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.h>
  26 #include <linux/delay.h>
  27 #include <linux/export.h>
  28 #include <linux/fs.h>
  29 #include <linux/anon_inodes.h>
  30 #include <linux/cpu.h>
  31 #include <linux/cpumask.h>
  32 #include <linux/spinlock.h>
  33 #include <linux/page-flags.h>
  34 #include <linux/srcu.h>
  35 #include <linux/miscdevice.h>
  36 #include <linux/debugfs.h>
  37
  38 #include <asm/reg.h>
  39 #include <asm/cputable.h>
  40 #include <asm/cacheflush.h>
  41 #include <asm/tlbflush.h>
  42 #include <linux/uaccess.h>
  43 #include <asm/io.h>
  44 #include <asm/kvm_ppc.h>
  45 #include <asm/kvm_book3s.h>
  46 #include <asm/mmu_context.h>
  47 #include <asm/lppaca.h>
  48 #include <asm/processor.h>
  49 #include <asm/cputhreads.h>
  50 #include <asm/page.h>
  51 #include <asm/hvcall.h>
  52 #include <asm/switch_to.h>
  53 #include <asm/smp.h>
  54 #include <asm/dbell.h>
  55 #include <asm/hmi.h>
  56 #include <asm/pnv-pci.h>
  57 #include <asm/mmu.h>
  58 #include <asm/opal.h>
  59 #include <asm/xics.h>
  60 #include <linux/gfp.h>
  61 #include <linux/vmalloc.h>
  62 #include <linux/highmem.h>
  63 #include <linux/hugetlb.h>
  64 #include <linux/kvm_irqfd.h>
  65 #include <linux/irqbypass.h>
  66 #include <linux/module.h>
  67 #include <linux/compiler.h>
  68 #include <linux/of.h>
  69
  70 #include "book3s.h"
  71
  72 #define CREATE_TRACE_POINTS
  73 #include "trace_hv.h"
  74
  75 /* #define EXIT_DEBUG */
  76 /* #define EXIT_DEBUG_SIMPLE */
  77 /* #define EXIT_DEBUG_INT */
  78
  79 /* Used to indicate that a guest page fault needs to be handled */
  80 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  81 /* Used to indicate that a guest passthrough interrupt needs to be handled */
  82 #define RESUME_PASSTHROUGH      (RESUME_GUEST | RESUME_FLAG_ARCH2)
  83
  84 /* Used as a "null" value for timebase values */
  85 #define TB_NIL  (~(u64)0)
  86
  87 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
  88
  89 static int dynamic_mt_modes = 6;
  90 module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
  91 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
  92 static int target_smt_mode;
  93 module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
  94 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
  95
  96 #ifdef CONFIG_KVM_XICS
  97 static struct kernel_param_ops module_param_ops = {
  98         .set = param_set_int,
  99         .get = param_get_int,
 100 };
 101
 102 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass,
 103                                                         S_IRUGO | S_IWUSR);
 104 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
 105
 106 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect,
 107                                                         S_IRUGO | S_IWUSR);
 108 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
 109 #endif
 110
 111 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
 112 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
 113
 114 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
 115                 int *ip)
 116 {
 117         int i = *ip;
 118         struct kvm_vcpu *vcpu;
 119
 120         while (++i < MAX_SMT_THREADS) {
 121                 vcpu = READ_ONCE(vc->runnable_threads[i]);
 122                 if (vcpu) {
 123                         *ip = i;
 124                         return vcpu;
 125                 }
 126         }
 127         return NULL;
 128 }
 129
 130 /* Used to traverse the list of runnable threads for a given vcore */
 131 #define for_each_runnable_thread(i, vcpu, vc) \
 132         for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
 133
 134 static bool kvmppc_ipi_thread(int cpu)
 135 {
 136         unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
 137
 138         /* On POWER9 we can use msgsnd to IPI any cpu */
 139         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
 140                 msg |= get_hard_smp_processor_id(cpu);
 141                 smp_mb();
 142                 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 143                 return true;
 144         }
 145
 146         /* On POWER8 for IPIs to threads in the same core, use msgsnd */
 147         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
 148                 preempt_disable();
 149                 if (cpu_first_thread_sibling(cpu) ==
 150                     cpu_first_thread_sibling(smp_processor_id())) {
 151                         msg |= cpu_thread_in_core(cpu);
 152                         smp_mb();
 153                         __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 154                         preempt_enable();
 155                         return true;
 156                 }
 157                 preempt_enable();
 158         }
 159
 160 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
 161         if (cpu >= 0 && cpu < nr_cpu_ids) {
 162                 if (paca[cpu].kvm_hstate.xics_phys) {
 163                         xics_wake_cpu(cpu);
 164                         return true;
 165                 }
 166                 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
 167                 return true;
 168         }
 169 #endif
 170
 171         return false;
 172 }
 173
 174 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
 175 {
 176         int cpu;
 177         struct swait_queue_head *wqp;
 178
 179         wqp = kvm_arch_vcpu_wq(vcpu);
 180         if (swait_active(wqp)) {
 181                 swake_up(wqp);
 182                 ++vcpu->stat.halt_wakeup;
 183         }
 184
 185         cpu = READ_ONCE(vcpu->arch.thread_cpu);
 186         if (cpu >= 0 && kvmppc_ipi_thread(cpu))
 187                 return;
 188
 189         /* CPU points to the first thread of the core */
 190         cpu = vcpu->cpu;
 191         if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
 192                 smp_send_reschedule(cpu);
 193 }
 194
 195 /*
 196  * We use the vcpu_load/put functions to measure stolen time.
 197  * Stolen time is counted as time when either the vcpu is able to
 198  * run as part of a virtual core, but the task running the vcore
 199  * is preempted or sleeping, or when the vcpu needs something done
 200  * in the kernel by the task running the vcpu, but that task is
 201  * preempted or sleeping.  Those two things have to be counted
 202  * separately, since one of the vcpu tasks will take on the job
 203  * of running the core, and the other vcpu tasks in the vcore will
 204  * sleep waiting for it to do that, but that sleep shouldn't count
 205  * as stolen time.
 206  *
 207  * Hence we accumulate stolen time when the vcpu can run as part of
 208  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 209  * needs its task to do other things in the kernel (for example,
 210  * service a page fault) in busy_stolen.  We don't accumulate
 211  * stolen time for a vcore when it is inactive, or for a vcpu
 212  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 213  * a misnomer; it means that the vcpu task is not executing in
 214  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 215  * the kernel.  We don't have any way of dividing up that time
 216  * between time that the vcpu is genuinely stopped, time that
 217  * the task is actively working on behalf of the vcpu, and time
 218  * that the task is preempted, so we don't count any of it as
 219  * stolen.
 220  *
 221  * Updates to busy_stolen are protected by arch.tbacct_lock;
 222  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
 223  * lock.  The stolen times are measured in units of timebase ticks.
 224  * (Note that the != TB_NIL checks below are purely defensive;
 225  * they should never fail.)
 226  */
 227
 228 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
 229 {
 230         unsigned long flags;
 231
 232         spin_lock_irqsave(&vc->stoltb_lock, flags);
 233         vc->preempt_tb = mftb();
 234         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 235 }
 236
 237 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
 238 {
 239         unsigned long flags;
 240
 241         spin_lock_irqsave(&vc->stoltb_lock, flags);
 242         if (vc->preempt_tb != TB_NIL) {
 243                 vc->stolen_tb += mftb() - vc->preempt_tb;
 244                 vc->preempt_tb = TB_NIL;
 245         }
 246         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 247 }
 248
 249 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 250 {
 251         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 252         unsigned long flags;
 253
 254         /*
 255          * We can test vc->runner without taking the vcore lock,
 256          * because only this task ever sets vc->runner to this
 257          * vcpu, and once it is set to this vcpu, only this task
 258          * ever sets it to NULL.
 259          */
 260         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 261                 kvmppc_core_end_stolen(vc);
 262
 263         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 264         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 265             vcpu->arch.busy_preempt != TB_NIL) {
 266                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 267                 vcpu->arch.busy_preempt = TB_NIL;
 268         }
 269         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 270 }
 271
 272 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 273 {
 274         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 275         unsigned long flags;
 276
 277         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 278                 kvmppc_core_start_stolen(vc);
 279
 280         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 281         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 282                 vcpu->arch.busy_preempt = mftb();
 283         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 284 }
 285
 286 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
 287 {
 288         /*
 289          * Check for illegal transactional state bit combination
 290          * and if we find it, force the TS field to a safe state.
 291          */
 292         if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
 293                 msr &= ~MSR_TS_MASK;
 294         vcpu->arch.shregs.msr = msr;
 295         kvmppc_end_cede(vcpu);
 296 }
 297
 298 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 299 {
 300         vcpu->arch.pvr = pvr;
 301 }
 302
 303 /* Dummy value used in computing PCR value below */
 304 #define PCR_ARCH_300    (PCR_ARCH_207 << 1)
 305
 306 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 307 {
 308         unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
 309         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 310
 311         /* We can (emulate) our own architecture version and anything older */
 312         if (cpu_has_feature(CPU_FTR_ARCH_300))
 313                 host_pcr_bit = PCR_ARCH_300;
 314         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
 315                 host_pcr_bit = PCR_ARCH_207;
 316         else if (cpu_has_feature(CPU_FTR_ARCH_206))
 317                 host_pcr_bit = PCR_ARCH_206;
 318         else
 319                 host_pcr_bit = PCR_ARCH_205;
 320
 321         /* Determine lowest PCR bit needed to run guest in given PVR level */
 322         guest_pcr_bit = host_pcr_bit;
 323         if (arch_compat) {
 324                 switch (arch_compat) {
 325                 case PVR_ARCH_205:
 326                         guest_pcr_bit = PCR_ARCH_205;
 327                         break;
 328                 case PVR_ARCH_206:
 329                 case PVR_ARCH_206p:
 330                         guest_pcr_bit = PCR_ARCH_206;
 331                         break;
 332                 case PVR_ARCH_207:
 333                         guest_pcr_bit = PCR_ARCH_207;
 334                         break;
 335                 case PVR_ARCH_300:
 336                         guest_pcr_bit = PCR_ARCH_300;
 337                         break;
 338                 default:
 339                         return -EINVAL;
 340                 }
 341         }
 342
 343         /* Check requested PCR bits don't exceed our capabilities */
 344         if (guest_pcr_bit > host_pcr_bit)
 345                 return -EINVAL;
 346
 347         spin_lock(&vc->lock);
 348         vc->arch_compat = arch_compat;
 349         /* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
 350         vc->pcr = host_pcr_bit - guest_pcr_bit;
 351         spin_unlock(&vc->lock);
 352
 353         return 0;
 354 }
 355
 356 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 357 {
 358         int r;
 359
 360         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 361         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 362                vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
 363         for (r = 0; r < 16; ++r)
 364                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 365                        r, kvmppc_get_gpr(vcpu, r),
 366                        r+16, kvmppc_get_gpr(vcpu, r+16));
 367         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 368                vcpu->arch.ctr, vcpu->arch.lr);
 369         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 370                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 371         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 372                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 373         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 374                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 375         pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
 376                vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
 377         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 378         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 379                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 380         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 381         for (r = 0; r < vcpu->arch.slb_max; ++r)
 382                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 383                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 384         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 385                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 386                vcpu->arch.last_inst);
 387 }
 388
 389 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 390 {
 391         struct kvm_vcpu *ret;
 392
 393         mutex_lock(&kvm->lock);
 394         ret = kvm_get_vcpu_by_id(kvm, id);
 395         mutex_unlock(&kvm->lock);
 396         return ret;
 397 }
 398
 399 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 400 {
 401         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 402         vpa->yield_count = cpu_to_be32(1);
 403 }
 404
 405 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 406                    unsigned long addr, unsigned long len)
 407 {
 408         /* check address is cacheline aligned */
 409         if (addr & (L1_CACHE_BYTES - 1))
 410                 return -EINVAL;
 411         spin_lock(&vcpu->arch.vpa_update_lock);
 412         if (v->next_gpa != addr || v->len != len) {
 413                 v->next_gpa = addr;
 414                 v->len = addr ? len : 0;
 415                 v->update_pending = 1;
 416         }
 417         spin_unlock(&vcpu->arch.vpa_update_lock);
 418         return 0;
 419 }
 420
 421 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 422 struct reg_vpa {
 423         u32 dummy;
 424         union {
 425                 __be16 hword;
 426                 __be32 word;
 427         } length;
 428 };
 429
 430 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 431 {
 432         if (vpap->update_pending)
 433                 return vpap->next_gpa != 0;
 434         return vpap->pinned_addr != NULL;
 435 }
 436
 437 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 438                                        unsigned long flags,
 439                                        unsigned long vcpuid, unsigned long vpa)
 440 {
 441         struct kvm *kvm = vcpu->kvm;
 442         unsigned long len, nb;
 443         void *va;
 444         struct kvm_vcpu *tvcpu;
 445         int err;
 446         int subfunc;
 447         struct kvmppc_vpa *vpap;
 448
 449         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 450         if (!tvcpu)
 451                 return H_PARAMETER;
 452
 453         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 454         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 455             subfunc == H_VPA_REG_SLB) {
 456                 /* Registering new area - address must be cache-line aligned */
 457                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 458                         return H_PARAMETER;
 459
 460                 /* convert logical addr to kernel addr and read length */
 461                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 462                 if (va == NULL)
 463                         return H_PARAMETER;
 464                 if (subfunc == H_VPA_REG_VPA)
 465                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 466                 else
 467                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 468                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 469
 470                 /* Check length */
 471                 if (len > nb || len < sizeof(struct reg_vpa))
 472                         return H_PARAMETER;
 473         } else {
 474                 vpa = 0;
 475                 len = 0;
 476         }
 477
 478         err = H_PARAMETER;
 479         vpap = NULL;
 480         spin_lock(&tvcpu->arch.vpa_update_lock);
 481
 482         switch (subfunc) {
 483         case H_VPA_REG_VPA:             /* register VPA */
 484                 if (len < sizeof(struct lppaca))
 485                         break;
 486                 vpap = &tvcpu->arch.vpa;
 487                 err = 0;
 488                 break;
 489
 490         case H_VPA_REG_DTL:             /* register DTL */
 491                 if (len < sizeof(struct dtl_entry))
 492                         break;
 493                 len -= len % sizeof(struct dtl_entry);
 494
 495                 /* Check that they have previously registered a VPA */
 496                 err = H_RESOURCE;
 497                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 498                         break;
 499
 500                 vpap = &tvcpu->arch.dtl;
 501                 err = 0;
 502                 break;
 503
 504         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 505                 /* Check that they have previously registered a VPA */
 506                 err = H_RESOURCE;
 507                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 508                         break;
 509
 510                 vpap = &tvcpu->arch.slb_shadow;
 511                 err = 0;
 512                 break;
 513
 514         case H_VPA_DEREG_VPA:           /* deregister VPA */
 515                 /* Check they don't still have a DTL or SLB buf registered */
 516                 err = H_RESOURCE;
 517                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 518                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 519                         break;
 520
 521                 vpap = &tvcpu->arch.vpa;
 522                 err = 0;
 523                 break;
 524
 525         case H_VPA_DEREG_DTL:           /* deregister DTL */
 526                 vpap = &tvcpu->arch.dtl;
 527                 err = 0;
 528                 break;
 529
 530         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 531                 vpap = &tvcpu->arch.slb_shadow;
 532                 err = 0;
 533                 break;
 534         }
 535
 536         if (vpap) {
 537                 vpap->next_gpa = vpa;
 538                 vpap->len = len;
 539                 vpap->update_pending = 1;
 540         }
 541
 542         spin_unlock(&tvcpu->arch.vpa_update_lock);
 543
 544         return err;
 545 }
 546
 547 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 548 {
 549         struct kvm *kvm = vcpu->kvm;
 550         void *va;
 551         unsigned long nb;
 552         unsigned long gpa;
 553
 554         /*
 555          * We need to pin the page pointed to by vpap->next_gpa,
 556          * but we can't call kvmppc_pin_guest_page under the lock
 557          * as it does get_user_pages() and down_read().  So we
 558          * have to drop the lock, pin the page, then get the lock
 559          * again and check that a new area didn't get registered
 560          * in the meantime.
 561          */
 562         for (;;) {
 563                 gpa = vpap->next_gpa;
 564                 spin_unlock(&vcpu->arch.vpa_update_lock);
 565                 va = NULL;
 566                 nb = 0;
 567                 if (gpa)
 568                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 569                 spin_lock(&vcpu->arch.vpa_update_lock);
 570                 if (gpa == vpap->next_gpa)
 571                         break;
 572                 /* sigh... unpin that one and try again */
 573                 if (va)
 574                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 575         }
 576
 577         vpap->update_pending = 0;
 578         if (va && nb < vpap->len) {
 579                 /*
 580                  * If it's now too short, it must be that userspace
 581                  * has changed the mappings underlying guest memory,
 582                  * so unregister the region.
 583                  */
 584                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 585                 va = NULL;
 586         }
 587         if (vpap->pinned_addr)
 588                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 589                                         vpap->dirty);
 590         vpap->gpa = gpa;
 591         vpap->pinned_addr = va;
 592         vpap->dirty = false;
 593         if (va)
 594                 vpap->pinned_end = va + vpap->len;
 595 }
 596
 597 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 598 {
 599         if (!(vcpu->arch.vpa.update_pending ||
 600               vcpu->arch.slb_shadow.update_pending ||
 601               vcpu->arch.dtl.update_pending))
 602                 return;
 603
 604         spin_lock(&vcpu->arch.vpa_update_lock);
 605         if (vcpu->arch.vpa.update_pending) {
 606                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 607                 if (vcpu->arch.vpa.pinned_addr)
 608                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 609         }
 610         if (vcpu->arch.dtl.update_pending) {
 611                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 612                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 613                 vcpu->arch.dtl_index = 0;
 614         }
 615         if (vcpu->arch.slb_shadow.update_pending)
 616                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 617         spin_unlock(&vcpu->arch.vpa_update_lock);
 618 }
 619
 620 /*
 621  * Return the accumulated stolen time for the vcore up until `now'.
 622  * The caller should hold the vcore lock.
 623  */
 624 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 625 {
 626         u64 p;
 627         unsigned long flags;
 628
 629         spin_lock_irqsave(&vc->stoltb_lock, flags);
 630         p = vc->stolen_tb;
 631         if (vc->vcore_state != VCORE_INACTIVE &&
 632             vc->preempt_tb != TB_NIL)
 633                 p += now - vc->preempt_tb;
 634         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 635         return p;
 636 }
 637
 638 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 639                                     struct kvmppc_vcore *vc)
 640 {
 641         struct dtl_entry *dt;
 642         struct lppaca *vpa;
 643         unsigned long stolen;
 644         unsigned long core_stolen;
 645         u64 now;
 646
 647         dt = vcpu->arch.dtl_ptr;
 648         vpa = vcpu->arch.vpa.pinned_addr;
 649         now = mftb();
 650         core_stolen = vcore_stolen_time(vc, now);
 651         stolen = core_stolen - vcpu->arch.stolen_logged;
 652         vcpu->arch.stolen_logged = core_stolen;
 653         spin_lock_irq(&vcpu->arch.tbacct_lock);
 654         stolen += vcpu->arch.busy_stolen;
 655         vcpu->arch.busy_stolen = 0;
 656         spin_unlock_irq(&vcpu->arch.tbacct_lock);
 657         if (!dt || !vpa)
 658                 return;
 659         memset(dt, 0, sizeof(struct dtl_entry));
 660         dt->dispatch_reason = 7;
 661         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 662         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 663         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 664         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 665         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 666         ++dt;
 667         if (dt == vcpu->arch.dtl.pinned_end)
 668                 dt = vcpu->arch.dtl.pinned_addr;
 669         vcpu->arch.dtl_ptr = dt;
 670         /* order writing *dt vs. writing vpa->dtl_idx */
 671         smp_wmb();
 672         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 673         vcpu->arch.dtl.dirty = true;
 674 }
 675
 676 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 677 {
 678         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 679                 return true;
 680         if ((!vcpu->arch.vcore->arch_compat) &&
 681             cpu_has_feature(CPU_FTR_ARCH_207S))
 682                 return true;
 683         return false;
 684 }
 685
 686 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 687                              unsigned long resource, unsigned long value1,
 688                              unsigned long value2)
 689 {
 690         switch (resource) {
 691         case H_SET_MODE_RESOURCE_SET_CIABR:
 692                 if (!kvmppc_power8_compatible(vcpu))
 693                         return H_P2;
 694                 if (value2)
 695                         return H_P4;
 696                 if (mflags)
 697                         return H_UNSUPPORTED_FLAG_START;
 698                 /* Guests can't breakpoint the hypervisor */
 699                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 700                         return H_P3;
 701                 vcpu->arch.ciabr  = value1;
 702                 return H_SUCCESS;
 703         case H_SET_MODE_RESOURCE_SET_DAWR:
 704                 if (!kvmppc_power8_compatible(vcpu))
 705                         return H_P2;
 706                 if (mflags)
 707                         return H_UNSUPPORTED_FLAG_START;
 708                 if (value2 & DABRX_HYP)
 709                         return H_P4;
 710                 vcpu->arch.dawr  = value1;
 711                 vcpu->arch.dawrx = value2;
 712                 return H_SUCCESS;
 713         default:
 714                 return H_TOO_HARD;
 715         }
 716 }
 717
 718 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 719 {
 720         struct kvmppc_vcore *vcore = target->arch.vcore;
 721
 722         /*
 723          * We expect to have been called by the real mode handler
 724          * (kvmppc_rm_h_confer()) which would have directly returned
 725          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 726          * have useful work to do and should not confer) so we don't
 727          * recheck that here.
 728          */
 729
 730         spin_lock(&vcore->lock);
 731         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 732             vcore->vcore_state != VCORE_INACTIVE &&
 733             vcore->runner)
 734                 target = vcore->runner;
 735         spin_unlock(&vcore->lock);
 736
 737         return kvm_vcpu_yield_to(target);
 738 }
 739
 740 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 741 {
 742         int yield_count = 0;
 743         struct lppaca *lppaca;
 744
 745         spin_lock(&vcpu->arch.vpa_update_lock);
 746         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 747         if (lppaca)
 748                 yield_count = be32_to_cpu(lppaca->yield_count);
 749         spin_unlock(&vcpu->arch.vpa_update_lock);
 750         return yield_count;
 751 }
 752
 753 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 754 {
 755         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 756         unsigned long target, ret = H_SUCCESS;
 757         int yield_count;
 758         struct kvm_vcpu *tvcpu;
 759         int idx, rc;
 760
 761         if (req <= MAX_HCALL_OPCODE &&
 762             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 763                 return RESUME_HOST;
 764
 765         switch (req) {
 766         case H_CEDE:
 767                 break;
 768         case H_PROD:
 769                 target = kvmppc_get_gpr(vcpu, 4);
 770                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 771                 if (!tvcpu) {
 772                         ret = H_PARAMETER;
 773                         break;
 774                 }
 775                 tvcpu->arch.prodded = 1;
 776                 smp_mb();
 777                 if (tvcpu->arch.ceded)
 778                         kvmppc_fast_vcpu_kick_hv(tvcpu);
 779                 break;
 780         case H_CONFER:
 781                 target = kvmppc_get_gpr(vcpu, 4);
 782                 if (target == -1)
 783                         break;
 784                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 785                 if (!tvcpu) {
 786                         ret = H_PARAMETER;
 787                         break;
 788                 }
 789                 yield_count = kvmppc_get_gpr(vcpu, 5);
 790                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 791                         break;
 792                 kvm_arch_vcpu_yield_to(tvcpu);
 793                 break;
 794         case H_REGISTER_VPA:
 795                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 796                                         kvmppc_get_gpr(vcpu, 5),
 797                                         kvmppc_get_gpr(vcpu, 6));
 798                 break;
 799         case H_RTAS:
 800                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 801                         return RESUME_HOST;
 802
 803                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 804                 rc = kvmppc_rtas_hcall(vcpu);
 805                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 806
 807                 if (rc == -ENOENT)
 808                         return RESUME_HOST;
 809                 else if (rc == 0)
 810                         break;
 811
 812                 /* Send the error out to userspace via KVM_RUN */
 813                 return rc;
 814         case H_LOGICAL_CI_LOAD:
 815                 ret = kvmppc_h_logical_ci_load(vcpu);
 816                 if (ret == H_TOO_HARD)
 817                         return RESUME_HOST;
 818                 break;
 819         case H_LOGICAL_CI_STORE:
 820                 ret = kvmppc_h_logical_ci_store(vcpu);
 821                 if (ret == H_TOO_HARD)
 822                         return RESUME_HOST;
 823                 break;
 824         case H_SET_MODE:
 825                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 826                                         kvmppc_get_gpr(vcpu, 5),
 827                                         kvmppc_get_gpr(vcpu, 6),
 828                                         kvmppc_get_gpr(vcpu, 7));
 829                 if (ret == H_TOO_HARD)
 830                         return RESUME_HOST;
 831                 break;
 832         case H_XIRR:
 833         case H_CPPR:
 834         case H_EOI:
 835         case H_IPI:
 836         case H_IPOLL:
 837         case H_XIRR_X:
 838                 if (kvmppc_xics_enabled(vcpu)) {
 839                         ret = kvmppc_xics_hcall(vcpu, req);
 840                         break;
 841                 }
 842                 return RESUME_HOST;
 843         case H_PUT_TCE:
 844                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
 845                                                 kvmppc_get_gpr(vcpu, 5),
 846                                                 kvmppc_get_gpr(vcpu, 6));
 847                 if (ret == H_TOO_HARD)
 848                         return RESUME_HOST;
 849                 break;
 850         case H_PUT_TCE_INDIRECT:
 851                 ret = kvmppc_h_put_tce_indirect(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_STUFF_TCE:
 859                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
 860                                                 kvmppc_get_gpr(vcpu, 5),
 861                                                 kvmppc_get_gpr(vcpu, 6),
 862                                                 kvmppc_get_gpr(vcpu, 7));
 863                 if (ret == H_TOO_HARD)
 864                         return RESUME_HOST;
 865                 break;
 866         default:
 867                 return RESUME_HOST;
 868         }
 869         kvmppc_set_gpr(vcpu, 3, ret);
 870         vcpu->arch.hcall_needed = 0;
 871         return RESUME_GUEST;
 872 }
 873
 874 static int kvmppc_hcall_impl_hv(unsigned long cmd)
 875 {
 876         switch (cmd) {
 877         case H_CEDE:
 878         case H_PROD:
 879         case H_CONFER:
 880         case H_REGISTER_VPA:
 881         case H_SET_MODE:
 882         case H_LOGICAL_CI_LOAD:
 883         case H_LOGICAL_CI_STORE:
 884 #ifdef CONFIG_KVM_XICS
 885         case H_XIRR:
 886         case H_CPPR:
 887         case H_EOI:
 888         case H_IPI:
 889         case H_IPOLL:
 890         case H_XIRR_X:
 891 #endif
 892                 return 1;
 893         }
 894
 895         /* See if it's in the real-mode table */
 896         return kvmppc_hcall_impl_hv_realmode(cmd);
 897 }
 898
 899 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
 900                                         struct kvm_vcpu *vcpu)
 901 {
 902         u32 last_inst;
 903
 904         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
 905                                         EMULATE_DONE) {
 906                 /*
 907                  * Fetch failed, so return to guest and
 908                  * try executing it again.
 909                  */
 910                 return RESUME_GUEST;
 911         }
 912
 913         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
 914                 run->exit_reason = KVM_EXIT_DEBUG;
 915                 run->debug.arch.address = kvmppc_get_pc(vcpu);
 916                 return RESUME_HOST;
 917         } else {
 918                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 919                 return RESUME_GUEST;
 920         }
 921 }
 922
 923 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
 924                                  struct task_struct *tsk)
 925 {
 926         int r = RESUME_HOST;
 927
 928         vcpu->stat.sum_exits++;
 929
 930         /*
 931          * This can happen if an interrupt occurs in the last stages
 932          * of guest entry or the first stages of guest exit (i.e. after
 933          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
 934          * and before setting it to KVM_GUEST_MODE_HOST_HV).
 935          * That can happen due to a bug, or due to a machine check
 936          * occurring at just the wrong time.
 937          */
 938         if (vcpu->arch.shregs.msr & MSR_HV) {
 939                 printk(KERN_EMERG "KVM trap in HV mode!\n");
 940                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 941                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 942                         vcpu->arch.shregs.msr);
 943                 kvmppc_dump_regs(vcpu);
 944                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
 945                 run->hw.hardware_exit_reason = vcpu->arch.trap;
 946                 return RESUME_HOST;
 947         }
 948         run->exit_reason = KVM_EXIT_UNKNOWN;
 949         run->ready_for_interrupt_injection = 1;
 950         switch (vcpu->arch.trap) {
 951         /* We're good on these - the host merely wanted to get our attention */
 952         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 953                 vcpu->stat.dec_exits++;
 954                 r = RESUME_GUEST;
 955                 break;
 956         case BOOK3S_INTERRUPT_EXTERNAL:
 957         case BOOK3S_INTERRUPT_H_DOORBELL:
 958         case BOOK3S_INTERRUPT_H_VIRT:
 959                 vcpu->stat.ext_intr_exits++;
 960                 r = RESUME_GUEST;
 961                 break;
 962         /* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
 963         case BOOK3S_INTERRUPT_HMI:
 964         case BOOK3S_INTERRUPT_PERFMON:
 965                 r = RESUME_GUEST;
 966                 break;
 967         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 968                 /*
 969                  * Deliver a machine check interrupt to the guest.
 970                  * We have to do this, even if the host has handled the
 971                  * machine check, because machine checks use SRR0/1 and
 972                  * the interrupt might have trashed guest state in them.
 973                  */
 974                 kvmppc_book3s_queue_irqprio(vcpu,
 975                                             BOOK3S_INTERRUPT_MACHINE_CHECK);
 976                 r = RESUME_GUEST;
 977                 break;
 978         case BOOK3S_INTERRUPT_PROGRAM:
 979         {
 980                 ulong flags;
 981                 /*
 982                  * Normally program interrupts are delivered directly
 983                  * to the guest by the hardware, but we can get here
 984                  * as a result of a hypervisor emulation interrupt
 985                  * (e40) getting turned into a 700 by BML RTAS.
 986                  */
 987                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 988                 kvmppc_core_queue_program(vcpu, flags);
 989                 r = RESUME_GUEST;
 990                 break;
 991         }
 992         case BOOK3S_INTERRUPT_SYSCALL:
 993         {
 994                 /* hcall - punt to userspace */
 995                 int i;
 996
 997                 /* hypercall with MSR_PR has already been handled in rmode,
 998                  * and never reaches here.
 999                  */
1000
1001                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1002                 for (i = 0; i < 9; ++i)
1003                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1004                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1005                 vcpu->arch.hcall_needed = 1;
1006                 r = RESUME_HOST;
1007                 break;
1008         }
1009         /*
1010          * We get these next two if the guest accesses a page which it thinks
1011          * it has mapped but which is not actually present, either because
1012          * it is for an emulated I/O device or because the corresonding
1013          * host page has been paged out.  Any other HDSI/HISI interrupts
1014          * have been handled already.
1015          */
1016         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1017                 r = RESUME_PAGE_FAULT;
1018                 break;
1019         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1020                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1021                 vcpu->arch.fault_dsisr = 0;
1022                 r = RESUME_PAGE_FAULT;
1023                 break;
1024         /*
1025          * This occurs if the guest executes an illegal instruction.
1026          * If the guest debug is disabled, generate a program interrupt
1027          * to the guest. If guest debug is enabled, we need to check
1028          * whether the instruction is a software breakpoint instruction.
1029          * Accordingly return to Guest or Host.
1030          */
1031         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1032                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1033                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1034                                 swab32(vcpu->arch.emul_inst) :
1035                                 vcpu->arch.emul_inst;
1036                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1037                         r = kvmppc_emulate_debug_inst(run, vcpu);
1038                 } else {
1039                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1040                         r = RESUME_GUEST;
1041                 }
1042                 break;
1043         /*
1044          * This occurs if the guest (kernel or userspace), does something that
1045          * is prohibited by HFSCR.  We just generate a program interrupt to
1046          * the guest.
1047          */
1048         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1049                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1050                 r = RESUME_GUEST;
1051                 break;
1052         case BOOK3S_INTERRUPT_HV_RM_HARD:
1053                 r = RESUME_PASSTHROUGH;
1054                 break;
1055         default:
1056                 kvmppc_dump_regs(vcpu);
1057                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1058                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1059                         vcpu->arch.shregs.msr);
1060                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1061                 r = RESUME_HOST;
1062                 break;
1063         }
1064
1065         return r;
1066 }
1067
1068 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1069                                             struct kvm_sregs *sregs)
1070 {
1071         int i;
1072
1073         memset(sregs, 0, sizeof(struct kvm_sregs));
1074         sregs->pvr = vcpu->arch.pvr;
1075         for (i = 0; i < vcpu->arch.slb_max; i++) {
1076                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1077                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1078         }
1079
1080         return 0;
1081 }
1082
1083 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1084                                             struct kvm_sregs *sregs)
1085 {
1086         int i, j;
1087
1088         /* Only accept the same PVR as the host's, since we can't spoof it */
1089         if (sregs->pvr != vcpu->arch.pvr)
1090                 return -EINVAL;
1091
1092         j = 0;
1093         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1094                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1095                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1096                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1097                         ++j;
1098                 }
1099         }
1100         vcpu->arch.slb_max = j;
1101
1102         return 0;
1103 }
1104
1105 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1106                 bool preserve_top32)
1107 {
1108         struct kvm *kvm = vcpu->kvm;
1109         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1110         u64 mask;
1111
1112         mutex_lock(&kvm->lock);
1113         spin_lock(&vc->lock);
1114         /*
1115          * If ILE (interrupt little-endian) has changed, update the
1116          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1117          */
1118         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1119                 struct kvm_vcpu *vcpu;
1120                 int i;
1121
1122                 kvm_for_each_vcpu(i, vcpu, kvm) {
1123                         if (vcpu->arch.vcore != vc)
1124                                 continue;
1125                         if (new_lpcr & LPCR_ILE)
1126                                 vcpu->arch.intr_msr |= MSR_LE;
1127                         else
1128                                 vcpu->arch.intr_msr &= ~MSR_LE;
1129                 }
1130         }
1131
1132         /*
1133          * Userspace can only modify DPFD (default prefetch depth),
1134          * ILE (interrupt little-endian) and TC (translation control).
1135          * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1136          */
1137         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1138         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1139                 mask |= LPCR_AIL;
1140
1141         /* Broken 32-bit version of LPCR must not clear top bits */
1142         if (preserve_top32)
1143                 mask &= 0xFFFFFFFF;
1144         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1145         spin_unlock(&vc->lock);
1146         mutex_unlock(&kvm->lock);
1147 }
1148
1149 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1150                                  union kvmppc_one_reg *val)
1151 {
1152         int r = 0;
1153         long int i;
1154
1155         switch (id) {
1156         case KVM_REG_PPC_DEBUG_INST:
1157                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1158                 break;
1159         case KVM_REG_PPC_HIOR:
1160                 *val = get_reg_val(id, 0);
1161                 break;
1162         case KVM_REG_PPC_DABR:
1163                 *val = get_reg_val(id, vcpu->arch.dabr);
1164                 break;
1165         case KVM_REG_PPC_DABRX:
1166                 *val = get_reg_val(id, vcpu->arch.dabrx);
1167                 break;
1168         case KVM_REG_PPC_DSCR:
1169                 *val = get_reg_val(id, vcpu->arch.dscr);
1170                 break;
1171         case KVM_REG_PPC_PURR:
1172                 *val = get_reg_val(id, vcpu->arch.purr);
1173                 break;
1174         case KVM_REG_PPC_SPURR:
1175                 *val = get_reg_val(id, vcpu->arch.spurr);
1176                 break;
1177         case KVM_REG_PPC_AMR:
1178                 *val = get_reg_val(id, vcpu->arch.amr);
1179                 break;
1180         case KVM_REG_PPC_UAMOR:
1181                 *val = get_reg_val(id, vcpu->arch.uamor);
1182                 break;
1183         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1184                 i = id - KVM_REG_PPC_MMCR0;
1185                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1186                 break;
1187         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1188                 i = id - KVM_REG_PPC_PMC1;
1189                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1190                 break;
1191         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1192                 i = id - KVM_REG_PPC_SPMC1;
1193                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1194                 break;
1195         case KVM_REG_PPC_SIAR:
1196                 *val = get_reg_val(id, vcpu->arch.siar);
1197                 break;
1198         case KVM_REG_PPC_SDAR:
1199                 *val = get_reg_val(id, vcpu->arch.sdar);
1200                 break;
1201         case KVM_REG_PPC_SIER:
1202                 *val = get_reg_val(id, vcpu->arch.sier);
1203                 break;
1204         case KVM_REG_PPC_IAMR:
1205                 *val = get_reg_val(id, vcpu->arch.iamr);
1206                 break;
1207         case KVM_REG_PPC_PSPB:
1208                 *val = get_reg_val(id, vcpu->arch.pspb);
1209                 break;
1210         case KVM_REG_PPC_DPDES:
1211                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1212                 break;
1213         case KVM_REG_PPC_VTB:
1214                 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1215                 break;
1216         case KVM_REG_PPC_DAWR:
1217                 *val = get_reg_val(id, vcpu->arch.dawr);
1218                 break;
1219         case KVM_REG_PPC_DAWRX:
1220                 *val = get_reg_val(id, vcpu->arch.dawrx);
1221                 break;
1222         case KVM_REG_PPC_CIABR:
1223                 *val = get_reg_val(id, vcpu->arch.ciabr);
1224                 break;
1225         case KVM_REG_PPC_CSIGR:
1226                 *val = get_reg_val(id, vcpu->arch.csigr);
1227                 break;
1228         case KVM_REG_PPC_TACR:
1229                 *val = get_reg_val(id, vcpu->arch.tacr);
1230                 break;
1231         case KVM_REG_PPC_TCSCR:
1232                 *val = get_reg_val(id, vcpu->arch.tcscr);
1233                 break;
1234         case KVM_REG_PPC_PID:
1235                 *val = get_reg_val(id, vcpu->arch.pid);
1236                 break;
1237         case KVM_REG_PPC_ACOP:
1238                 *val = get_reg_val(id, vcpu->arch.acop);
1239                 break;
1240         case KVM_REG_PPC_WORT:
1241                 *val = get_reg_val(id, vcpu->arch.wort);
1242                 break;
1243         case KVM_REG_PPC_TIDR:
1244                 *val = get_reg_val(id, vcpu->arch.tid);
1245                 break;
1246         case KVM_REG_PPC_PSSCR:
1247                 *val = get_reg_val(id, vcpu->arch.psscr);
1248                 break;
1249         case KVM_REG_PPC_VPA_ADDR:
1250                 spin_lock(&vcpu->arch.vpa_update_lock);
1251                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1252                 spin_unlock(&vcpu->arch.vpa_update_lock);
1253                 break;
1254         case KVM_REG_PPC_VPA_SLB:
1255                 spin_lock(&vcpu->arch.vpa_update_lock);
1256                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1257                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1258                 spin_unlock(&vcpu->arch.vpa_update_lock);
1259                 break;
1260         case KVM_REG_PPC_VPA_DTL:
1261                 spin_lock(&vcpu->arch.vpa_update_lock);
1262                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1263                 val->vpaval.length = vcpu->arch.dtl.len;
1264                 spin_unlock(&vcpu->arch.vpa_update_lock);
1265                 break;
1266         case KVM_REG_PPC_TB_OFFSET:
1267                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1268                 break;
1269         case KVM_REG_PPC_LPCR:
1270         case KVM_REG_PPC_LPCR_64:
1271                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1272                 break;
1273         case KVM_REG_PPC_PPR:
1274                 *val = get_reg_val(id, vcpu->arch.ppr);
1275                 break;
1276 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1277         case KVM_REG_PPC_TFHAR:
1278                 *val = get_reg_val(id, vcpu->arch.tfhar);
1279                 break;
1280         case KVM_REG_PPC_TFIAR:
1281                 *val = get_reg_val(id, vcpu->arch.tfiar);
1282                 break;
1283         case KVM_REG_PPC_TEXASR:
1284                 *val = get_reg_val(id, vcpu->arch.texasr);
1285                 break;
1286         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1287                 i = id - KVM_REG_PPC_TM_GPR0;
1288                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1289                 break;
1290         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1291         {
1292                 int j;
1293                 i = id - KVM_REG_PPC_TM_VSR0;
1294                 if (i < 32)
1295                         for (j = 0; j < TS_FPRWIDTH; j++)
1296                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1297                 else {
1298                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1299                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1300                         else
1301                                 r = -ENXIO;
1302                 }
1303                 break;
1304         }
1305         case KVM_REG_PPC_TM_CR:
1306                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1307                 break;
1308         case KVM_REG_PPC_TM_XER:
1309                 *val = get_reg_val(id, vcpu->arch.xer_tm);
1310                 break;
1311         case KVM_REG_PPC_TM_LR:
1312                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1313                 break;
1314         case KVM_REG_PPC_TM_CTR:
1315                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1316                 break;
1317         case KVM_REG_PPC_TM_FPSCR:
1318                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1319                 break;
1320         case KVM_REG_PPC_TM_AMR:
1321                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1322                 break;
1323         case KVM_REG_PPC_TM_PPR:
1324                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1325                 break;
1326         case KVM_REG_PPC_TM_VRSAVE:
1327                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1328                 break;
1329         case KVM_REG_PPC_TM_VSCR:
1330                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1331                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1332                 else
1333                         r = -ENXIO;
1334                 break;
1335         case KVM_REG_PPC_TM_DSCR:
1336                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1337                 break;
1338         case KVM_REG_PPC_TM_TAR:
1339                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1340                 break;
1341 #endif
1342         case KVM_REG_PPC_ARCH_COMPAT:
1343                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1344                 break;
1345         default:
1346                 r = -EINVAL;
1347                 break;
1348         }
1349
1350         return r;
1351 }
1352
1353 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1354                                  union kvmppc_one_reg *val)
1355 {
1356         int r = 0;
1357         long int i;
1358         unsigned long addr, len;
1359
1360         switch (id) {
1361         case KVM_REG_PPC_HIOR:
1362                 /* Only allow this to be set to zero */
1363                 if (set_reg_val(id, *val))
1364                         r = -EINVAL;
1365                 break;
1366         case KVM_REG_PPC_DABR:
1367                 vcpu->arch.dabr = set_reg_val(id, *val);
1368                 break;
1369         case KVM_REG_PPC_DABRX:
1370                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1371                 break;
1372         case KVM_REG_PPC_DSCR:
1373                 vcpu->arch.dscr = set_reg_val(id, *val);
1374                 break;
1375         case KVM_REG_PPC_PURR:
1376                 vcpu->arch.purr = set_reg_val(id, *val);
1377                 break;
1378         case KVM_REG_PPC_SPURR:
1379                 vcpu->arch.spurr = set_reg_val(id, *val);
1380                 break;
1381         case KVM_REG_PPC_AMR:
1382                 vcpu->arch.amr = set_reg_val(id, *val);
1383                 break;
1384         case KVM_REG_PPC_UAMOR:
1385                 vcpu->arch.uamor = set_reg_val(id, *val);
1386                 break;
1387         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1388                 i = id - KVM_REG_PPC_MMCR0;
1389                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1390                 break;
1391         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1392                 i = id - KVM_REG_PPC_PMC1;
1393                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1394                 break;
1395         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1396                 i = id - KVM_REG_PPC_SPMC1;
1397                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1398                 break;
1399         case KVM_REG_PPC_SIAR:
1400                 vcpu->arch.siar = set_reg_val(id, *val);
1401                 break;
1402         case KVM_REG_PPC_SDAR:
1403                 vcpu->arch.sdar = set_reg_val(id, *val);
1404                 break;
1405         case KVM_REG_PPC_SIER:
1406                 vcpu->arch.sier = set_reg_val(id, *val);
1407                 break;
1408         case KVM_REG_PPC_IAMR:
1409                 vcpu->arch.iamr = set_reg_val(id, *val);
1410                 break;
1411         case KVM_REG_PPC_PSPB:
1412                 vcpu->arch.pspb = set_reg_val(id, *val);
1413                 break;
1414         case KVM_REG_PPC_DPDES:
1415                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1416                 break;
1417         case KVM_REG_PPC_VTB:
1418                 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1419                 break;
1420         case KVM_REG_PPC_DAWR:
1421                 vcpu->arch.dawr = set_reg_val(id, *val);
1422                 break;
1423         case KVM_REG_PPC_DAWRX:
1424                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1425                 break;
1426         case KVM_REG_PPC_CIABR:
1427                 vcpu->arch.ciabr = set_reg_val(id, *val);
1428                 /* Don't allow setting breakpoints in hypervisor code */
1429                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1430                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1431                 break;
1432         case KVM_REG_PPC_CSIGR:
1433                 vcpu->arch.csigr = set_reg_val(id, *val);
1434                 break;
1435         case KVM_REG_PPC_TACR:
1436                 vcpu->arch.tacr = set_reg_val(id, *val);
1437                 break;
1438         case KVM_REG_PPC_TCSCR:
1439                 vcpu->arch.tcscr = set_reg_val(id, *val);
1440                 break;
1441         case KVM_REG_PPC_PID:
1442                 vcpu->arch.pid = set_reg_val(id, *val);
1443                 break;
1444         case KVM_REG_PPC_ACOP:
1445                 vcpu->arch.acop = set_reg_val(id, *val);
1446                 break;
1447         case KVM_REG_PPC_WORT:
1448                 vcpu->arch.wort = set_reg_val(id, *val);
1449                 break;
1450         case KVM_REG_PPC_TIDR:
1451                 vcpu->arch.tid = set_reg_val(id, *val);
1452                 break;
1453         case KVM_REG_PPC_PSSCR:
1454                 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1455                 break;
1456         case KVM_REG_PPC_VPA_ADDR:
1457                 addr = set_reg_val(id, *val);
1458                 r = -EINVAL;
1459                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1460                               vcpu->arch.dtl.next_gpa))
1461                         break;
1462                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1463                 break;
1464         case KVM_REG_PPC_VPA_SLB:
1465                 addr = val->vpaval.addr;
1466                 len = val->vpaval.length;
1467                 r = -EINVAL;
1468                 if (addr && !vcpu->arch.vpa.next_gpa)
1469                         break;
1470                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1471                 break;
1472         case KVM_REG_PPC_VPA_DTL:
1473                 addr = val->vpaval.addr;
1474                 len = val->vpaval.length;
1475                 r = -EINVAL;
1476                 if (addr && (len < sizeof(struct dtl_entry) ||
1477                              !vcpu->arch.vpa.next_gpa))
1478                         break;
1479                 len -= len % sizeof(struct dtl_entry);
1480                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1481                 break;
1482         case KVM_REG_PPC_TB_OFFSET:
1483                 /* round up to multiple of 2^24 */
1484                 vcpu->arch.vcore->tb_offset =
1485                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1486                 break;
1487         case KVM_REG_PPC_LPCR:
1488                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1489                 break;
1490         case KVM_REG_PPC_LPCR_64:
1491                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1492                 break;
1493         case KVM_REG_PPC_PPR:
1494                 vcpu->arch.ppr = set_reg_val(id, *val);
1495                 break;
1496 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1497         case KVM_REG_PPC_TFHAR:
1498                 vcpu->arch.tfhar = set_reg_val(id, *val);
1499                 break;
1500         case KVM_REG_PPC_TFIAR:
1501                 vcpu->arch.tfiar = set_reg_val(id, *val);
1502                 break;
1503         case KVM_REG_PPC_TEXASR:
1504                 vcpu->arch.texasr = set_reg_val(id, *val);
1505                 break;
1506         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1507                 i = id - KVM_REG_PPC_TM_GPR0;
1508                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1509                 break;
1510         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1511         {
1512                 int j;
1513                 i = id - KVM_REG_PPC_TM_VSR0;
1514                 if (i < 32)
1515                         for (j = 0; j < TS_FPRWIDTH; j++)
1516                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1517                 else
1518                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1519                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1520                         else
1521                                 r = -ENXIO;
1522                 break;
1523         }
1524         case KVM_REG_PPC_TM_CR:
1525                 vcpu->arch.cr_tm = set_reg_val(id, *val);
1526                 break;
1527         case KVM_REG_PPC_TM_XER:
1528                 vcpu->arch.xer_tm = set_reg_val(id, *val);
1529                 break;
1530         case KVM_REG_PPC_TM_LR:
1531                 vcpu->arch.lr_tm = set_reg_val(id, *val);
1532                 break;
1533         case KVM_REG_PPC_TM_CTR:
1534                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1535                 break;
1536         case KVM_REG_PPC_TM_FPSCR:
1537                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1538                 break;
1539         case KVM_REG_PPC_TM_AMR:
1540                 vcpu->arch.amr_tm = set_reg_val(id, *val);
1541                 break;
1542         case KVM_REG_PPC_TM_PPR:
1543                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1544                 break;
1545         case KVM_REG_PPC_TM_VRSAVE:
1546                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1547                 break;
1548         case KVM_REG_PPC_TM_VSCR:
1549                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1550                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1551                 else
1552                         r = - ENXIO;
1553                 break;
1554         case KVM_REG_PPC_TM_DSCR:
1555                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1556                 break;
1557         case KVM_REG_PPC_TM_TAR:
1558                 vcpu->arch.tar_tm = set_reg_val(id, *val);
1559                 break;
1560 #endif
1561         case KVM_REG_PPC_ARCH_COMPAT:
1562                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1563                 break;
1564         default:
1565                 r = -EINVAL;
1566                 break;
1567         }
1568
1569         return r;
1570 }
1571
1572 /*
1573  * On POWER9, threads are independent and can be in different partitions.
1574  * Therefore we consider each thread to be a subcore.
1575  * There is a restriction that all threads have to be in the same
1576  * MMU mode (radix or HPT), unfortunately, but since we only support
1577  * HPT guests on a HPT host so far, that isn't an impediment yet.
1578  */
1579 static int threads_per_vcore(void)
1580 {
1581         if (cpu_has_feature(CPU_FTR_ARCH_300))
1582                 return 1;
1583         return threads_per_subcore;
1584 }
1585
1586 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1587 {
1588         struct kvmppc_vcore *vcore;
1589
1590         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1591
1592         if (vcore == NULL)
1593                 return NULL;
1594
1595         spin_lock_init(&vcore->lock);
1596         spin_lock_init(&vcore->stoltb_lock);
1597         init_swait_queue_head(&vcore->wq);
1598         vcore->preempt_tb = TB_NIL;
1599         vcore->lpcr = kvm->arch.lpcr;
1600         vcore->first_vcpuid = core * threads_per_vcore();
1601         vcore->kvm = kvm;
1602         INIT_LIST_HEAD(&vcore->preempt_list);
1603
1604         return vcore;
1605 }
1606
1607 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1608 static struct debugfs_timings_element {
1609         const char *name;
1610         size_t offset;
1611 } timings[] = {
1612         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
1613         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
1614         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
1615         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
1616         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
1617 };
1618
1619 #define N_TIMINGS       (sizeof(timings) / sizeof(timings[0]))
1620
1621 struct debugfs_timings_state {
1622         struct kvm_vcpu *vcpu;
1623         unsigned int    buflen;
1624         char            buf[N_TIMINGS * 100];
1625 };
1626
1627 static int debugfs_timings_open(struct inode *inode, struct file *file)
1628 {
1629         struct kvm_vcpu *vcpu = inode->i_private;
1630         struct debugfs_timings_state *p;
1631
1632         p = kzalloc(sizeof(*p), GFP_KERNEL);
1633         if (!p)
1634                 return -ENOMEM;
1635
1636         kvm_get_kvm(vcpu->kvm);
1637         p->vcpu = vcpu;
1638         file->private_data = p;
1639
1640         return nonseekable_open(inode, file);
1641 }
1642
1643 static int debugfs_timings_release(struct inode *inode, struct file *file)
1644 {
1645         struct debugfs_timings_state *p = file->private_data;
1646
1647         kvm_put_kvm(p->vcpu->kvm);
1648         kfree(p);
1649         return 0;
1650 }
1651
1652 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1653                                     size_t len, loff_t *ppos)
1654 {
1655         struct debugfs_timings_state *p = file->private_data;
1656         struct kvm_vcpu *vcpu = p->vcpu;
1657         char *s, *buf_end;
1658         struct kvmhv_tb_accumulator tb;
1659         u64 count;
1660         loff_t pos;
1661         ssize_t n;
1662         int i, loops;
1663         bool ok;
1664
1665         if (!p->buflen) {
1666                 s = p->buf;
1667                 buf_end = s + sizeof(p->buf);
1668                 for (i = 0; i < N_TIMINGS; ++i) {
1669                         struct kvmhv_tb_accumulator *acc;
1670
1671                         acc = (struct kvmhv_tb_accumulator *)
1672                                 ((unsigned long)vcpu + timings[i].offset);
1673                         ok = false;
1674                         for (loops = 0; loops < 1000; ++loops) {
1675                                 count = acc->seqcount;
1676                                 if (!(count & 1)) {
1677                                         smp_rmb();
1678                                         tb = *acc;
1679                                         smp_rmb();
1680                                         if (count == acc->seqcount) {
1681                                                 ok = true;
1682                                                 break;
1683                                         }
1684                                 }
1685                                 udelay(1);
1686                         }
1687                         if (!ok)
1688                                 snprintf(s, buf_end - s, "%s: stuck\n",
1689                                         timings[i].name);
1690                         else
1691                                 snprintf(s, buf_end - s,
1692                                         "%s: %llu %llu %llu %llu\n",
1693                                         timings[i].name, count / 2,
1694                                         tb_to_ns(tb.tb_total),
1695                                         tb_to_ns(tb.tb_min),
1696                                         tb_to_ns(tb.tb_max));
1697                         s += strlen(s);
1698                 }
1699                 p->buflen = s - p->buf;
1700         }
1701
1702         pos = *ppos;
1703         if (pos >= p->buflen)
1704                 return 0;
1705         if (len > p->buflen - pos)
1706                 len = p->buflen - pos;
1707         n = copy_to_user(buf, p->buf + pos, len);
1708         if (n) {
1709                 if (n == len)
1710                         return -EFAULT;
1711                 len -= n;
1712         }
1713         *ppos = pos + len;
1714         return len;
1715 }
1716
1717 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1718                                      size_t len, loff_t *ppos)
1719 {
1720         return -EACCES;
1721 }
1722
1723 static const struct file_operations debugfs_timings_ops = {
1724         .owner   = THIS_MODULE,
1725         .open    = debugfs_timings_open,
1726         .release = debugfs_timings_release,
1727         .read    = debugfs_timings_read,
1728         .write   = debugfs_timings_write,
1729         .llseek  = generic_file_llseek,
1730 };
1731
1732 /* Create a debugfs directory for the vcpu */
1733 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1734 {
1735         char buf[16];
1736         struct kvm *kvm = vcpu->kvm;
1737
1738         snprintf(buf, sizeof(buf), "vcpu%u", id);
1739         if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1740                 return;
1741         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1742         if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1743                 return;
1744         vcpu->arch.debugfs_timings =
1745                 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1746                                     vcpu, &debugfs_timings_ops);
1747 }
1748
1749 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1750 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1751 {
1752 }
1753 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1754
1755 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1756                                                    unsigned int id)
1757 {
1758         struct kvm_vcpu *vcpu;
1759         int err = -EINVAL;
1760         int core;
1761         struct kvmppc_vcore *vcore;
1762
1763         core = id / threads_per_vcore();
1764         if (core >= KVM_MAX_VCORES)
1765                 goto out;
1766
1767         err = -ENOMEM;
1768         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1769         if (!vcpu)
1770                 goto out;
1771
1772         err = kvm_vcpu_init(vcpu, kvm, id);
1773         if (err)
1774                 goto free_vcpu;
1775
1776         vcpu->arch.shared = &vcpu->arch.shregs;
1777 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1778         /*
1779          * The shared struct is never shared on HV,
1780          * so we can always use host endianness
1781          */
1782 #ifdef __BIG_ENDIAN__
1783         vcpu->arch.shared_big_endian = true;
1784 #else
1785         vcpu->arch.shared_big_endian = false;
1786 #endif
1787 #endif
1788         vcpu->arch.mmcr[0] = MMCR0_FC;
1789         vcpu->arch.ctrl = CTRL_RUNLATCH;
1790         /* default to host PVR, since we can't spoof it */
1791         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1792         spin_lock_init(&vcpu->arch.vpa_update_lock);
1793         spin_lock_init(&vcpu->arch.tbacct_lock);
1794         vcpu->arch.busy_preempt = TB_NIL;
1795         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1796
1797         kvmppc_mmu_book3s_hv_init(vcpu);
1798
1799         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1800
1801         init_waitqueue_head(&vcpu->arch.cpu_run);
1802
1803         mutex_lock(&kvm->lock);
1804         vcore = kvm->arch.vcores[core];
1805         if (!vcore) {
1806                 vcore = kvmppc_vcore_create(kvm, core);
1807                 kvm->arch.vcores[core] = vcore;
1808                 kvm->arch.online_vcores++;
1809         }
1810         mutex_unlock(&kvm->lock);
1811
1812         if (!vcore)
1813                 goto free_vcpu;
1814
1815         spin_lock(&vcore->lock);
1816         ++vcore->num_threads;
1817         spin_unlock(&vcore->lock);
1818         vcpu->arch.vcore = vcore;
1819         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1820         vcpu->arch.thread_cpu = -1;
1821         vcpu->arch.prev_cpu = -1;
1822
1823         vcpu->arch.cpu_type = KVM_CPU_3S_64;
1824         kvmppc_sanity_check(vcpu);
1825
1826         debugfs_vcpu_init(vcpu, id);
1827
1828         return vcpu;
1829
1830 free_vcpu:
1831         kmem_cache_free(kvm_vcpu_cache, vcpu);
1832 out:
1833         return ERR_PTR(err);
1834 }
1835
1836 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1837 {
1838         if (vpa->pinned_addr)
1839                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1840                                         vpa->dirty);
1841 }
1842
1843 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1844 {
1845         spin_lock(&vcpu->arch.vpa_update_lock);
1846         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1847         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1848         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1849         spin_unlock(&vcpu->arch.vpa_update_lock);
1850         kvm_vcpu_uninit(vcpu);
1851         kmem_cache_free(kvm_vcpu_cache, vcpu);
1852 }
1853
1854 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1855 {
1856         /* Indicate we want to get back into the guest */
1857         return 1;
1858 }
1859
1860 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1861 {
1862         unsigned long dec_nsec, now;
1863
1864         now = get_tb();
1865         if (now > vcpu->arch.dec_expires) {
1866                 /* decrementer has already gone negative */
1867                 kvmppc_core_queue_dec(vcpu);
1868                 kvmppc_core_prepare_to_enter(vcpu);
1869                 return;
1870         }
1871         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1872                    / tb_ticks_per_sec;
1873         hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
1874         vcpu->arch.timer_running = 1;
1875 }
1876
1877 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1878 {
1879         vcpu->arch.ceded = 0;
1880         if (vcpu->arch.timer_running) {
1881                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1882                 vcpu->arch.timer_running = 0;
1883         }
1884 }
1885
1886 extern void __kvmppc_vcore_entry(void);
1887
1888 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1889                                    struct kvm_vcpu *vcpu)
1890 {
1891         u64 now;
1892
1893         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1894                 return;
1895         spin_lock_irq(&vcpu->arch.tbacct_lock);
1896         now = mftb();
1897         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1898                 vcpu->arch.stolen_logged;
1899         vcpu->arch.busy_preempt = now;
1900         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1901         spin_unlock_irq(&vcpu->arch.tbacct_lock);
1902         --vc->n_runnable;
1903         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
1904 }
1905
1906 static int kvmppc_grab_hwthread(int cpu)
1907 {
1908         struct paca_struct *tpaca;
1909         long timeout = 10000;
1910
1911         tpaca = &paca[cpu];
1912
1913         /* Ensure the thread won't go into the kernel if it wakes */
1914         tpaca->kvm_hstate.kvm_vcpu = NULL;
1915         tpaca->kvm_hstate.kvm_vcore = NULL;
1916         tpaca->kvm_hstate.napping = 0;
1917         smp_wmb();
1918         tpaca->kvm_hstate.hwthread_req = 1;
1919
1920         /*
1921          * If the thread is already executing in the kernel (e.g. handling
1922          * a stray interrupt), wait for it to get back to nap mode.
1923          * The smp_mb() is to ensure that our setting of hwthread_req
1924          * is visible before we look at hwthread_state, so if this
1925          * races with the code at system_reset_pSeries and the thread
1926          * misses our setting of hwthread_req, we are sure to see its
1927          * setting of hwthread_state, and vice versa.
1928          */
1929         smp_mb();
1930         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1931                 if (--timeout <= 0) {
1932                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
1933                         return -EBUSY;
1934                 }
1935                 udelay(1);
1936         }
1937         return 0;
1938 }
1939
1940 static void kvmppc_release_hwthread(int cpu)
1941 {
1942         struct paca_struct *tpaca;
1943
1944         tpaca = &paca[cpu];
1945         tpaca->kvm_hstate.hwthread_req = 0;
1946         tpaca->kvm_hstate.kvm_vcpu = NULL;
1947         tpaca->kvm_hstate.kvm_vcore = NULL;
1948         tpaca->kvm_hstate.kvm_split_mode = NULL;
1949 }
1950
1951 static void do_nothing(void *x)
1952 {
1953 }
1954
1955 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
1956 {
1957         int i;
1958
1959         cpu = cpu_first_thread_sibling(cpu);
1960         cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
1961         /*
1962          * Make sure setting of bit in need_tlb_flush precedes
1963          * testing of cpu_in_guest bits.  The matching barrier on
1964          * the other side is the first smp_mb() in kvmppc_run_core().
1965          */
1966         smp_mb();
1967         for (i = 0; i < threads_per_core; ++i)
1968                 if (cpumask_test_cpu(cpu + i, &kvm->arch.cpu_in_guest))
1969                         smp_call_function_single(cpu + i, do_nothing, NULL, 1);
1970 }
1971
1972 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
1973 {
1974         int cpu;
1975         struct paca_struct *tpaca;
1976         struct kvmppc_vcore *mvc = vc->master_vcore;
1977         struct kvm *kvm = vc->kvm;
1978
1979         cpu = vc->pcpu;
1980         if (vcpu) {
1981                 if (vcpu->arch.timer_running) {
1982                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1983                         vcpu->arch.timer_running = 0;
1984                 }
1985                 cpu += vcpu->arch.ptid;
1986                 vcpu->cpu = mvc->pcpu;
1987                 vcpu->arch.thread_cpu = cpu;
1988
1989                 /*
1990                  * With radix, the guest can do TLB invalidations itself,
1991                  * and it could choose to use the local form (tlbiel) if
1992                  * it is invalidating a translation that has only ever been
1993                  * used on one vcpu.  However, that doesn't mean it has
1994                  * only ever been used on one physical cpu, since vcpus
1995                  * can move around between pcpus.  To cope with this, when
1996                  * a vcpu moves from one pcpu to another, we need to tell
1997                  * any vcpus running on the same core as this vcpu previously
1998                  * ran to flush the TLB.  The TLB is shared between threads,
1999                  * so we use a single bit in .need_tlb_flush for all 4 threads.
2000                  */
2001                 if (kvm_is_radix(kvm) && vcpu->arch.prev_cpu != cpu) {
2002                         if (vcpu->arch.prev_cpu >= 0 &&
2003                             cpu_first_thread_sibling(vcpu->arch.prev_cpu) !=
2004                             cpu_first_thread_sibling(cpu))
2005                                 radix_flush_cpu(kvm, vcpu->arch.prev_cpu, vcpu);
2006                         vcpu->arch.prev_cpu = cpu;
2007                 }
2008                 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2009         }
2010         tpaca = &paca[cpu];
2011         tpaca->kvm_hstate.kvm_vcpu = vcpu;
2012         tpaca->kvm_hstate.ptid = cpu - mvc->pcpu;
2013         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2014         smp_wmb();
2015         tpaca->kvm_hstate.kvm_vcore = mvc;
2016         if (cpu != smp_processor_id())
2017                 kvmppc_ipi_thread(cpu);
2018 }
2019
2020 static void kvmppc_wait_for_nap(void)
2021 {
2022         int cpu = smp_processor_id();
2023         int i, loops;
2024         int n_threads = threads_per_vcore();
2025
2026         if (n_threads <= 1)
2027                 return;
2028         for (loops = 0; loops < 1000000; ++loops) {
2029                 /*
2030                  * Check if all threads are finished.
2031                  * We set the vcore pointer when starting a thread
2032                  * and the thread clears it when finished, so we look
2033                  * for any threads that still have a non-NULL vcore ptr.
2034                  */
2035                 for (i = 1; i < n_threads; ++i)
2036                         if (paca[cpu + i].kvm_hstate.kvm_vcore)
2037                                 break;
2038                 if (i == n_threads) {
2039                         HMT_medium();
2040                         return;
2041                 }
2042                 HMT_low();
2043         }
2044         HMT_medium();
2045         for (i = 1; i < n_threads; ++i)
2046                 if (paca[cpu + i].kvm_hstate.kvm_vcore)
2047                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2048 }
2049
2050 /*
2051  * Check that we are on thread 0 and that any other threads in
2052  * this core are off-line.  Then grab the threads so they can't
2053  * enter the kernel.
2054  */
2055 static int on_primary_thread(void)
2056 {
2057         int cpu = smp_processor_id();
2058         int thr;
2059
2060         /* Are we on a primary subcore? */
2061         if (cpu_thread_in_subcore(cpu))
2062                 return 0;
2063
2064         thr = 0;
2065         while (++thr < threads_per_subcore)
2066                 if (cpu_online(cpu + thr))
2067                         return 0;
2068
2069         /* Grab all hw threads so they can't go into the kernel */
2070         for (thr = 1; thr < threads_per_subcore; ++thr) {
2071                 if (kvmppc_grab_hwthread(cpu + thr)) {
2072                         /* Couldn't grab one; let the others go */
2073                         do {
2074                                 kvmppc_release_hwthread(cpu + thr);
2075                         } while (--thr > 0);
2076                         return 0;
2077                 }
2078         }
2079         return 1;
2080 }
2081
2082 /*
2083  * A list of virtual cores for each physical CPU.
2084  * These are vcores that could run but their runner VCPU tasks are
2085  * (or may be) preempted.
2086  */
2087 struct preempted_vcore_list {
2088         struct list_head        list;
2089         spinlock_t              lock;
2090 };
2091
2092 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2093
2094 static void init_vcore_lists(void)
2095 {
2096         int cpu;
2097
2098         for_each_possible_cpu(cpu) {
2099                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2100                 spin_lock_init(&lp->lock);
2101                 INIT_LIST_HEAD(&lp->list);
2102         }
2103 }
2104
2105 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2106 {
2107         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2108
2109         vc->vcore_state = VCORE_PREEMPT;
2110         vc->pcpu = smp_processor_id();
2111         if (vc->num_threads < threads_per_vcore()) {
2112                 spin_lock(&lp->lock);
2113                 list_add_tail(&vc->preempt_list, &lp->list);
2114                 spin_unlock(&lp->lock);
2115         }
2116
2117         /* Start accumulating stolen time */
2118         kvmppc_core_start_stolen(vc);
2119 }
2120
2121 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2122 {
2123         struct preempted_vcore_list *lp;
2124
2125         kvmppc_core_end_stolen(vc);
2126         if (!list_empty(&vc->preempt_list)) {
2127                 lp = &per_cpu(preempted_vcores, vc->pcpu);
2128                 spin_lock(&lp->lock);
2129                 list_del_init(&vc->preempt_list);
2130                 spin_unlock(&lp->lock);
2131         }
2132         vc->vcore_state = VCORE_INACTIVE;
2133 }
2134
2135 /*
2136  * This stores information about the virtual cores currently
2137  * assigned to a physical core.
2138  */
2139 struct core_info {
2140         int             n_subcores;
2141         int             max_subcore_threads;
2142         int             total_threads;
2143         int             subcore_threads[MAX_SUBCORES];
2144         struct kvm      *subcore_vm[MAX_SUBCORES];
2145         struct list_head vcs[MAX_SUBCORES];
2146 };
2147
2148 /*
2149  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2150  * respectively in 2-way micro-threading (split-core) mode.
2151  */
2152 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2153
2154 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2155 {
2156         int sub;
2157
2158         memset(cip, 0, sizeof(*cip));
2159         cip->n_subcores = 1;
2160         cip->max_subcore_threads = vc->num_threads;
2161         cip->total_threads = vc->num_threads;
2162         cip->subcore_threads[0] = vc->num_threads;
2163         cip->subcore_vm[0] = vc->kvm;
2164         for (sub = 0; sub < MAX_SUBCORES; ++sub)
2165                 INIT_LIST_HEAD(&cip->vcs[sub]);
2166         list_add_tail(&vc->preempt_list, &cip->vcs[0]);
2167 }
2168
2169 static bool subcore_config_ok(int n_subcores, int n_threads)
2170 {
2171         /* Can only dynamically split if unsplit to begin with */
2172         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2173                 return false;
2174         if (n_subcores > MAX_SUBCORES)
2175                 return false;
2176         if (n_subcores > 1) {
2177                 if (!(dynamic_mt_modes & 2))
2178                         n_subcores = 4;
2179                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2180                         return false;
2181         }
2182
2183         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2184 }
2185
2186 static void init_master_vcore(struct kvmppc_vcore *vc)
2187 {
2188         vc->master_vcore = vc;
2189         vc->entry_exit_map = 0;
2190         vc->in_guest = 0;
2191         vc->napping_threads = 0;
2192         vc->conferring_threads = 0;
2193 }
2194
2195 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2196 {
2197         int n_threads = vc->num_threads;
2198         int sub;
2199
2200         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2201                 return false;
2202
2203         if (n_threads < cip->max_subcore_threads)
2204                 n_threads = cip->max_subcore_threads;
2205         if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2206                 return false;
2207         cip->max_subcore_threads = n_threads;
2208
2209         sub = cip->n_subcores;
2210         ++cip->n_subcores;
2211         cip->total_threads += vc->num_threads;
2212         cip->subcore_threads[sub] = vc->num_threads;
2213         cip->subcore_vm[sub] = vc->kvm;
2214         init_master_vcore(vc);
2215         list_move_tail(&vc->preempt_list, &cip->vcs[sub]);
2216
2217         return true;
2218 }
2219
2220 /*
2221  * Work out whether it is possible to piggyback the execution of
2222  * vcore *pvc onto the execution of the other vcores described in *cip.
2223  */
2224 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2225                           int target_threads)
2226 {
2227         if (cip->total_threads + pvc->num_threads > target_threads)
2228                 return false;
2229
2230         return can_dynamic_split(pvc, cip);
2231 }
2232
2233 static void prepare_threads(struct kvmppc_vcore *vc)
2234 {
2235         int i;
2236         struct kvm_vcpu *vcpu;
2237
2238         for_each_runnable_thread(i, vcpu, vc) {
2239                 if (signal_pending(vcpu->arch.run_task))
2240                         vcpu->arch.ret = -EINTR;
2241                 else if (vcpu->arch.vpa.update_pending ||
2242                          vcpu->arch.slb_shadow.update_pending ||
2243                          vcpu->arch.dtl.update_pending)
2244                         vcpu->arch.ret = RESUME_GUEST;
2245                 else
2246                         continue;
2247                 kvmppc_remove_runnable(vc, vcpu);
2248                 wake_up(&vcpu->arch.cpu_run);
2249         }
2250 }
2251
2252 static void collect_piggybacks(struct core_info *cip, int target_threads)
2253 {
2254         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2255         struct kvmppc_vcore *pvc, *vcnext;
2256
2257         spin_lock(&lp->lock);
2258         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2259                 if (!spin_trylock(&pvc->lock))
2260                         continue;
2261                 prepare_threads(pvc);
2262                 if (!pvc->n_runnable) {
2263                         list_del_init(&pvc->preempt_list);
2264                         if (pvc->runner == NULL) {
2265                                 pvc->vcore_state = VCORE_INACTIVE;
2266                                 kvmppc_core_end_stolen(pvc);
2267                         }
2268                         spin_unlock(&pvc->lock);
2269                         continue;
2270                 }
2271                 if (!can_piggyback(pvc, cip, target_threads)) {
2272                         spin_unlock(&pvc->lock);
2273                         continue;
2274                 }
2275                 kvmppc_core_end_stolen(pvc);
2276                 pvc->vcore_state = VCORE_PIGGYBACK;
2277                 if (cip->total_threads >= target_threads)
2278                         break;
2279         }
2280         spin_unlock(&lp->lock);
2281 }
2282
2283 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2284 {
2285         int still_running = 0, i;
2286         u64 now;
2287         long ret;
2288         struct kvm_vcpu *vcpu;
2289
2290         spin_lock(&vc->lock);
2291         now = get_tb();
2292         for_each_runnable_thread(i, vcpu, vc) {
2293                 /* cancel pending dec exception if dec is positive */
2294                 if (now < vcpu->arch.dec_expires &&
2295                     kvmppc_core_pending_dec(vcpu))
2296                         kvmppc_core_dequeue_dec(vcpu);
2297
2298                 trace_kvm_guest_exit(vcpu);
2299
2300                 ret = RESUME_GUEST;
2301                 if (vcpu->arch.trap)
2302                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2303                                                     vcpu->arch.run_task);
2304
2305                 vcpu->arch.ret = ret;
2306                 vcpu->arch.trap = 0;
2307
2308                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2309                         if (vcpu->arch.pending_exceptions)
2310                                 kvmppc_core_prepare_to_enter(vcpu);
2311                         if (vcpu->arch.ceded)
2312                                 kvmppc_set_timer(vcpu);
2313                         else
2314                                 ++still_running;
2315                 } else {
2316                         kvmppc_remove_runnable(vc, vcpu);
2317                         wake_up(&vcpu->arch.cpu_run);
2318                 }
2319         }
2320         list_del_init(&vc->preempt_list);
2321         if (!is_master) {
2322                 if (still_running > 0) {
2323                         kvmppc_vcore_preempt(vc);
2324                 } else if (vc->runner) {
2325                         vc->vcore_state = VCORE_PREEMPT;
2326                         kvmppc_core_start_stolen(vc);
2327                 } else {
2328                         vc->vcore_state = VCORE_INACTIVE;
2329                 }
2330                 if (vc->n_runnable > 0 && vc->runner == NULL) {
2331                         /* make sure there's a candidate runner awake */
2332                         i = -1;
2333                         vcpu = next_runnable_thread(vc, &i);
2334                         wake_up(&vcpu->arch.cpu_run);
2335                 }
2336         }
2337         spin_unlock(&vc->lock);
2338 }
2339
2340 /*
2341  * Clear core from the list of active host cores as we are about to
2342  * enter the guest. Only do this if it is the primary thread of the
2343  * core (not if a subcore) that is entering the guest.
2344  */
2345 static inline int kvmppc_clear_host_core(unsigned int cpu)
2346 {
2347         int core;
2348
2349         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2350                 return 0;
2351         /*
2352          * Memory barrier can be omitted here as we will do a smp_wmb()
2353          * later in kvmppc_start_thread and we need ensure that state is
2354          * visible to other CPUs only after we enter guest.
2355          */
2356         core = cpu >> threads_shift;
2357         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2358         return 0;
2359 }
2360
2361 /*
2362  * Advertise this core as an active host core since we exited the guest
2363  * Only need to do this if it is the primary thread of the core that is
2364  * exiting.
2365  */
2366 static inline int kvmppc_set_host_core(unsigned int cpu)
2367 {
2368         int core;
2369
2370         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2371                 return 0;
2372
2373         /*
2374          * Memory barrier can be omitted here because we do a spin_unlock
2375          * immediately after this which provides the memory barrier.
2376          */
2377         core = cpu >> threads_shift;
2378         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2379         return 0;
2380 }
2381
2382 /*
2383  * Run a set of guest threads on a physical core.
2384  * Called with vc->lock held.
2385  */
2386 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2387 {
2388         struct kvm_vcpu *vcpu;
2389         int i;
2390         int srcu_idx;
2391         struct core_info core_info;
2392         struct kvmppc_vcore *pvc, *vcnext;
2393         struct kvm_split_mode split_info, *sip;
2394         int split, subcore_size, active;
2395         int sub;
2396         bool thr0_done;
2397         unsigned long cmd_bit, stat_bit;
2398         int pcpu, thr;
2399         int target_threads;
2400         int controlled_threads;
2401
2402         /*
2403          * Remove from the list any threads that have a signal pending
2404          * or need a VPA update done
2405          */
2406         prepare_threads(vc);
2407
2408         /* if the runner is no longer runnable, let the caller pick a new one */
2409         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2410                 return;
2411
2412         /*
2413          * Initialize *vc.
2414          */
2415         init_master_vcore(vc);
2416         vc->preempt_tb = TB_NIL;
2417
2418         /*
2419          * Number of threads that we will be controlling: the same as
2420          * the number of threads per subcore, except on POWER9,
2421          * where it's 1 because the threads are (mostly) independent.
2422          */
2423         controlled_threads = threads_per_vcore();
2424
2425         /*
2426          * Make sure we are running on primary threads, and that secondary
2427          * threads are offline.  Also check if the number of threads in this
2428          * guest are greater than the current system threads per guest.
2429          */
2430         if ((controlled_threads > 1) &&
2431             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2432                 for_each_runnable_thread(i, vcpu, vc) {
2433                         vcpu->arch.ret = -EBUSY;
2434                         kvmppc_remove_runnable(vc, vcpu);
2435                         wake_up(&vcpu->arch.cpu_run);
2436                 }
2437                 goto out;
2438         }
2439
2440         /*
2441          * See if we could run any other vcores on the physical core
2442          * along with this one.
2443          */
2444         init_core_info(&core_info, vc);
2445         pcpu = smp_processor_id();
2446         target_threads = controlled_threads;
2447         if (target_smt_mode && target_smt_mode < target_threads)
2448                 target_threads = target_smt_mode;
2449         if (vc->num_threads < target_threads)
2450                 collect_piggybacks(&core_info, target_threads);
2451
2452         /* Decide on micro-threading (split-core) mode */
2453         subcore_size = threads_per_subcore;
2454         cmd_bit = stat_bit = 0;
2455         split = core_info.n_subcores;
2456         sip = NULL;
2457         if (split > 1) {
2458                 /* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2459                 if (split == 2 && (dynamic_mt_modes & 2)) {
2460                         cmd_bit = HID0_POWER8_1TO2LPAR;
2461                         stat_bit = HID0_POWER8_2LPARMODE;
2462                 } else {
2463                         split = 4;
2464                         cmd_bit = HID0_POWER8_1TO4LPAR;
2465                         stat_bit = HID0_POWER8_4LPARMODE;
2466                 }
2467                 subcore_size = MAX_SMT_THREADS / split;
2468                 sip = &split_info;
2469                 memset(&split_info, 0, sizeof(split_info));
2470                 split_info.rpr = mfspr(SPRN_RPR);
2471                 split_info.pmmar = mfspr(SPRN_PMMAR);
2472                 split_info.ldbar = mfspr(SPRN_LDBAR);
2473                 split_info.subcore_size = subcore_size;
2474                 for (sub = 0; sub < core_info.n_subcores; ++sub)
2475                         split_info.master_vcs[sub] =
2476                                 list_first_entry(&core_info.vcs[sub],
2477                                         struct kvmppc_vcore, preempt_list);
2478                 /* order writes to split_info before kvm_split_mode pointer */
2479                 smp_wmb();
2480         }
2481         pcpu = smp_processor_id();
2482         for (thr = 0; thr < controlled_threads; ++thr)
2483                 paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2484
2485         /* Initiate micro-threading (split-core) if required */
2486         if (cmd_bit) {
2487                 unsigned long hid0 = mfspr(SPRN_HID0);
2488
2489                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2490                 mb();
2491                 mtspr(SPRN_HID0, hid0);
2492                 isync();
2493                 for (;;) {
2494                         hid0 = mfspr(SPRN_HID0);
2495                         if (hid0 & stat_bit)
2496                                 break;
2497                         cpu_relax();
2498                 }
2499         }
2500
2501         kvmppc_clear_host_core(pcpu);
2502
2503         /* Start all the threads */
2504         active = 0;
2505         for (sub = 0; sub < core_info.n_subcores; ++sub) {
2506                 thr = subcore_thread_map[sub];
2507                 thr0_done = false;
2508                 active |= 1 << thr;
2509                 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list) {
2510                         pvc->pcpu = pcpu + thr;
2511                         for_each_runnable_thread(i, vcpu, pvc) {
2512                                 kvmppc_start_thread(vcpu, pvc);
2513                                 kvmppc_create_dtl_entry(vcpu, pvc);
2514                                 trace_kvm_guest_enter(vcpu);
2515                                 if (!vcpu->arch.ptid)
2516                                         thr0_done = true;
2517                                 active |= 1 << (thr + vcpu->arch.ptid);
2518                         }
2519                         /*
2520                          * We need to start the first thread of each subcore
2521                          * even if it doesn't have a vcpu.
2522                          */
2523                         if (pvc->master_vcore == pvc && !thr0_done)
2524                                 kvmppc_start_thread(NULL, pvc);
2525                         thr += pvc->num_threads;
2526                 }
2527         }
2528
2529         /*
2530          * Ensure that split_info.do_nap is set after setting
2531          * the vcore pointer in the PACA of the secondaries.
2532          */
2533         smp_mb();
2534         if (cmd_bit)
2535                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
2536
2537         /*
2538          * When doing micro-threading, poke the inactive threads as well.
2539          * This gets them to the nap instruction after kvm_do_nap,
2540          * which reduces the time taken to unsplit later.
2541          */
2542         if (split > 1)
2543                 for (thr = 1; thr < threads_per_subcore; ++thr)
2544                         if (!(active & (1 << thr)))
2545                                 kvmppc_ipi_thread(pcpu + thr);
2546
2547         vc->vcore_state = VCORE_RUNNING;
2548         preempt_disable();
2549
2550         trace_kvmppc_run_core(vc, 0);
2551
2552         for (sub = 0; sub < core_info.n_subcores; ++sub)
2553                 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list)
2554                         spin_unlock(&pvc->lock);
2555
2556         guest_enter();
2557
2558         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2559
2560         __kvmppc_vcore_entry();
2561
2562         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2563
2564         spin_lock(&vc->lock);
2565         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2566         vc->vcore_state = VCORE_EXITING;
2567
2568         /* wait for secondary threads to finish writing their state to memory */
2569         kvmppc_wait_for_nap();
2570
2571         /* Return to whole-core mode if we split the core earlier */
2572         if (split > 1) {
2573                 unsigned long hid0 = mfspr(SPRN_HID0);
2574                 unsigned long loops = 0;
2575
2576                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2577                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2578                 mb();
2579                 mtspr(SPRN_HID0, hid0);
2580                 isync();
2581                 for (;;) {
2582                         hid0 = mfspr(SPRN_HID0);
2583                         if (!(hid0 & stat_bit))
2584                                 break;
2585                         cpu_relax();
2586                         ++loops;
2587                 }
2588                 split_info.do_nap = 0;
2589         }
2590
2591         /* Let secondaries go back to the offline loop */
2592         for (i = 0; i < controlled_threads; ++i) {
2593                 kvmppc_release_hwthread(pcpu + i);
2594                 if (sip && sip->napped[i])
2595                         kvmppc_ipi_thread(pcpu + i);
2596                 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
2597         }
2598
2599         kvmppc_set_host_core(pcpu);
2600
2601         spin_unlock(&vc->lock);
2602
2603         /* make sure updates to secondary vcpu structs are visible now */
2604         smp_mb();
2605         guest_exit();
2606
2607         for (sub = 0; sub < core_info.n_subcores; ++sub)
2608                 list_for_each_entry_safe(pvc, vcnext, &core_info.vcs[sub],
2609                                          preempt_list)
2610                         post_guest_process(pvc, pvc == vc);
2611
2612         spin_lock(&vc->lock);
2613         preempt_enable();
2614
2615  out:
2616         vc->vcore_state = VCORE_INACTIVE;
2617         trace_kvmppc_run_core(vc, 1);
2618 }
2619
2620 /*
2621  * Wait for some other vcpu thread to execute us, and
2622  * wake us up when we need to handle something in the host.
2623  */
2624 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2625                                  struct kvm_vcpu *vcpu, int wait_state)
2626 {
2627         DEFINE_WAIT(wait);
2628
2629         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2630         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2631                 spin_unlock(&vc->lock);
2632                 schedule();
2633                 spin_lock(&vc->lock);
2634         }
2635         finish_wait(&vcpu->arch.cpu_run, &wait);
2636 }
2637
2638 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
2639 {
2640         /* 10us base */
2641         if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
2642                 vc->halt_poll_ns = 10000;
2643         else
2644                 vc->halt_poll_ns *= halt_poll_ns_grow;
2645 }
2646
2647 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
2648 {
2649         if (halt_poll_ns_shrink == 0)
2650                 vc->halt_poll_ns = 0;
2651         else
2652                 vc->halt_poll_ns /= halt_poll_ns_shrink;
2653 }
2654
2655 /*
2656  * Check to see if any of the runnable vcpus on the vcore have pending
2657  * exceptions or are no longer ceded
2658  */
2659 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
2660 {
2661         struct kvm_vcpu *vcpu;
2662         int i;
2663
2664         for_each_runnable_thread(i, vcpu, vc) {
2665                 if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded ||
2666                     vcpu->arch.prodded)
2667                         return 1;
2668         }
2669
2670         return 0;
2671 }
2672
2673 /*
2674  * All the vcpus in this vcore are idle, so wait for a decrementer
2675  * or external interrupt to one of the vcpus.  vc->lock is held.
2676  */
2677 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2678 {
2679         ktime_t cur, start_poll, start_wait;
2680         int do_sleep = 1;
2681         u64 block_ns;
2682         DECLARE_SWAITQUEUE(wait);
2683
2684         /* Poll for pending exceptions and ceded state */
2685         cur = start_poll = ktime_get();
2686         if (vc->halt_poll_ns) {
2687                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
2688                 ++vc->runner->stat.halt_attempted_poll;
2689
2690                 vc->vcore_state = VCORE_POLLING;
2691                 spin_unlock(&vc->lock);
2692
2693                 do {
2694                         if (kvmppc_vcore_check_block(vc)) {
2695                                 do_sleep = 0;
2696                                 break;
2697                         }
2698                         cur = ktime_get();
2699                 } while (single_task_running() && ktime_before(cur, stop));
2700
2701                 spin_lock(&vc->lock);
2702                 vc->vcore_state = VCORE_INACTIVE;
2703
2704                 if (!do_sleep) {
2705                         ++vc->runner->stat.halt_successful_poll;
2706                         goto out;
2707                 }
2708         }
2709
2710         prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2711
2712         if (kvmppc_vcore_check_block(vc)) {
2713                 finish_swait(&vc->wq, &wait);
2714                 do_sleep = 0;
2715                 /* If we polled, count this as a successful poll */
2716                 if (vc->halt_poll_ns)
2717                         ++vc->runner->stat.halt_successful_poll;
2718                 goto out;
2719         }
2720
2721         start_wait = ktime_get();
2722
2723         vc->vcore_state = VCORE_SLEEPING;
2724         trace_kvmppc_vcore_blocked(vc, 0);
2725         spin_unlock(&vc->lock);
2726         schedule();
2727         finish_swait(&vc->wq, &wait);
2728         spin_lock(&vc->lock);
2729         vc->vcore_state = VCORE_INACTIVE;
2730         trace_kvmppc_vcore_blocked(vc, 1);
2731         ++vc->runner->stat.halt_successful_wait;
2732
2733         cur = ktime_get();
2734
2735 out:
2736         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
2737
2738         /* Attribute wait time */
2739         if (do_sleep) {
2740                 vc->runner->stat.halt_wait_ns +=
2741                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
2742                 /* Attribute failed poll time */
2743                 if (vc->halt_poll_ns)
2744                         vc->runner->stat.halt_poll_fail_ns +=
2745                                 ktime_to_ns(start_wait) -
2746                                 ktime_to_ns(start_poll);
2747         } else {
2748                 /* Attribute successful poll time */
2749                 if (vc->halt_poll_ns)
2750                         vc->runner->stat.halt_poll_success_ns +=
2751                                 ktime_to_ns(cur) -
2752                                 ktime_to_ns(start_poll);
2753         }
2754
2755         /* Adjust poll time */
2756         if (halt_poll_ns) {
2757                 if (block_ns <= vc->halt_poll_ns)
2758                         ;
2759                 /* We slept and blocked for longer than the max halt time */
2760                 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
2761                         shrink_halt_poll_ns(vc);
2762                 /* We slept and our poll time is too small */
2763                 else if (vc->halt_poll_ns < halt_poll_ns &&
2764                                 block_ns < halt_poll_ns)
2765                         grow_halt_poll_ns(vc);
2766                 if (vc->halt_poll_ns > halt_poll_ns)
2767                         vc->halt_poll_ns = halt_poll_ns;
2768         } else
2769                 vc->halt_poll_ns = 0;
2770
2771         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
2772 }
2773
2774 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2775 {
2776         int n_ceded, i;
2777         struct kvmppc_vcore *vc;
2778         struct kvm_vcpu *v;
2779
2780         trace_kvmppc_run_vcpu_enter(vcpu);
2781
2782         kvm_run->exit_reason = 0;
2783         vcpu->arch.ret = RESUME_GUEST;
2784         vcpu->arch.trap = 0;
2785         kvmppc_update_vpas(vcpu);
2786
2787         /*
2788          * Synchronize with other threads in this virtual core
2789          */
2790         vc = vcpu->arch.vcore;
2791         spin_lock(&vc->lock);
2792         vcpu->arch.ceded = 0;
2793         vcpu->arch.run_task = current;
2794         vcpu->arch.kvm_run = kvm_run;
2795         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
2796         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
2797         vcpu->arch.busy_preempt = TB_NIL;
2798         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
2799         ++vc->n_runnable;
2800
2801         /*
2802          * This happens the first time this is called for a vcpu.
2803          * If the vcore is already running, we may be able to start
2804          * this thread straight away and have it join in.
2805          */
2806         if (!signal_pending(current)) {
2807                 if (vc->vcore_state == VCORE_PIGGYBACK) {
2808                         struct kvmppc_vcore *mvc = vc->master_vcore;
2809                         if (spin_trylock(&mvc->lock)) {
2810                                 if (mvc->vcore_state == VCORE_RUNNING &&
2811                                     !VCORE_IS_EXITING(mvc)) {
2812                                         kvmppc_create_dtl_entry(vcpu, vc);
2813                                         kvmppc_start_thread(vcpu, vc);
2814                                         trace_kvm_guest_enter(vcpu);
2815                                 }
2816                                 spin_unlock(&mvc->lock);
2817                         }
2818                 } else if (vc->vcore_state == VCORE_RUNNING &&
2819                            !VCORE_IS_EXITING(vc)) {
2820                         kvmppc_create_dtl_entry(vcpu, vc);
2821                         kvmppc_start_thread(vcpu, vc);
2822                         trace_kvm_guest_enter(vcpu);
2823                 } else if (vc->vcore_state == VCORE_SLEEPING) {
2824                         swake_up(&vc->wq);
2825                 }
2826
2827         }
2828
2829         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2830                !signal_pending(current)) {
2831                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2832                         kvmppc_vcore_end_preempt(vc);
2833
2834                 if (vc->vcore_state != VCORE_INACTIVE) {
2835                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
2836                         continue;
2837                 }
2838                 for_each_runnable_thread(i, v, vc) {
2839                         kvmppc_core_prepare_to_enter(v);
2840                         if (signal_pending(v->arch.run_task)) {
2841                                 kvmppc_remove_runnable(vc, v);
2842                                 v->stat.signal_exits++;
2843                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2844                                 v->arch.ret = -EINTR;
2845                                 wake_up(&v->arch.cpu_run);
2846                         }
2847                 }
2848                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2849                         break;
2850                 n_ceded = 0;
2851                 for_each_runnable_thread(i, v, vc) {
2852                         if (!v->arch.pending_exceptions && !v->arch.prodded)
2853                                 n_ceded += v->arch.ceded;
2854                         else
2855                                 v->arch.ceded = 0;
2856                 }
2857                 vc->runner = vcpu;
2858                 if (n_ceded == vc->n_runnable) {
2859                         kvmppc_vcore_blocked(vc);
2860                 } else if (need_resched()) {
2861                         kvmppc_vcore_preempt(vc);
2862                         /* Let something else run */
2863                         cond_resched_lock(&vc->lock);
2864                         if (vc->vcore_state == VCORE_PREEMPT)
2865                                 kvmppc_vcore_end_preempt(vc);
2866                 } else {
2867                         kvmppc_run_core(vc);
2868                 }
2869                 vc->runner = NULL;
2870         }
2871
2872         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2873                (vc->vcore_state == VCORE_RUNNING ||
2874                 vc->vcore_state == VCORE_EXITING ||
2875                 vc->vcore_state == VCORE_PIGGYBACK))
2876                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
2877
2878         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2879                 kvmppc_vcore_end_preempt(vc);
2880
2881         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2882                 kvmppc_remove_runnable(vc, vcpu);
2883                 vcpu->stat.signal_exits++;
2884                 kvm_run->exit_reason = KVM_EXIT_INTR;
2885                 vcpu->arch.ret = -EINTR;
2886         }
2887
2888         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2889                 /* Wake up some vcpu to run the core */
2890                 i = -1;
2891                 v = next_runnable_thread(vc, &i);
2892                 wake_up(&v->arch.cpu_run);
2893         }
2894
2895         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2896         spin_unlock(&vc->lock);
2897         return vcpu->arch.ret;
2898 }
2899
2900 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2901 {
2902         int r;
2903         int srcu_idx;
2904
2905         if (!vcpu->arch.sane) {
2906                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2907                 return -EINVAL;
2908         }
2909
2910         kvmppc_core_prepare_to_enter(vcpu);
2911
2912         /* No need to go into the guest when all we'll do is come back out */
2913         if (signal_pending(current)) {
2914                 run->exit_reason = KVM_EXIT_INTR;
2915                 return -EINTR;
2916         }
2917
2918         atomic_inc(&vcpu->kvm->arch.vcpus_running);
2919         /* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2920         smp_mb();
2921
2922         /* On the first time here, set up HTAB and VRMA */
2923         if (!kvm_is_radix(vcpu->kvm) && !vcpu->kvm->arch.hpte_setup_done) {
2924                 r = kvmppc_hv_setup_htab_rma(vcpu);
2925                 if (r)
2926                         goto out;
2927         }
2928
2929         flush_all_to_thread(current);
2930
2931         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2932         vcpu->arch.pgdir = current->mm->pgd;
2933         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2934
2935         do {
2936                 r = kvmppc_run_vcpu(run, vcpu);
2937
2938                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2939                     !(vcpu->arch.shregs.msr & MSR_PR)) {
2940                         trace_kvm_hcall_enter(vcpu);
2941                         r = kvmppc_pseries_do_hcall(vcpu);
2942                         trace_kvm_hcall_exit(vcpu, r);
2943                         kvmppc_core_prepare_to_enter(vcpu);
2944                 } else if (r == RESUME_PAGE_FAULT) {
2945                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2946                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
2947                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2948                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2949                 } else if (r == RESUME_PASSTHROUGH)
2950                         r = kvmppc_xics_rm_complete(vcpu, 0);
2951         } while (is_kvmppc_resume_guest(r));
2952
2953  out:
2954         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2955         atomic_dec(&vcpu->kvm->arch.vcpus_running);
2956         return r;
2957 }
2958
2959 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2960                                      int linux_psize)
2961 {
2962         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2963
2964         if (!def->shift)
2965                 return;
2966         (*sps)->page_shift = def->shift;
2967         (*sps)->slb_enc = def->sllp;
2968         (*sps)->enc[0].page_shift = def->shift;
2969         (*sps)->enc[0].pte_enc = def->penc[linux_psize];
2970         /*
2971          * Add 16MB MPSS support if host supports it
2972          */
2973         if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2974                 (*sps)->enc[1].page_shift = 24;
2975                 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2976         }
2977         (*sps)++;
2978 }
2979
2980 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2981                                          struct kvm_ppc_smmu_info *info)
2982 {
2983         struct kvm_ppc_one_seg_page_size *sps;
2984
2985         /*
2986          * Since we don't yet support HPT guests on a radix host,
2987          * return an error if the host uses radix.
2988          */
2989         if (radix_enabled())
2990                 return -EINVAL;
2991
2992         info->flags = KVM_PPC_PAGE_SIZES_REAL;
2993         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2994                 info->flags |= KVM_PPC_1T_SEGMENTS;
2995         info->slb_size = mmu_slb_size;
2996
2997         /* We only support these sizes for now, and no muti-size segments */
2998         sps = &info->sps[0];
2999         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
3000         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
3001         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
3002
3003         return 0;
3004 }
3005
3006 /*
3007  * Get (and clear) the dirty memory log for a memory slot.
3008  */
3009 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
3010                                          struct kvm_dirty_log *log)
3011 {
3012         struct kvm_memslots *slots;
3013         struct kvm_memory_slot *memslot;
3014         int i, r;
3015         unsigned long n;
3016         unsigned long *buf;
3017         struct kvm_vcpu *vcpu;
3018
3019         mutex_lock(&kvm->slots_lock);
3020
3021         r = -EINVAL;
3022         if (log->slot >= KVM_USER_MEM_SLOTS)
3023                 goto out;
3024
3025         slots = kvm_memslots(kvm);
3026         memslot = id_to_memslot(slots, log->slot);
3027         r = -ENOENT;
3028         if (!memslot->dirty_bitmap)
3029                 goto out;
3030
3031         /*
3032          * Use second half of bitmap area because radix accumulates
3033          * bits in the first half.
3034          */
3035         n = kvm_dirty_bitmap_bytes(memslot);
3036         buf = memslot->dirty_bitmap + n / sizeof(long);
3037         memset(buf, 0, n);
3038
3039         if (kvm_is_radix(kvm))
3040                 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
3041         else
3042                 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
3043         if (r)
3044                 goto out;
3045
3046         /* Harvest dirty bits from VPA and DTL updates */
3047         /* Note: we never modify the SLB shadow buffer areas */
3048         kvm_for_each_vcpu(i, vcpu, kvm) {
3049                 spin_lock(&vcpu->arch.vpa_update_lock);
3050                 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
3051                 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
3052                 spin_unlock(&vcpu->arch.vpa_update_lock);
3053         }
3054
3055         r = -EFAULT;
3056         if (copy_to_user(log->dirty_bitmap, buf, n))
3057                 goto out;
3058
3059         r = 0;
3060 out:
3061         mutex_unlock(&kvm->slots_lock);
3062         return r;
3063 }
3064
3065 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
3066                                         struct kvm_memory_slot *dont)
3067 {
3068         if (!dont || free->arch.rmap != dont->arch.rmap) {
3069                 vfree(free->arch.rmap);
3070                 free->arch.rmap = NULL;
3071         }
3072 }
3073
3074 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
3075                                          unsigned long npages)
3076 {
3077         /*
3078          * For now, if radix_enabled() then we only support radix guests,
3079          * and in that case we don't need the rmap array.
3080          */
3081         if (radix_enabled()) {
3082                 slot->arch.rmap = NULL;
3083                 return 0;
3084         }
3085
3086         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
3087         if (!slot->arch.rmap)
3088                 return -ENOMEM;
3089
3090         return 0;
3091 }
3092
3093 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
3094                                         struct kvm_memory_slot *memslot,
3095                                         const struct kvm_userspace_memory_region *mem)
3096 {
3097         return 0;
3098 }
3099
3100 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
3101                                 const struct kvm_userspace_memory_region *mem,
3102                                 const struct kvm_memory_slot *old,
3103                                 const struct kvm_memory_slot *new)
3104 {
3105         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
3106         struct kvm_memslots *slots;
3107         struct kvm_memory_slot *memslot;
3108
3109         /*
3110          * If we are making a new memslot, it might make
3111          * some address that was previously cached as emulated
3112          * MMIO be no longer emulated MMIO, so invalidate
3113          * all the caches of emulated MMIO translations.
3114          */
3115         if (npages)
3116                 atomic64_inc(&kvm->arch.mmio_update);
3117
3118         if (npages && old->npages && !kvm_is_radix(kvm)) {
3119                 /*
3120                  * If modifying a memslot, reset all the rmap dirty bits.
3121                  * If this is a new memslot, we don't need to do anything
3122                  * since the rmap array starts out as all zeroes,
3123                  * i.e. no pages are dirty.
3124                  */
3125                 slots = kvm_memslots(kvm);
3126                 memslot = id_to_memslot(slots, mem->slot);
3127                 kvmppc_hv_get_dirty_log_hpt(kvm, memslot, NULL);
3128         }
3129 }
3130
3131 /*
3132  * Update LPCR values in kvm->arch and in vcores.
3133  * Caller must hold kvm->lock.
3134  */
3135 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3136 {
3137         long int i;
3138         u32 cores_done = 0;
3139
3140         if ((kvm->arch.lpcr & mask) == lpcr)
3141                 return;
3142
3143         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3144
3145         for (i = 0; i < KVM_MAX_VCORES; ++i) {
3146                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3147                 if (!vc)
3148                         continue;
3149                 spin_lock(&vc->lock);
3150                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3151                 spin_unlock(&vc->lock);
3152                 if (++cores_done >= kvm->arch.online_vcores)
3153                         break;
3154         }
3155 }
3156
3157 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3158 {
3159         return;
3160 }
3161
3162 static void kvmppc_setup_partition_table(struct kvm *kvm)
3163 {
3164         unsigned long dw0, dw1;
3165
3166         if (!kvm_is_radix(kvm)) {
3167                 /* PS field - page size for VRMA */
3168                 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
3169                         ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
3170                 /* HTABSIZE and HTABORG fields */
3171                 dw0 |= kvm->arch.sdr1;
3172
3173                 /* Second dword as set by userspace */
3174                 dw1 = kvm->arch.process_table;
3175         } else {
3176                 dw0 = PATB_HR | radix__get_tree_size() |
3177                         __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
3178                 dw1 = PATB_GR | kvm->arch.process_table;
3179         }
3180
3181         mmu_partition_table_set_entry(kvm->arch.lpid, dw0, dw1);
3182 }
3183
3184 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3185 {
3186         int err = 0;
3187         struct kvm *kvm = vcpu->kvm;
3188         unsigned long hva;
3189         struct kvm_memory_slot *memslot;
3190         struct vm_area_struct *vma;
3191         unsigned long lpcr = 0, senc;
3192         unsigned long psize, porder;
3193         int srcu_idx;
3194
3195         mutex_lock(&kvm->lock);
3196         if (kvm->arch.hpte_setup_done)
3197                 goto out;       /* another vcpu beat us to it */
3198
3199         /* Allocate hashed page table (if not done already) and reset it */
3200         if (!kvm->arch.hpt.virt) {
3201                 int order = KVM_DEFAULT_HPT_ORDER;
3202                 struct kvm_hpt_info info;
3203
3204                 err = kvmppc_allocate_hpt(&info, order);
3205                 /* If we get here, it means userspace didn't specify a
3206                  * size explicitly.  So, try successively smaller
3207                  * sizes if the default failed. */
3208                 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
3209                         err  = kvmppc_allocate_hpt(&info, order);
3210
3211                 if (err < 0) {
3212                         pr_err("KVM: Couldn't alloc HPT\n");
3213                         goto out;
3214                 }
3215
3216                 kvmppc_set_hpt(kvm, &info);
3217         }
3218
3219         /* Look up the memslot for guest physical address 0 */
3220         srcu_idx = srcu_read_lock(&kvm->srcu);
3221         memslot = gfn_to_memslot(kvm, 0);
3222
3223         /* We must have some memory at 0 by now */
3224         err = -EINVAL;
3225         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3226                 goto out_srcu;
3227
3228         /* Look up the VMA for the start of this memory slot */
3229         hva = memslot->userspace_addr;
3230         down_read(&current->mm->mmap_sem);
3231         vma = find_vma(current->mm, hva);
3232         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3233                 goto up_out;
3234
3235         psize = vma_kernel_pagesize(vma);
3236         porder = __ilog2(psize);
3237
3238         up_read(&current->mm->mmap_sem);
3239
3240         /* We can handle 4k, 64k or 16M pages in the VRMA */
3241         err = -EINVAL;
3242         if (!(psize == 0x1000 || psize == 0x10000 ||
3243               psize == 0x1000000))
3244                 goto out_srcu;
3245
3246         senc = slb_pgsize_encoding(psize);
3247         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3248                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3249         /* Create HPTEs in the hash page table for the VRMA */
3250         kvmppc_map_vrma(vcpu, memslot, porder);
3251
3252         /* Update VRMASD field in the LPCR */
3253         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3254                 /* the -4 is to account for senc values starting at 0x10 */
3255                 lpcr = senc << (LPCR_VRMASD_SH - 4);
3256                 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3257         } else {
3258                 kvmppc_setup_partition_table(kvm);
3259         }
3260
3261         /* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
3262         smp_wmb();
3263         kvm->arch.hpte_setup_done = 1;
3264         err = 0;
3265  out_srcu:
3266         srcu_read_unlock(&kvm->srcu, srcu_idx);
3267  out:
3268         mutex_unlock(&kvm->lock);
3269         return err;
3270
3271  up_out:
3272         up_read(&current->mm->mmap_sem);
3273         goto out_srcu;
3274 }
3275
3276 #ifdef CONFIG_KVM_XICS
3277 /*
3278  * Allocate a per-core structure for managing state about which cores are
3279  * running in the host versus the guest and for exchanging data between
3280  * real mode KVM and CPU running in the host.
3281  * This is only done for the first VM.
3282  * The allocated structure stays even if all VMs have stopped.
3283  * It is only freed when the kvm-hv module is unloaded.
3284  * It's OK for this routine to fail, we just don't support host
3285  * core operations like redirecting H_IPI wakeups.
3286  */
3287 void kvmppc_alloc_host_rm_ops(void)
3288 {
3289         struct kvmppc_host_rm_ops *ops;
3290         unsigned long l_ops;
3291         int cpu, core;
3292         int size;
3293
3294         /* Not the first time here ? */
3295         if (kvmppc_host_rm_ops_hv != NULL)
3296                 return;
3297
3298         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3299         if (!ops)
3300                 return;
3301
3302         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3303         ops->rm_core = kzalloc(size, GFP_KERNEL);
3304
3305         if (!ops->rm_core) {
3306                 kfree(ops);
3307                 return;
3308         }
3309
3310         get_online_cpus();
3311
3312         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3313                 if (!cpu_online(cpu))
3314                         continue;
3315
3316                 core = cpu >> threads_shift;
3317                 ops->rm_core[core].rm_state.in_host = 1;
3318         }
3319
3320         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3321
3322         /*
3323          * Make the contents of the kvmppc_host_rm_ops structure visible
3324          * to other CPUs before we assign it to the global variable.
3325          * Do an atomic assignment (no locks used here), but if someone
3326          * beats us to it, just free our copy and return.
3327          */
3328         smp_wmb();
3329         l_ops = (unsigned long) ops;
3330
3331         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3332                 put_online_cpus();
3333                 kfree(ops->rm_core);
3334                 kfree(ops);
3335                 return;
3336         }
3337
3338         cpuhp_setup_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE,
3339                                   "ppc/kvm_book3s:prepare",
3340                                   kvmppc_set_host_core,
3341                                   kvmppc_clear_host_core);
3342         put_online_cpus();
3343 }
3344
3345 void kvmppc_free_host_rm_ops(void)
3346 {
3347         if (kvmppc_host_rm_ops_hv) {
3348                 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
3349                 kfree(kvmppc_host_rm_ops_hv->rm_core);
3350                 kfree(kvmppc_host_rm_ops_hv);
3351                 kvmppc_host_rm_ops_hv = NULL;
3352         }
3353 }
3354 #endif
3355
3356 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3357 {
3358         unsigned long lpcr, lpid;
3359         char buf[32];
3360         int ret;
3361
3362         /* Allocate the guest's logical partition ID */
3363
3364         lpid = kvmppc_alloc_lpid();
3365         if ((long)lpid < 0)
3366                 return -ENOMEM;
3367         kvm->arch.lpid = lpid;
3368
3369         kvmppc_alloc_host_rm_ops();
3370
3371         /*
3372          * Since we don't flush the TLB when tearing down a VM,
3373          * and this lpid might have previously been used,
3374          * make sure we flush on each core before running the new VM.
3375          * On POWER9, the tlbie in mmu_partition_table_set_entry()
3376          * does this flush for us.
3377          */
3378         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3379                 cpumask_setall(&kvm->arch.need_tlb_flush);
3380
3381         /* Start out with the default set of hcalls enabled */
3382         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3383                sizeof(kvm->arch.enabled_hcalls));
3384
3385         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3386                 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3387
3388         /* Init LPCR for virtual RMA mode */
3389         kvm->arch.host_lpid = mfspr(SPRN_LPID);
3390         kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3391         lpcr &= LPCR_PECE | LPCR_LPES;
3392         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3393                 LPCR_VPM0 | LPCR_VPM1;
3394         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3395                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3396         /* On POWER8 turn on online bit to enable PURR/SPURR */
3397         if (cpu_has_feature(CPU_FTR_ARCH_207S))
3398                 lpcr |= LPCR_ONL;
3399         /*
3400          * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
3401          * Set HVICE bit to enable hypervisor virtualization interrupts.
3402          */
3403         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3404                 lpcr &= ~LPCR_VPM0;
3405                 lpcr |= LPCR_HVICE;
3406         }
3407
3408         /*
3409          * For now, if the host uses radix, the guest must be radix.
3410          */
3411         if (radix_enabled()) {
3412                 kvm->arch.radix = 1;
3413                 lpcr &= ~LPCR_VPM1;
3414                 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
3415                 ret = kvmppc_init_vm_radix(kvm);
3416                 if (ret) {
3417                         kvmppc_free_lpid(kvm->arch.lpid);
3418                         return ret;
3419                 }
3420                 kvmppc_setup_partition_table(kvm);
3421         }
3422
3423         kvm->arch.lpcr = lpcr;
3424
3425         /* Initialization for future HPT resizes */
3426         kvm->arch.resize_hpt = NULL;
3427
3428         /*
3429          * Work out how many sets the TLB has, for the use of
3430          * the TLB invalidation loop in book3s_hv_rmhandlers.S.
3431          */
3432         if (kvm_is_radix(kvm))
3433                 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;     /* 128 */
3434         else if (cpu_has_feature(CPU_FTR_ARCH_300))
3435                 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;      /* 256 */
3436         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
3437                 kvm->arch.tlb_sets = POWER8_TLB_SETS;           /* 512 */
3438         else
3439                 kvm->arch.tlb_sets = POWER7_TLB_SETS;           /* 128 */
3440
3441         /*
3442          * Track that we now have a HV mode VM active. This blocks secondary
3443          * CPU threads from coming online.
3444          * On POWER9, we only need to do this for HPT guests on a radix
3445          * host, which is not yet supported.
3446          */
3447         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3448                 kvm_hv_vm_activated();
3449
3450         /*
3451          * Create a debugfs directory for the VM
3452          */
3453         snprintf(buf, sizeof(buf), "vm%d", current->pid);
3454         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3455         if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3456                 kvmppc_mmu_debugfs_init(kvm);
3457
3458         return 0;
3459 }
3460
3461 static void kvmppc_free_vcores(struct kvm *kvm)
3462 {
3463         long int i;
3464
3465         for (i = 0; i < KVM_MAX_VCORES; ++i)
3466                 kfree(kvm->arch.vcores[i]);
3467         kvm->arch.online_vcores = 0;
3468 }
3469
3470 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3471 {
3472         debugfs_remove_recursive(kvm->arch.debugfs_dir);
3473
3474         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3475                 kvm_hv_vm_deactivated();
3476
3477         kvmppc_free_vcores(kvm);
3478
3479         kvmppc_free_lpid(kvm->arch.lpid);
3480
3481         if (kvm_is_radix(kvm))
3482                 kvmppc_free_radix(kvm);
3483         else
3484                 kvmppc_free_hpt(&kvm->arch.hpt);
3485
3486         kvmppc_free_pimap(kvm);
3487 }
3488
3489 /* We don't need to emulate any privileged instructions or dcbz */
3490 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3491                                      unsigned int inst, int *advance)
3492 {
3493         return EMULATE_FAIL;
3494 }
3495
3496 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3497                                         ulong spr_val)
3498 {
3499         return EMULATE_FAIL;
3500 }
3501
3502 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3503                                         ulong *spr_val)
3504 {
3505         return EMULATE_FAIL;
3506 }
3507
3508 static int kvmppc_core_check_processor_compat_hv(void)
3509 {
3510         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3511             !cpu_has_feature(CPU_FTR_ARCH_206))
3512                 return -EIO;
3513
3514         return 0;
3515 }
3516
3517 #ifdef CONFIG_KVM_XICS
3518
3519 void kvmppc_free_pimap(struct kvm *kvm)
3520 {
3521         kfree(kvm->arch.pimap);
3522 }
3523
3524 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
3525 {
3526         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
3527 }
3528
3529 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3530 {
3531         struct irq_desc *desc;
3532         struct kvmppc_irq_map *irq_map;
3533         struct kvmppc_passthru_irqmap *pimap;
3534         struct irq_chip *chip;
3535         int i;
3536
3537         if (!kvm_irq_bypass)
3538                 return 1;
3539
3540         desc = irq_to_desc(host_irq);
3541         if (!desc)
3542                 return -EIO;
3543
3544         mutex_lock(&kvm->lock);
3545
3546         pimap = kvm->arch.pimap;
3547         if (pimap == NULL) {
3548                 /* First call, allocate structure to hold IRQ map */
3549                 pimap = kvmppc_alloc_pimap();
3550                 if (pimap == NULL) {
3551                         mutex_unlock(&kvm->lock);
3552                         return -ENOMEM;
3553                 }
3554                 kvm->arch.pimap = pimap;
3555         }
3556
3557         /*
3558          * For now, we only support interrupts for which the EOI operation
3559          * is an OPAL call followed by a write to XIRR, since that's
3560          * what our real-mode EOI code does.
3561          */
3562         chip = irq_data_get_irq_chip(&desc->irq_data);
3563         if (!chip || !is_pnv_opal_msi(chip)) {
3564                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
3565                         host_irq, guest_gsi);
3566                 mutex_unlock(&kvm->lock);
3567                 return -ENOENT;
3568         }
3569
3570         /*
3571          * See if we already have an entry for this guest IRQ number.
3572          * If it's mapped to a hardware IRQ number, that's an error,
3573          * otherwise re-use this entry.
3574          */
3575         for (i = 0; i < pimap->n_mapped; i++) {
3576                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
3577                         if (pimap->mapped[i].r_hwirq) {
3578                                 mutex_unlock(&kvm->lock);
3579                                 return -EINVAL;
3580                         }
3581                         break;
3582                 }
3583         }
3584
3585         if (i == KVMPPC_PIRQ_MAPPED) {
3586                 mutex_unlock(&kvm->lock);
3587                 return -EAGAIN;         /* table is full */
3588         }
3589
3590         irq_map = &pimap->mapped[i];
3591
3592         irq_map->v_hwirq = guest_gsi;
3593         irq_map->desc = desc;
3594
3595         /*
3596          * Order the above two stores before the next to serialize with
3597          * the KVM real mode handler.
3598          */
3599         smp_wmb();
3600         irq_map->r_hwirq = desc->irq_data.hwirq;
3601
3602         if (i == pimap->n_mapped)
3603                 pimap->n_mapped++;
3604
3605         kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
3606
3607         mutex_unlock(&kvm->lock);
3608
3609         return 0;
3610 }
3611
3612 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3613 {
3614         struct irq_desc *desc;
3615         struct kvmppc_passthru_irqmap *pimap;
3616         int i;
3617
3618         if (!kvm_irq_bypass)
3619                 return 0;
3620
3621         desc = irq_to_desc(host_irq);
3622         if (!desc)
3623                 return -EIO;
3624
3625         mutex_lock(&kvm->lock);
3626
3627         if (kvm->arch.pimap == NULL) {
3628                 mutex_unlock(&kvm->lock);
3629                 return 0;
3630         }
3631         pimap = kvm->arch.pimap;
3632
3633         for (i = 0; i < pimap->n_mapped; i++) {
3634                 if (guest_gsi == pimap->mapped[i].v_hwirq)
3635                         break;
3636         }
3637
3638         if (i == pimap->n_mapped) {
3639                 mutex_unlock(&kvm->lock);
3640                 return -ENODEV;
3641         }
3642
3643         kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
3644
3645         /* invalidate the entry */
3646         pimap->mapped[i].r_hwirq = 0;
3647
3648         /*
3649          * We don't free this structure even when the count goes to
3650          * zero. The structure is freed when we destroy the VM.
3651          */
3652
3653         mutex_unlock(&kvm->lock);
3654         return 0;
3655 }
3656
3657 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
3658                                              struct irq_bypass_producer *prod)
3659 {
3660         int ret = 0;
3661         struct kvm_kernel_irqfd *irqfd =
3662                 container_of(cons, struct kvm_kernel_irqfd, consumer);
3663
3664         irqfd->producer = prod;
3665
3666         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
3667         if (ret)
3668                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
3669                         prod->irq, irqfd->gsi, ret);
3670
3671         return ret;
3672 }
3673
3674 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
3675                                               struct irq_bypass_producer *prod)
3676 {
3677         int ret;
3678         struct kvm_kernel_irqfd *irqfd =
3679                 container_of(cons, struct kvm_kernel_irqfd, consumer);
3680
3681         irqfd->producer = NULL;
3682
3683         /*
3684          * When producer of consumer is unregistered, we change back to
3685          * default external interrupt handling mode - KVM real mode
3686          * will switch back to host.
3687          */
3688         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
3689         if (ret)
3690                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
3691                         prod->irq, irqfd->gsi, ret);
3692 }
3693 #endif
3694
3695 static long kvm_arch_vm_ioctl_hv(struct file *filp,
3696                                  unsigned int ioctl, unsigned long arg)
3697 {
3698         struct kvm *kvm __maybe_unused = filp->private_data;
3699         void __user *argp = (void __user *)arg;
3700         long r;
3701
3702         switch (ioctl) {
3703
3704         case KVM_PPC_ALLOCATE_HTAB: {
3705                 u32 htab_order;
3706
3707                 r = -EFAULT;
3708                 if (get_user(htab_order, (u32 __user *)argp))
3709                         break;
3710                 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
3711                 if (r)
3712                         break;
3713                 r = 0;
3714                 break;
3715         }
3716
3717         case KVM_PPC_GET_HTAB_FD: {
3718                 struct kvm_get_htab_fd ghf;
3719
3720                 r = -EFAULT;
3721                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
3722                         break;
3723                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
3724                 break;
3725         }
3726
3727         case KVM_PPC_RESIZE_HPT_PREPARE: {
3728                 struct kvm_ppc_resize_hpt rhpt;
3729
3730                 r = -EFAULT;
3731                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
3732                         break;
3733
3734                 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
3735                 break;
3736         }
3737
3738         case KVM_PPC_RESIZE_HPT_COMMIT: {
3739                 struct kvm_ppc_resize_hpt rhpt;
3740
3741                 r = -EFAULT;
3742                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
3743                         break;
3744
3745                 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
3746                 break;
3747         }
3748
3749         default:
3750                 r = -ENOTTY;
3751         }
3752
3753         return r;
3754 }
3755
3756 /*
3757  * List of hcall numbers to enable by default.
3758  * For compatibility with old userspace, we enable by default
3759  * all hcalls that were implemented before the hcall-enabling
3760  * facility was added.  Note this list should not include H_RTAS.
3761  */
3762 static unsigned int default_hcall_list[] = {
3763         H_REMOVE,
3764         H_ENTER,
3765         H_READ,
3766         H_PROTECT,
3767         H_BULK_REMOVE,
3768         H_GET_TCE,
3769         H_PUT_TCE,
3770         H_SET_DABR,
3771         H_SET_XDABR,
3772         H_CEDE,
3773         H_PROD,
3774         H_CONFER,
3775         H_REGISTER_VPA,
3776 #ifdef CONFIG_KVM_XICS
3777         H_EOI,
3778         H_CPPR,
3779         H_IPI,
3780         H_IPOLL,
3781         H_XIRR,
3782         H_XIRR_X,
3783 #endif
3784         0
3785 };
3786
3787 static void init_default_hcalls(void)
3788 {
3789         int i;
3790         unsigned int hcall;
3791
3792         for (i = 0; default_hcall_list[i]; ++i) {
3793                 hcall = default_hcall_list[i];
3794                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
3795                 __set_bit(hcall / 4, default_enabled_hcalls);
3796         }
3797 }
3798
3799 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
3800 {
3801         unsigned long lpcr;
3802         int radix;
3803
3804         /* If not on a POWER9, reject it */
3805         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3806                 return -ENODEV;
3807
3808         /* If any unknown flags set, reject it */
3809         if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
3810                 return -EINVAL;
3811
3812         /* We can't change a guest to/from radix yet */
3813         radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
3814         if (radix != kvm_is_radix(kvm))
3815                 return -EINVAL;
3816
3817         /* GR (guest radix) bit in process_table field must match */
3818         if (!!(cfg->process_table & PATB_GR) != radix)
3819                 return -EINVAL;
3820
3821         /* Process table size field must be reasonable, i.e. <= 24 */
3822         if ((cfg->process_table & PRTS_MASK) > 24)
3823                 return -EINVAL;
3824
3825         kvm->arch.process_table = cfg->process_table;
3826         kvmppc_setup_partition_table(kvm);
3827
3828         lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
3829         kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
3830
3831         return 0;
3832 }
3833
3834 static struct kvmppc_ops kvm_ops_hv = {
3835         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
3836         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
3837         .get_one_reg = kvmppc_get_one_reg_hv,
3838         .set_one_reg = kvmppc_set_one_reg_hv,
3839         .vcpu_load   = kvmppc_core_vcpu_load_hv,
3840         .vcpu_put    = kvmppc_core_vcpu_put_hv,
3841         .set_msr     = kvmppc_set_msr_hv,
3842         .vcpu_run    = kvmppc_vcpu_run_hv,
3843         .vcpu_create = kvmppc_core_vcpu_create_hv,
3844         .vcpu_free   = kvmppc_core_vcpu_free_hv,
3845         .check_requests = kvmppc_core_check_requests_hv,
3846         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
3847         .flush_memslot  = kvmppc_core_flush_memslot_hv,
3848         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
3849         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
3850         .unmap_hva = kvm_unmap_hva_hv,
3851         .unmap_hva_range = kvm_unmap_hva_range_hv,
3852         .age_hva  = kvm_age_hva_hv,
3853         .test_age_hva = kvm_test_age_hva_hv,
3854         .set_spte_hva = kvm_set_spte_hva_hv,
3855         .mmu_destroy  = kvmppc_mmu_destroy_hv,
3856         .free_memslot = kvmppc_core_free_memslot_hv,
3857         .create_memslot = kvmppc_core_create_memslot_hv,
3858         .init_vm =  kvmppc_core_init_vm_hv,
3859         .destroy_vm = kvmppc_core_destroy_vm_hv,
3860         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
3861         .emulate_op = kvmppc_core_emulate_op_hv,
3862         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
3863         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
3864         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
3865         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
3866         .hcall_implemented = kvmppc_hcall_impl_hv,
3867 #ifdef CONFIG_KVM_XICS
3868         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
3869         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
3870 #endif
3871         .configure_mmu = kvmhv_configure_mmu,
3872         .get_rmmu_info = kvmhv_get_rmmu_info,
3873 };
3874
3875 static int kvm_init_subcore_bitmap(void)
3876 {
3877         int i, j;
3878         int nr_cores = cpu_nr_cores();
3879         struct sibling_subcore_state *sibling_subcore_state;
3880
3881         for (i = 0; i < nr_cores; i++) {
3882                 int first_cpu = i * threads_per_core;
3883                 int node = cpu_to_node(first_cpu);
3884
3885                 /* Ignore if it is already allocated. */
3886                 if (paca[first_cpu].sibling_subcore_state)
3887                         continue;
3888
3889                 sibling_subcore_state =
3890                         kmalloc_node(sizeof(struct sibling_subcore_state),
3891                                                         GFP_KERNEL, node);
3892                 if (!sibling_subcore_state)
3893                         return -ENOMEM;
3894
3895                 memset(sibling_subcore_state, 0,
3896                                 sizeof(struct sibling_subcore_state));
3897
3898                 for (j = 0; j < threads_per_core; j++) {
3899                         int cpu = first_cpu + j;
3900
3901                         paca[cpu].sibling_subcore_state = sibling_subcore_state;
3902                 }
3903         }
3904         return 0;
3905 }
3906
3907 static int kvmppc_radix_possible(void)
3908 {
3909         return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
3910 }
3911
3912 static int kvmppc_book3s_init_hv(void)
3913 {
3914         int r;
3915         /*
3916          * FIXME!! Do we need to check on all cpus ?
3917          */
3918         r = kvmppc_core_check_processor_compat_hv();
3919         if (r < 0)
3920                 return -ENODEV;
3921
3922         r = kvm_init_subcore_bitmap();
3923         if (r)
3924                 return r;
3925
3926         /*
3927          * We need a way of accessing the XICS interrupt controller,
3928          * either directly, via paca[cpu].kvm_hstate.xics_phys, or
3929          * indirectly, via OPAL.
3930          */
3931 #ifdef CONFIG_SMP
3932         if (!get_paca()->kvm_hstate.xics_phys) {
3933                 struct device_node *np;
3934
3935                 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
3936                 if (!np) {
3937                         pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
3938                         return -ENODEV;
3939                 }
3940         }
3941 #endif
3942
3943         kvm_ops_hv.owner = THIS_MODULE;
3944         kvmppc_hv_ops = &kvm_ops_hv;
3945
3946         init_default_hcalls();
3947
3948         init_vcore_lists();
3949
3950         r = kvmppc_mmu_hv_init();
3951         if (r)
3952                 return r;
3953
3954         if (kvmppc_radix_possible())
3955                 r = kvmppc_radix_init();
3956         return r;
3957 }
3958
3959 static void kvmppc_book3s_exit_hv(void)
3960 {
3961         kvmppc_free_host_rm_ops();
3962         if (kvmppc_radix_possible())
3963                 kvmppc_radix_exit();
3964         kvmppc_hv_ops = NULL;
3965 }
3966
3967 module_init(kvmppc_book3s_init_hv);
3968 module_exit(kvmppc_book3s_exit_hv);
3969 MODULE_LICENSE("GPL");
3970 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3971 MODULE_ALIAS("devname:kvm");
3972