->lock);
4250
4251         return 0;
4252 }
4253
4254 static struct kvmppc_ops kvm_ops_hv = {
4255         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
4256         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
4257         .get_one_reg = kvmppc_get_one_reg_hv,
4258         .set_one_reg = kvmppc_set_one_reg_hv,
4259         .vcpu_load   = kvmppc_core_vcpu_load_hv,
4260         .vcpu_put    = kvmppc_core_vcpu_put_hv,
4261         .set_msr     = kvmppc_set_msr_hv,
4262         .vcpu_run    = kvmppc_vcpu_run_hv,
4263         .vcpu_create = kvmppc_core_vcpu_create_hv,
4264         .vcpu_free   = kvmppc_core_vcpu_free_hv,
4265         .check_requests = kvmppc_core_check_requests_hv,
4266         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
4267         .flush_memslot  = kvmppc_core_flush_memslot_hv,
4268         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
4269         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
4270         .unmap_hva = kvm_unmap_hva_hv,
4271         .unmap_hva_range = kvm_unmap_hva_range_hv,
4272         .age_hva  = kvm_age_hva_hv,
4273         .test_age_hva = kvm_test_age_hva_hv,
4274         .set_spte_hva = kvm_set_spte_hva_hv,
4275         .mmu_destroy  = kvmppc_mmu_destroy_hv,
4276         .free_memslot = kvmppc_core_free_memslot_hv,
4277         .create_memslot = kvmppc_core_create_memslot_hv,
4278         .init_vm =  kvmppc_core_init_vm_hv,
4279         .destroy_vm = kvmppc_core_destroy_vm_hv,
4280         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
4281         .emulate_op = kvmppc_core_emulate_op_hv,
4282         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
4283         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
4284         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
4285         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
4286         .hcall_implemented = kvmppc_hcall_impl_hv,
4287 #ifdef CONFIG_KVM_XICS
4288         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
4289         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
4290 #endif
4291         .configure_mmu = kvmhv_configure_mmu,
4292         .get_rmmu_info = kvmhv_get_rmmu_info,
4293         .set_smt_mode = kvmhv_set_smt_mode,
4294 };
4295
4296 static int kvm_init_subcore_bitmap(void)
4297 {
4298         int i, j;
4299         int nr_cores = cpu_nr_cores();
4300         struct sibling_subcore_state *sibling_subcore_state;
4301
4302         for (i = 0; i < nr_cores; i++) {
4303                 int first_cpu = i * threads_per_core;
4304                 int node = cpu_to_node(first_cpu);
4305
4306                 /* Ignore if it is already allocated. */
4307                 if (paca[first_cpu].sibling_subcore_state)
4308                         continue;
4309
4310                 sibling_subcore_state =
4311                         kmalloc_node(sizeof(struct sibling_subcore_state),
4312                                                         GFP_KERNEL, node);
4313                 if (!sibling_subcore_state)
4314                         return -ENOMEM;
4315
4316                 memset(sibling_subcore_state, 0,
4317                                 sizeof(struct sibling_subcore_state));
4318
4319                 for (j = 0; j < threads_per_core; j++) {
4320                         int cpu = first_cpu + j;
4321
4322                         paca[cpu].sibling_subcore_state = sibling_subcore_state;
4323                 }
4324         }
4325         return 0;
4326 }
4327
4328 static int kvmppc_radix_possible(void)
4329 {
4330         return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
4331 }
4332
4333 static int kvmppc_book3s_init_hv(void)
4334 {
4335         int r;
4336         /*
4337          * FIXME!! Do we need to check on all cpus ?
4338          */
4339         r = kvmppc_core_check_processor_compat_hv();
4340         if (r < 0)
4341                 return -ENODEV;
4342
4343         r = kvm_init_subcore_bitmap();
4344         if (r)
4345                 return r;
4346
4347         /*
4348          * We need a way of accessing the XICS interrupt controller,
4349          * either directly, via paca[cpu].kvm_hstate.xics_phys, or
4350          * indirectly, via OPAL.
4351          */
4352 #ifdef CONFIG_SMP
4353         if (!xive_enabled() && !local_paca->kvm_hstate.xics_phys) {
4354                 struct device_node *np;
4355
4356                 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
4357                 if (!np) {
4358                         pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
4359                         return -ENODEV;
4360                 }
4361                 /* presence of intc confirmed - node can be dropped again */
4362                 of_node_put(np);
4363         }
4364 #endif
4365
4366         kvm_ops_hv.owner = THIS_MODULE;
4367         kvmppc_hv_ops = &kvm_ops_hv;
4368
4369         init_default_hcalls();
4370
4371         init_vcore_lists();
4372
4373         r = kvmppc_mmu_hv_init();
4374         if (r)
4375                 return r;
4376
4377         if (kvmppc_radix_possible())
4378                 r = kvmppc_radix_init();
4379         return r;
4380 }
4381
4382 static void kvmppc_book3s_exit_hv(void)
4383 {
4384         kvmppc_free_host_rm_ops();
4385         if (kvmppc_radix_possible())
4386                 kvmppc_radix_exit();
4387         kvmppc_hv_ops = NULL;
4388 }
4389
4390 module_init(kvmppc_book3s_init_hv);
4391 module_exit(kvmppc_book3s_exit_hv);
4392 MODULE_LICENSE("GPL");
4393 MODULE_ALIAS_MISCDEV(KVM_MINOR);
4394 MODULE_ALIAS("devname:kvm");
4395