arch/powerpc/kvm/book3s_hv.c

   1 /*
   2  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
   3  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
   4  *
   5  * Authors:
   6  *    Paul Mackerras <paulus@au1.ibm.com>
   7  *    Alexander Graf <agraf@suse.de>
   8  *    Kevin Wolf <mail@kevin-wolf.de>
   9  *
  10  * Description: KVM functions specific to running on Book 3S
  11  * processors in hypervisor mode (specifically POWER7 and later).
  12  *
  13  * This file is derived from arch/powerpc/kvm/book3s.c,
  14  * by Alexander Graf <agraf@suse.de>.
  15  *
  16  * This program is free software; you can redistribute it and/or modify
  17  * it under the terms of the GNU General Public License, version 2, as
  18  * published by the Free Software Foundation.
  19  */
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/err.h>
  23 #include <linux/slab.h>
  24 #include <linux/preempt.h>
  25 #include <linux/sched/signal.h>
  26 #include <linux/sched/stat.h>
  27 #include <linux/delay.h>
  28 #include <linux/export.h>
  29 #include <linux/fs.h>
  30 #include <linux/anon_inodes.h>
  31 #include <linux/cpu.h>
  32 #include <linux/cpumask.h>
  33 #include <linux/spinlock.h>
  34 #include <linux/page-flags.h>
  35 #include <linux/srcu.h>
  36 #include <linux/miscdevice.h>
  37 #include <linux/debugfs.h>
  38 #include <linux/gfp.h>
  39 #include <linux/vmalloc.h>
  40 #include <linux/highmem.h>
  41 #include <linux/hugetlb.h>
  42 #include <linux/kvm_irqfd.h>
  43 #include <linux/irqbypass.h>
  44 #include <linux/module.h>
  45 #include <linux/compiler.h>
  46 #include <linux/of.h>
  47
  48 #include <asm/reg.h>
  49 #include <asm/cputable.h>
  50 #include <asm/cacheflush.h>
  51 #include <asm/tlbflush.h>
  52 #include <linux/uaccess.h>
  53 #include <asm/io.h>
  54 #include <asm/kvm_ppc.h>
  55 #include <asm/kvm_book3s.h>
  56 #include <asm/mmu_context.h>
  57 #include <asm/lppaca.h>
  58 #include <asm/processor.h>
  59 #include <asm/cputhreads.h>
  60 #include <asm/page.h>
  61 #include <asm/hvcall.h>
  62 #include <asm/switch_to.h>
  63 #include <asm/smp.h>
  64 #include <asm/dbell.h>
  65 #include <asm/hmi.h>
  66 #include <asm/pnv-pci.h>
  67 #include <asm/mmu.h>
  68 #include <asm/opal.h>
  69 #include <asm/xics.h>
  70
  71 #include "book3s.h"
  72
  73 #define CREATE_TRACE_POINTS
  74 #include "trace_hv.h"
  75
  76 /* #define EXIT_DEBUG */
  77 /* #define EXIT_DEBUG_SIMPLE */
  78 /* #define EXIT_DEBUG_INT */
  79
  80 /* Used to indicate that a guest page fault needs to be handled */
  81 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  82 /* Used to indicate that a guest passthrough interrupt needs to be handled */
  83 #define RESUME_PASSTHROUGH      (RESUME_GUEST | RESUME_FLAG_ARCH2)
  84
  85 /* Used as a "null" value for timebase values */
  86 #define TB_NIL  (~(u64)0)
  87
  88 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
  89
  90 static int dynamic_mt_modes = 6;
  91 module_param(dynamic_mt_modes, int, S_IRUGO | S_IWUSR);
  92 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
  93 static int target_smt_mode;
  94 module_param(target_smt_mode, int, S_IRUGO | S_IWUSR);
  95 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
  96
  97 #ifdef CONFIG_KVM_XICS
  98 static struct kernel_param_ops module_param_ops = {
  99         .set = param_set_int,
 100         .get = param_get_int,
 101 };
 102
 103 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass,
 104                                                         S_IRUGO | S_IWUSR);
 105 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
 106
 107 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect,
 108                                                         S_IRUGO | S_IWUSR);
 109 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
 110 #endif
 111
 112 static void kvmppc_end_cede(struct kvm_vcpu *vcpu);
 113 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
 114
 115 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
 116                 int *ip)
 117 {
 118         int i = *ip;
 119         struct kvm_vcpu *vcpu;
 120
 121         while (++i < MAX_SMT_THREADS) {
 122                 vcpu = READ_ONCE(vc->runnable_threads[i]);
 123                 if (vcpu) {
 124                         *ip = i;
 125                         return vcpu;
 126                 }
 127         }
 128         return NULL;
 129 }
 130
 131 /* Used to traverse the list of runnable threads for a given vcore */
 132 #define for_each_runnable_thread(i, vcpu, vc) \
 133         for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
 134
 135 static bool kvmppc_ipi_thread(int cpu)
 136 {
 137         unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
 138
 139         /* On POWER9 we can use msgsnd to IPI any cpu */
 140         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
 141                 msg |= get_hard_smp_processor_id(cpu);
 142                 smp_mb();
 143                 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 144                 return true;
 145         }
 146
 147         /* On POWER8 for IPIs to threads in the same core, use msgsnd */
 148         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
 149                 preempt_disable();
 150                 if (cpu_first_thread_sibling(cpu) ==
 151                     cpu_first_thread_sibling(smp_processor_id())) {
 152                         msg |= cpu_thread_in_core(cpu);
 153                         smp_mb();
 154                         __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 155                         preempt_enable();
 156                         return true;
 157                 }
 158                 preempt_enable();
 159         }
 160
 161 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
 162         if (cpu >= 0 && cpu < nr_cpu_ids) {
 163                 if (paca[cpu].kvm_hstate.xics_phys) {
 164                         xics_wake_cpu(cpu);
 165                         return true;
 166                 }
 167                 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
 168                 return true;
 169         }
 170 #endif
 171
 172         return false;
 173 }
 174
 175 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
 176 {
 177         int cpu;
 178         struct swait_queue_head *wqp;
 179
 180         wqp = kvm_arch_vcpu_wq(vcpu);
 181         if (swait_active(wqp)) {
 182                 swake_up(wqp);
 183                 ++vcpu->stat.halt_wakeup;
 184         }
 185
 186         cpu = READ_ONCE(vcpu->arch.thread_cpu);
 187         if (cpu >= 0 && kvmppc_ipi_thread(cpu))
 188                 return;
 189
 190         /* CPU points to the first thread of the core */
 191         cpu = vcpu->cpu;
 192         if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
 193                 smp_send_reschedule(cpu);
 194 }
 195
 196 /*
 197  * We use the vcpu_load/put functions to measure stolen time.
 198  * Stolen time is counted as time when either the vcpu is able to
 199  * run as part of a virtual core, but the task running the vcore
 200  * is preempted or sleeping, or when the vcpu needs something done
 201  * in the kernel by the task running the vcpu, but that task is
 202  * preempted or sleeping.  Those two things have to be counted
 203  * separately, since one of the vcpu tasks will take on the job
 204  * of running the core, and the other vcpu tasks in the vcore will
 205  * sleep waiting for it to do that, but that sleep shouldn't count
 206  * as stolen time.
 207  *
 208  * Hence we accumulate stolen time when the vcpu can run as part of
 209  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 210  * needs its task to do other things in the kernel (for example,
 211  * service a page fault) in busy_stolen.  We don't accumulate
 212  * stolen time for a vcore when it is inactive, or for a vcpu
 213  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 214  * a misnomer; it means that the vcpu task is not executing in
 215  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 216  * the kernel.  We don't have any way of dividing up that time
 217  * between time that the vcpu is genuinely stopped, time that
 218  * the task is actively working on behalf of the vcpu, and time
 219  * that the task is preempted, so we don't count any of it as
 220  * stolen.
 221  *
 222  * Updates to busy_stolen are protected by arch.tbacct_lock;
 223  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
 224  * lock.  The stolen times are measured in units of timebase ticks.
 225  * (Note that the != TB_NIL checks below are purely defensive;
 226  * they should never fail.)
 227  */
 228
 229 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
 230 {
 231         unsigned long flags;
 232
 233         spin_lock_irqsave(&vc->stoltb_lock, flags);
 234         vc->preempt_tb = mftb();
 235         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 236 }
 237
 238 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
 239 {
 240         unsigned long flags;
 241
 242         spin_lock_irqsave(&vc->stoltb_lock, flags);
 243         if (vc->preempt_tb != TB_NIL) {
 244                 vc->stolen_tb += mftb() - vc->preempt_tb;
 245                 vc->preempt_tb = TB_NIL;
 246         }
 247         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 248 }
 249
 250 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 251 {
 252         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 253         unsigned long flags;
 254
 255         /*
 256          * We can test vc->runner without taking the vcore lock,
 257          * because only this task ever sets vc->runner to this
 258          * vcpu, and once it is set to this vcpu, only this task
 259          * ever sets it to NULL.
 260          */
 261         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 262                 kvmppc_core_end_stolen(vc);
 263
 264         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 265         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 266             vcpu->arch.busy_preempt != TB_NIL) {
 267                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 268                 vcpu->arch.busy_preempt = TB_NIL;
 269         }
 270         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 271 }
 272
 273 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 274 {
 275         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 276         unsigned long flags;
 277
 278         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 279                 kvmppc_core_start_stolen(vc);
 280
 281         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 282         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 283                 vcpu->arch.busy_preempt = mftb();
 284         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 285 }
 286
 287 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
 288 {
 289         /*
 290          * Check for illegal transactional state bit combination
 291          * and if we find it, force the TS field to a safe state.
 292          */
 293         if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
 294                 msr &= ~MSR_TS_MASK;
 295         vcpu->arch.shregs.msr = msr;
 296         kvmppc_end_cede(vcpu);
 297 }
 298
 299 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 300 {
 301         vcpu->arch.pvr = pvr;
 302 }
 303
 304 /* Dummy value used in computing PCR value below */
 305 #define PCR_ARCH_300    (PCR_ARCH_207 << 1)
 306
 307 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 308 {
 309         unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
 310         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 311
 312         /* We can (emulate) our own architecture version and anything older */
 313         if (cpu_has_feature(CPU_FTR_ARCH_300))
 314                 host_pcr_bit = PCR_ARCH_300;
 315         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
 316                 host_pcr_bit = PCR_ARCH_207;
 317         else if (cpu_has_feature(CPU_FTR_ARCH_206))
 318                 host_pcr_bit = PCR_ARCH_206;
 319         else
 320                 host_pcr_bit = PCR_ARCH_205;
 321
 322         /* Determine lowest PCR bit needed to run guest in given PVR level */
 323         guest_pcr_bit = host_pcr_bit;
 324         if (arch_compat) {
 325                 switch (arch_compat) {
 326                 case PVR_ARCH_205:
 327                         guest_pcr_bit = PCR_ARCH_205;
 328                         break;
 329                 case PVR_ARCH_206:
 330                 case PVR_ARCH_206p:
 331                         guest_pcr_bit = PCR_ARCH_206;
 332                         break;
 333                 case PVR_ARCH_207:
 334                         guest_pcr_bit = PCR_ARCH_207;
 335                         break;
 336                 case PVR_ARCH_300:
 337                         guest_pcr_bit = PCR_ARCH_300;
 338                         break;
 339                 default:
 340                         return -EINVAL;
 341                 }
 342         }
 343
 344         /* Check requested PCR bits don't exceed our capabilities */
 345         if (guest_pcr_bit > host_pcr_bit)
 346                 return -EINVAL;
 347
 348         spin_lock(&vc->lock);
 349         vc->arch_compat = arch_compat;
 350         /* Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit */
 351         vc->pcr = host_pcr_bit - guest_pcr_bit;
 352         spin_unlock(&vc->lock);
 353
 354         return 0;
 355 }
 356
 357 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 358 {
 359         int r;
 360
 361         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 362         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 363                vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap);
 364         for (r = 0; r < 16; ++r)
 365                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 366                        r, kvmppc_get_gpr(vcpu, r),
 367                        r+16, kvmppc_get_gpr(vcpu, r+16));
 368         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 369                vcpu->arch.ctr, vcpu->arch.lr);
 370         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 371                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 372         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 373                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 374         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 375                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 376         pr_err("cr = %.8x  xer = %.16lx  dsisr = %.8x\n",
 377                vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr);
 378         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 379         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 380                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 381         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 382         for (r = 0; r < vcpu->arch.slb_max; ++r)
 383                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 384                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 385         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 386                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 387                vcpu->arch.last_inst);
 388 }
 389
 390 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 391 {
 392         struct kvm_vcpu *ret;
 393
 394         mutex_lock(&kvm->lock);
 395         ret = kvm_get_vcpu_by_id(kvm, id);
 396         mutex_unlock(&kvm->lock);
 397         return ret;
 398 }
 399
 400 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 401 {
 402         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 403         vpa->yield_count = cpu_to_be32(1);
 404 }
 405
 406 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 407                    unsigned long addr, unsigned long len)
 408 {
 409         /* check address is cacheline aligned */
 410         if (addr & (L1_CACHE_BYTES - 1))
 411                 return -EINVAL;
 412         spin_lock(&vcpu->arch.vpa_update_lock);
 413         if (v->next_gpa != addr || v->len != len) {
 414                 v->next_gpa = addr;
 415                 v->len = addr ? len : 0;
 416                 v->update_pending = 1;
 417         }
 418         spin_unlock(&vcpu->arch.vpa_update_lock);
 419         return 0;
 420 }
 421
 422 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 423 struct reg_vpa {
 424         u32 dummy;
 425         union {
 426                 __be16 hword;
 427                 __be32 word;
 428         } length;
 429 };
 430
 431 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 432 {
 433         if (vpap->update_pending)
 434                 return vpap->next_gpa != 0;
 435         return vpap->pinned_addr != NULL;
 436 }
 437
 438 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 439                                        unsigned long flags,
 440                                        unsigned long vcpuid, unsigned long vpa)
 441 {
 442         struct kvm *kvm = vcpu->kvm;
 443         unsigned long len, nb;
 444         void *va;
 445         struct kvm_vcpu *tvcpu;
 446         int err;
 447         int subfunc;
 448         struct kvmppc_vpa *vpap;
 449
 450         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 451         if (!tvcpu)
 452                 return H_PARAMETER;
 453
 454         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 455         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 456             subfunc == H_VPA_REG_SLB) {
 457                 /* Registering new area - address must be cache-line aligned */
 458                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 459                         return H_PARAMETER;
 460
 461                 /* convert logical addr to kernel addr and read length */
 462                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 463                 if (va == NULL)
 464                         return H_PARAMETER;
 465                 if (subfunc == H_VPA_REG_VPA)
 466                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 467                 else
 468                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 469                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 470
 471                 /* Check length */
 472                 if (len > nb || len < sizeof(struct reg_vpa))
 473                         return H_PARAMETER;
 474         } else {
 475                 vpa = 0;
 476                 len = 0;
 477         }
 478
 479         err = H_PARAMETER;
 480         vpap = NULL;
 481         spin_lock(&tvcpu->arch.vpa_update_lock);
 482
 483         switch (subfunc) {
 484         case H_VPA_REG_VPA:             /* register VPA */
 485                 if (len < sizeof(struct lppaca))
 486                         break;
 487                 vpap = &tvcpu->arch.vpa;
 488                 err = 0;
 489                 break;
 490
 491         case H_VPA_REG_DTL:             /* register DTL */
 492                 if (len < sizeof(struct dtl_entry))
 493                         break;
 494                 len -= len % sizeof(struct dtl_entry);
 495
 496                 /* Check that they have previously registered a VPA */
 497                 err = H_RESOURCE;
 498                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 499                         break;
 500
 501                 vpap = &tvcpu->arch.dtl;
 502                 err = 0;
 503                 break;
 504
 505         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 506                 /* Check that they have previously registered a VPA */
 507                 err = H_RESOURCE;
 508                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 509                         break;
 510
 511                 vpap = &tvcpu->arch.slb_shadow;
 512                 err = 0;
 513                 break;
 514
 515         case H_VPA_DEREG_VPA:           /* deregister VPA */
 516                 /* Check they don't still have a DTL or SLB buf registered */
 517                 err = H_RESOURCE;
 518                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 519                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 520                         break;
 521
 522                 vpap = &tvcpu->arch.vpa;
 523                 err = 0;
 524                 break;
 525
 526         case H_VPA_DEREG_DTL:           /* deregister DTL */
 527                 vpap = &tvcpu->arch.dtl;
 528                 err = 0;
 529                 break;
 530
 531         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 532                 vpap = &tvcpu->arch.slb_shadow;
 533                 err = 0;
 534                 break;
 535         }
 536
 537         if (vpap) {
 538                 vpap->next_gpa = vpa;
 539                 vpap->len = len;
 540                 vpap->update_pending = 1;
 541         }
 542
 543         spin_unlock(&tvcpu->arch.vpa_update_lock);
 544
 545         return err;
 546 }
 547
 548 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 549 {
 550         struct kvm *kvm = vcpu->kvm;
 551         void *va;
 552         unsigned long nb;
 553         unsigned long gpa;
 554
 555         /*
 556          * We need to pin the page pointed to by vpap->next_gpa,
 557          * but we can't call kvmppc_pin_guest_page under the lock
 558          * as it does get_user_pages() and down_read().  So we
 559          * have to drop the lock, pin the page, then get the lock
 560          * again and check that a new area didn't get registered
 561          * in the meantime.
 562          */
 563         for (;;) {
 564                 gpa = vpap->next_gpa;
 565                 spin_unlock(&vcpu->arch.vpa_update_lock);
 566                 va = NULL;
 567                 nb = 0;
 568                 if (gpa)
 569                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 570                 spin_lock(&vcpu->arch.vpa_update_lock);
 571                 if (gpa == vpap->next_gpa)
 572                         break;
 573                 /* sigh... unpin that one and try again */
 574                 if (va)
 575                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 576         }
 577
 578         vpap->update_pending = 0;
 579         if (va && nb < vpap->len) {
 580                 /*
 581                  * If it's now too short, it must be that userspace
 582                  * has changed the mappings underlying guest memory,
 583                  * so unregister the region.
 584                  */
 585                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 586                 va = NULL;
 587         }
 588         if (vpap->pinned_addr)
 589                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 590                                         vpap->dirty);
 591         vpap->gpa = gpa;
 592         vpap->pinned_addr = va;
 593         vpap->dirty = false;
 594         if (va)
 595                 vpap->pinned_end = va + vpap->len;
 596 }
 597
 598 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 599 {
 600         if (!(vcpu->arch.vpa.update_pending ||
 601               vcpu->arch.slb_shadow.update_pending ||
 602               vcpu->arch.dtl.update_pending))
 603                 return;
 604
 605         spin_lock(&vcpu->arch.vpa_update_lock);
 606         if (vcpu->arch.vpa.update_pending) {
 607                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 608                 if (vcpu->arch.vpa.pinned_addr)
 609                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 610         }
 611         if (vcpu->arch.dtl.update_pending) {
 612                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 613                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 614                 vcpu->arch.dtl_index = 0;
 615         }
 616         if (vcpu->arch.slb_shadow.update_pending)
 617                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 618         spin_unlock(&vcpu->arch.vpa_update_lock);
 619 }
 620
 621 /*
 622  * Return the accumulated stolen time for the vcore up until `now'.
 623  * The caller should hold the vcore lock.
 624  */
 625 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 626 {
 627         u64 p;
 628         unsigned long flags;
 629
 630         spin_lock_irqsave(&vc->stoltb_lock, flags);
 631         p = vc->stolen_tb;
 632         if (vc->vcore_state != VCORE_INACTIVE &&
 633             vc->preempt_tb != TB_NIL)
 634                 p += now - vc->preempt_tb;
 635         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 636         return p;
 637 }
 638
 639 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 640                                     struct kvmppc_vcore *vc)
 641 {
 642         struct dtl_entry *dt;
 643         struct lppaca *vpa;
 644         unsigned long stolen;
 645         unsigned long core_stolen;
 646         u64 now;
 647
 648         dt = vcpu->arch.dtl_ptr;
 649         vpa = vcpu->arch.vpa.pinned_addr;
 650         now = mftb();
 651         core_stolen = vcore_stolen_time(vc, now);
 652         stolen = core_stolen - vcpu->arch.stolen_logged;
 653         vcpu->arch.stolen_logged = core_stolen;
 654         spin_lock_irq(&vcpu->arch.tbacct_lock);
 655         stolen += vcpu->arch.busy_stolen;
 656         vcpu->arch.busy_stolen = 0;
 657         spin_unlock_irq(&vcpu->arch.tbacct_lock);
 658         if (!dt || !vpa)
 659                 return;
 660         memset(dt, 0, sizeof(struct dtl_entry));
 661         dt->dispatch_reason = 7;
 662         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 663         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 664         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 665         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 666         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 667         ++dt;
 668         if (dt == vcpu->arch.dtl.pinned_end)
 669                 dt = vcpu->arch.dtl.pinned_addr;
 670         vcpu->arch.dtl_ptr = dt;
 671         /* order writing *dt vs. writing vpa->dtl_idx */
 672         smp_wmb();
 673         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 674         vcpu->arch.dtl.dirty = true;
 675 }
 676
 677 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 678 {
 679         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 680                 return true;
 681         if ((!vcpu->arch.vcore->arch_compat) &&
 682             cpu_has_feature(CPU_FTR_ARCH_207S))
 683                 return true;
 684         return false;
 685 }
 686
 687 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 688                              unsigned long resource, unsigned long value1,
 689                              unsigned long value2)
 690 {
 691         switch (resource) {
 692         case H_SET_MODE_RESOURCE_SET_CIABR:
 693                 if (!kvmppc_power8_compatible(vcpu))
 694                         return H_P2;
 695                 if (value2)
 696                         return H_P4;
 697                 if (mflags)
 698                         return H_UNSUPPORTED_FLAG_START;
 699                 /* Guests can't breakpoint the hypervisor */
 700                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 701                         return H_P3;
 702                 vcpu->arch.ciabr  = value1;
 703                 return H_SUCCESS;
 704         case H_SET_MODE_RESOURCE_SET_DAWR:
 705                 if (!kvmppc_power8_compatible(vcpu))
 706                         return H_P2;
 707                 if (mflags)
 708                         return H_UNSUPPORTED_FLAG_START;
 709                 if (value2 & DABRX_HYP)
 710                         return H_P4;
 711                 vcpu->arch.dawr  = value1;
 712                 vcpu->arch.dawrx = value2;
 713                 return H_SUCCESS;
 714         default:
 715                 return H_TOO_HARD;
 716         }
 717 }
 718
 719 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 720 {
 721         struct kvmppc_vcore *vcore = target->arch.vcore;
 722
 723         /*
 724          * We expect to have been called by the real mode handler
 725          * (kvmppc_rm_h_confer()) which would have directly returned
 726          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 727          * have useful work to do and should not confer) so we don't
 728          * recheck that here.
 729          */
 730
 731         spin_lock(&vcore->lock);
 732         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 733             vcore->vcore_state != VCORE_INACTIVE &&
 734             vcore->runner)
 735                 target = vcore->runner;
 736         spin_unlock(&vcore->lock);
 737
 738         return kvm_vcpu_yield_to(target);
 739 }
 740
 741 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 742 {
 743         int yield_count = 0;
 744         struct lppaca *lppaca;
 745
 746         spin_lock(&vcpu->arch.vpa_update_lock);
 747         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 748         if (lppaca)
 749                 yield_count = be32_to_cpu(lppaca->yield_count);
 750         spin_unlock(&vcpu->arch.vpa_update_lock);
 751         return yield_count;
 752 }
 753
 754 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 755 {
 756         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 757         unsigned long target, ret = H_SUCCESS;
 758         int yield_count;
 759         struct kvm_vcpu *tvcpu;
 760         int idx, rc;
 761
 762         if (req <= MAX_HCALL_OPCODE &&
 763             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 764                 return RESUME_HOST;
 765
 766         switch (req) {
 767         case H_CEDE:
 768                 break;
 769         case H_PROD:
 770                 target = kvmppc_get_gpr(vcpu, 4);
 771                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 772                 if (!tvcpu) {
 773                         ret = H_PARAMETER;
 774                         break;
 775                 }
 776                 tvcpu->arch.prodded = 1;
 777                 smp_mb();
 778                 if (tvcpu->arch.ceded)
 779                         kvmppc_fast_vcpu_kick_hv(tvcpu);
 780                 break;
 781         case H_CONFER:
 782                 target = kvmppc_get_gpr(vcpu, 4);
 783                 if (target == -1)
 784                         break;
 785                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 786                 if (!tvcpu) {
 787                         ret = H_PARAMETER;
 788                         break;
 789                 }
 790                 yield_count = kvmppc_get_gpr(vcpu, 5);
 791                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 792                         break;
 793                 kvm_arch_vcpu_yield_to(tvcpu);
 794                 break;
 795         case H_REGISTER_VPA:
 796                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 797                                         kvmppc_get_gpr(vcpu, 5),
 798                                         kvmppc_get_gpr(vcpu, 6));
 799                 break;
 800         case H_RTAS:
 801                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 802                         return RESUME_HOST;
 803
 804                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 805                 rc = kvmppc_rtas_hcall(vcpu);
 806                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 807
 808                 if (rc == -ENOENT)
 809                         return RESUME_HOST;
 810                 else if (rc == 0)
 811                         break;
 812
 813                 /* Send the error out to userspace via KVM_RUN */
 814                 return rc;
 815         case H_LOGICAL_CI_LOAD:
 816                 ret = kvmppc_h_logical_ci_load(vcpu);
 817                 if (ret == H_TOO_HARD)
 818                         return RESUME_HOST;
 819                 break;
 820         case H_LOGICAL_CI_STORE:
 821                 ret = kvmppc_h_logical_ci_store(vcpu);
 822                 if (ret == H_TOO_HARD)
 823                         return RESUME_HOST;
 824                 break;
 825         case H_SET_MODE:
 826                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 827                                         kvmppc_get_gpr(vcpu, 5),
 828                                         kvmppc_get_gpr(vcpu, 6),
 829                                         kvmppc_get_gpr(vcpu, 7));
 830                 if (ret == H_TOO_HARD)
 831                         return RESUME_HOST;
 832                 break;
 833         case H_XIRR:
 834         case H_CPPR:
 835         case H_EOI:
 836         case H_IPI:
 837         case H_IPOLL:
 838         case H_XIRR_X:
 839                 if (kvmppc_xics_enabled(vcpu)) {
 840                         ret = kvmppc_xics_hcall(vcpu, req);
 841                         break;
 842                 }
 843                 return RESUME_HOST;
 844         case H_PUT_TCE:
 845                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
 846                                                 kvmppc_get_gpr(vcpu, 5),
 847                                                 kvmppc_get_gpr(vcpu, 6));
 848                 if (ret == H_TOO_HARD)
 849                         return RESUME_HOST;
 850                 break;
 851         case H_PUT_TCE_INDIRECT:
 852                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
 853                                                 kvmppc_get_gpr(vcpu, 5),
 854                                                 kvmppc_get_gpr(vcpu, 6),
 855                                                 kvmppc_get_gpr(vcpu, 7));
 856                 if (ret == H_TOO_HARD)
 857                         return RESUME_HOST;
 858                 break;
 859         case H_STUFF_TCE:
 860                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
 861                                                 kvmppc_get_gpr(vcpu, 5),
 862                                                 kvmppc_get_gpr(vcpu, 6),
 863                                                 kvmppc_get_gpr(vcpu, 7));
 864                 if (ret == H_TOO_HARD)
 865                         return RESUME_HOST;
 866                 break;
 867         default:
 868                 return RESUME_HOST;
 869         }
 870         kvmppc_set_gpr(vcpu, 3, ret);
 871         vcpu->arch.hcall_needed = 0;
 872         return RESUME_GUEST;
 873 }
 874
 875 static int kvmppc_hcall_impl_hv(unsigned long cmd)
 876 {
 877         switch (cmd) {
 878         case H_CEDE:
 879         case H_PROD:
 880         case H_CONFER:
 881         case H_REGISTER_VPA:
 882         case H_SET_MODE:
 883         case H_LOGICAL_CI_LOAD:
 884         case H_LOGICAL_CI_STORE:
 885 #ifdef CONFIG_KVM_XICS
 886         case H_XIRR:
 887         case H_CPPR:
 888         case H_EOI:
 889         case H_IPI:
 890         case H_IPOLL:
 891         case H_XIRR_X:
 892 #endif
 893                 return 1;
 894         }
 895
 896         /* See if it's in the real-mode table */
 897         return kvmppc_hcall_impl_hv_realmode(cmd);
 898 }
 899
 900 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
 901                                         struct kvm_vcpu *vcpu)
 902 {
 903         u32 last_inst;
 904
 905         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
 906                                         EMULATE_DONE) {
 907                 /*
 908                  * Fetch failed, so return to guest and
 909                  * try executing it again.
 910                  */
 911                 return RESUME_GUEST;
 912         }
 913
 914         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
 915                 run->exit_reason = KVM_EXIT_DEBUG;
 916                 run->debug.arch.address = kvmppc_get_pc(vcpu);
 917                 return RESUME_HOST;
 918         } else {
 919                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
 920                 return RESUME_GUEST;
 921         }
 922 }
 923
 924 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
 925                                  struct task_struct *tsk)
 926 {
 927         int r = RESUME_HOST;
 928
 929         vcpu->stat.sum_exits++;
 930
 931         /*
 932          * This can happen if an interrupt occurs in the last stages
 933          * of guest entry or the first stages of guest exit (i.e. after
 934          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
 935          * and before setting it to KVM_GUEST_MODE_HOST_HV).
 936          * That can happen due to a bug, or due to a machine check
 937          * occurring at just the wrong time.
 938          */
 939         if (vcpu->arch.shregs.msr & MSR_HV) {
 940                 printk(KERN_EMERG "KVM trap in HV mode!\n");
 941                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
 942                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
 943                         vcpu->arch.shregs.msr);
 944                 kvmppc_dump_regs(vcpu);
 945                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
 946                 run->hw.hardware_exit_reason = vcpu->arch.trap;
 947                 return RESUME_HOST;
 948         }
 949         run->exit_reason = KVM_EXIT_UNKNOWN;
 950         run->ready_for_interrupt_injection = 1;
 951         switch (vcpu->arch.trap) {
 952         /* We're good on these - the host merely wanted to get our attention */
 953         case BOOK3S_INTERRUPT_HV_DECREMENTER:
 954                 vcpu->stat.dec_exits++;
 955                 r = RESUME_GUEST;
 956                 break;
 957         case BOOK3S_INTERRUPT_EXTERNAL:
 958         case BOOK3S_INTERRUPT_H_DOORBELL:
 959         case BOOK3S_INTERRUPT_H_VIRT:
 960                 vcpu->stat.ext_intr_exits++;
 961                 r = RESUME_GUEST;
 962                 break;
 963         /* HMI is hypervisor interrupt and host has handled it. Resume guest.*/
 964         case BOOK3S_INTERRUPT_HMI:
 965         case BOOK3S_INTERRUPT_PERFMON:
 966                 r = RESUME_GUEST;
 967                 break;
 968         case BOOK3S_INTERRUPT_MACHINE_CHECK:
 969                 /*
 970                  * Deliver a machine check interrupt to the guest.
 971                  * We have to do this, even if the host has handled the
 972                  * machine check, because machine checks use SRR0/1 and
 973                  * the interrupt might have trashed guest state in them.
 974                  */
 975                 kvmppc_book3s_queue_irqprio(vcpu,
 976                                             BOOK3S_INTERRUPT_MACHINE_CHECK);
 977                 r = RESUME_GUEST;
 978                 break;
 979         case BOOK3S_INTERRUPT_PROGRAM:
 980         {
 981                 ulong flags;
 982                 /*
 983                  * Normally program interrupts are delivered directly
 984                  * to the guest by the hardware, but we can get here
 985                  * as a result of a hypervisor emulation interrupt
 986                  * (e40) getting turned into a 700 by BML RTAS.
 987                  */
 988                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
 989                 kvmppc_core_queue_program(vcpu, flags);
 990                 r = RESUME_GUEST;
 991                 break;
 992         }
 993         case BOOK3S_INTERRUPT_SYSCALL:
 994         {
 995                 /* hcall - punt to userspace */
 996                 int i;
 997
 998                 /* hypercall with MSR_PR has already been handled in rmode,
 999                  * and never reaches here.
1000                  */
1001
1002                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1003                 for (i = 0; i < 9; ++i)
1004                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1005                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1006                 vcpu->arch.hcall_needed = 1;
1007                 r = RESUME_HOST;
1008                 break;
1009         }
1010         /*
1011          * We get these next two if the guest accesses a page which it thinks
1012          * it has mapped but which is not actually present, either because
1013          * it is for an emulated I/O device or because the corresonding
1014          * host page has been paged out.  Any other HDSI/HISI interrupts
1015          * have been handled already.
1016          */
1017         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1018                 r = RESUME_PAGE_FAULT;
1019                 break;
1020         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1021                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1022                 vcpu->arch.fault_dsisr = 0;
1023                 r = RESUME_PAGE_FAULT;
1024                 break;
1025         /*
1026          * This occurs if the guest executes an illegal instruction.
1027          * If the guest debug is disabled, generate a program interrupt
1028          * to the guest. If guest debug is enabled, we need to check
1029          * whether the instruction is a software breakpoint instruction.
1030          * Accordingly return to Guest or Host.
1031          */
1032         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1033                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1034                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1035                                 swab32(vcpu->arch.emul_inst) :
1036                                 vcpu->arch.emul_inst;
1037                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1038                         r = kvmppc_emulate_debug_inst(run, vcpu);
1039                 } else {
1040                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1041                         r = RESUME_GUEST;
1042                 }
1043                 break;
1044         /*
1045          * This occurs if the guest (kernel or userspace), does something that
1046          * is prohibited by HFSCR.  We just generate a program interrupt to
1047          * the guest.
1048          */
1049         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1050                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1051                 r = RESUME_GUEST;
1052                 break;
1053         case BOOK3S_INTERRUPT_HV_RM_HARD:
1054                 r = RESUME_PASSTHROUGH;
1055                 break;
1056         default:
1057                 kvmppc_dump_regs(vcpu);
1058                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1059                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1060                         vcpu->arch.shregs.msr);
1061                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1062                 r = RESUME_HOST;
1063                 break;
1064         }
1065
1066         return r;
1067 }
1068
1069 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1070                                             struct kvm_sregs *sregs)
1071 {
1072         int i;
1073
1074         memset(sregs, 0, sizeof(struct kvm_sregs));
1075         sregs->pvr = vcpu->arch.pvr;
1076         for (i = 0; i < vcpu->arch.slb_max; i++) {
1077                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1078                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1079         }
1080
1081         return 0;
1082 }
1083
1084 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1085                                             struct kvm_sregs *sregs)
1086 {
1087         int i, j;
1088
1089         /* Only accept the same PVR as the host's, since we can't spoof it */
1090         if (sregs->pvr != vcpu->arch.pvr)
1091                 return -EINVAL;
1092
1093         j = 0;
1094         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1095                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1096                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1097                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1098                         ++j;
1099                 }
1100         }
1101         vcpu->arch.slb_max = j;
1102
1103         return 0;
1104 }
1105
1106 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1107                 bool preserve_top32)
1108 {
1109         struct kvm *kvm = vcpu->kvm;
1110         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1111         u64 mask;
1112
1113         mutex_lock(&kvm->lock);
1114         spin_lock(&vc->lock);
1115         /*
1116          * If ILE (interrupt little-endian) has changed, update the
1117          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1118          */
1119         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1120                 struct kvm_vcpu *vcpu;
1121                 int i;
1122
1123                 kvm_for_each_vcpu(i, vcpu, kvm) {
1124                         if (vcpu->arch.vcore != vc)
1125                                 continue;
1126                         if (new_lpcr & LPCR_ILE)
1127                                 vcpu->arch.intr_msr |= MSR_LE;
1128                         else
1129                                 vcpu->arch.intr_msr &= ~MSR_LE;
1130                 }
1131         }
1132
1133         /*
1134          * Userspace can only modify DPFD (default prefetch depth),
1135          * ILE (interrupt little-endian) and TC (translation control).
1136          * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1137          */
1138         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1139         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1140                 mask |= LPCR_AIL;
1141
1142         /* Broken 32-bit version of LPCR must not clear top bits */
1143         if (preserve_top32)
1144                 mask &= 0xFFFFFFFF;
1145         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1146         spin_unlock(&vc->lock);
1147         mutex_unlock(&kvm->lock);
1148 }
1149
1150 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1151                                  union kvmppc_one_reg *val)
1152 {
1153         int r = 0;
1154         long int i;
1155
1156         switch (id) {
1157         case KVM_REG_PPC_DEBUG_INST:
1158                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1159                 break;
1160         case KVM_REG_PPC_HIOR:
1161                 *val = get_reg_val(id, 0);
1162                 break;
1163         case KVM_REG_PPC_DABR:
1164                 *val = get_reg_val(id, vcpu->arch.dabr);
1165                 break;
1166         case KVM_REG_PPC_DABRX:
1167                 *val = get_reg_val(id, vcpu->arch.dabrx);
1168                 break;
1169         case KVM_REG_PPC_DSCR:
1170                 *val = get_reg_val(id, vcpu->arch.dscr);
1171                 break;
1172         case KVM_REG_PPC_PURR:
1173                 *val = get_reg_val(id, vcpu->arch.purr);
1174                 break;
1175         case KVM_REG_PPC_SPURR:
1176                 *val = get_reg_val(id, vcpu->arch.spurr);
1177                 break;
1178         case KVM_REG_PPC_AMR:
1179                 *val = get_reg_val(id, vcpu->arch.amr);
1180                 break;
1181         case KVM_REG_PPC_UAMOR:
1182                 *val = get_reg_val(id, vcpu->arch.uamor);
1183                 break;
1184         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1185                 i = id - KVM_REG_PPC_MMCR0;
1186                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1187                 break;
1188         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1189                 i = id - KVM_REG_PPC_PMC1;
1190                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1191                 break;
1192         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1193                 i = id - KVM_REG_PPC_SPMC1;
1194                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1195                 break;
1196         case KVM_REG_PPC_SIAR:
1197                 *val = get_reg_val(id, vcpu->arch.siar);
1198                 break;
1199         case KVM_REG_PPC_SDAR:
1200                 *val = get_reg_val(id, vcpu->arch.sdar);
1201                 break;
1202         case KVM_REG_PPC_SIER:
1203                 *val = get_reg_val(id, vcpu->arch.sier);
1204                 break;
1205         case KVM_REG_PPC_IAMR:
1206                 *val = get_reg_val(id, vcpu->arch.iamr);
1207                 break;
1208         case KVM_REG_PPC_PSPB:
1209                 *val = get_reg_val(id, vcpu->arch.pspb);
1210                 break;
1211         case KVM_REG_PPC_DPDES:
1212                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes);
1213                 break;
1214         case KVM_REG_PPC_VTB:
1215                 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1216                 break;
1217         case KVM_REG_PPC_DAWR:
1218                 *val = get_reg_val(id, vcpu->arch.dawr);
1219                 break;
1220         case KVM_REG_PPC_DAWRX:
1221                 *val = get_reg_val(id, vcpu->arch.dawrx);
1222                 break;
1223         case KVM_REG_PPC_CIABR:
1224                 *val = get_reg_val(id, vcpu->arch.ciabr);
1225                 break;
1226         case KVM_REG_PPC_CSIGR:
1227                 *val = get_reg_val(id, vcpu->arch.csigr);
1228                 break;
1229         case KVM_REG_PPC_TACR:
1230                 *val = get_reg_val(id, vcpu->arch.tacr);
1231                 break;
1232         case KVM_REG_PPC_TCSCR:
1233                 *val = get_reg_val(id, vcpu->arch.tcscr);
1234                 break;
1235         case KVM_REG_PPC_PID:
1236                 *val = get_reg_val(id, vcpu->arch.pid);
1237                 break;
1238         case KVM_REG_PPC_ACOP:
1239                 *val = get_reg_val(id, vcpu->arch.acop);
1240                 break;
1241         case KVM_REG_PPC_WORT:
1242                 *val = get_reg_val(id, vcpu->arch.wort);
1243                 break;
1244         case KVM_REG_PPC_TIDR:
1245                 *val = get_reg_val(id, vcpu->arch.tid);
1246                 break;
1247         case KVM_REG_PPC_PSSCR:
1248                 *val = get_reg_val(id, vcpu->arch.psscr);
1249                 break;
1250         case KVM_REG_PPC_VPA_ADDR:
1251                 spin_lock(&vcpu->arch.vpa_update_lock);
1252                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1253                 spin_unlock(&vcpu->arch.vpa_update_lock);
1254                 break;
1255         case KVM_REG_PPC_VPA_SLB:
1256                 spin_lock(&vcpu->arch.vpa_update_lock);
1257                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1258                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1259                 spin_unlock(&vcpu->arch.vpa_update_lock);
1260                 break;
1261         case KVM_REG_PPC_VPA_DTL:
1262                 spin_lock(&vcpu->arch.vpa_update_lock);
1263                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1264                 val->vpaval.length = vcpu->arch.dtl.len;
1265                 spin_unlock(&vcpu->arch.vpa_update_lock);
1266                 break;
1267         case KVM_REG_PPC_TB_OFFSET:
1268                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1269                 break;
1270         case KVM_REG_PPC_LPCR:
1271         case KVM_REG_PPC_LPCR_64:
1272                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1273                 break;
1274         case KVM_REG_PPC_PPR:
1275                 *val = get_reg_val(id, vcpu->arch.ppr);
1276                 break;
1277 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1278         case KVM_REG_PPC_TFHAR:
1279                 *val = get_reg_val(id, vcpu->arch.tfhar);
1280                 break;
1281         case KVM_REG_PPC_TFIAR:
1282                 *val = get_reg_val(id, vcpu->arch.tfiar);
1283                 break;
1284         case KVM_REG_PPC_TEXASR:
1285                 *val = get_reg_val(id, vcpu->arch.texasr);
1286                 break;
1287         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1288                 i = id - KVM_REG_PPC_TM_GPR0;
1289                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1290                 break;
1291         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1292         {
1293                 int j;
1294                 i = id - KVM_REG_PPC_TM_VSR0;
1295                 if (i < 32)
1296                         for (j = 0; j < TS_FPRWIDTH; j++)
1297                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1298                 else {
1299                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1300                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1301                         else
1302                                 r = -ENXIO;
1303                 }
1304                 break;
1305         }
1306         case KVM_REG_PPC_TM_CR:
1307                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1308                 break;
1309         case KVM_REG_PPC_TM_XER:
1310                 *val = get_reg_val(id, vcpu->arch.xer_tm);
1311                 break;
1312         case KVM_REG_PPC_TM_LR:
1313                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1314                 break;
1315         case KVM_REG_PPC_TM_CTR:
1316                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1317                 break;
1318         case KVM_REG_PPC_TM_FPSCR:
1319                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1320                 break;
1321         case KVM_REG_PPC_TM_AMR:
1322                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1323                 break;
1324         case KVM_REG_PPC_TM_PPR:
1325                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1326                 break;
1327         case KVM_REG_PPC_TM_VRSAVE:
1328                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1329                 break;
1330         case KVM_REG_PPC_TM_VSCR:
1331                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1332                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1333                 else
1334                         r = -ENXIO;
1335                 break;
1336         case KVM_REG_PPC_TM_DSCR:
1337                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1338                 break;
1339         case KVM_REG_PPC_TM_TAR:
1340                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1341                 break;
1342 #endif
1343         case KVM_REG_PPC_ARCH_COMPAT:
1344                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1345                 break;
1346         default:
1347                 r = -EINVAL;
1348                 break;
1349         }
1350
1351         return r;
1352 }
1353
1354 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1355                                  union kvmppc_one_reg *val)
1356 {
1357         int r = 0;
1358         long int i;
1359         unsigned long addr, len;
1360
1361         switch (id) {
1362         case KVM_REG_PPC_HIOR:
1363                 /* Only allow this to be set to zero */
1364                 if (set_reg_val(id, *val))
1365                         r = -EINVAL;
1366                 break;
1367         case KVM_REG_PPC_DABR:
1368                 vcpu->arch.dabr = set_reg_val(id, *val);
1369                 break;
1370         case KVM_REG_PPC_DABRX:
1371                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1372                 break;
1373         case KVM_REG_PPC_DSCR:
1374                 vcpu->arch.dscr = set_reg_val(id, *val);
1375                 break;
1376         case KVM_REG_PPC_PURR:
1377                 vcpu->arch.purr = set_reg_val(id, *val);
1378                 break;
1379         case KVM_REG_PPC_SPURR:
1380                 vcpu->arch.spurr = set_reg_val(id, *val);
1381                 break;
1382         case KVM_REG_PPC_AMR:
1383                 vcpu->arch.amr = set_reg_val(id, *val);
1384                 break;
1385         case KVM_REG_PPC_UAMOR:
1386                 vcpu->arch.uamor = set_reg_val(id, *val);
1387                 break;
1388         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1389                 i = id - KVM_REG_PPC_MMCR0;
1390                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1391                 break;
1392         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1393                 i = id - KVM_REG_PPC_PMC1;
1394                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1395                 break;
1396         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1397                 i = id - KVM_REG_PPC_SPMC1;
1398                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1399                 break;
1400         case KVM_REG_PPC_SIAR:
1401                 vcpu->arch.siar = set_reg_val(id, *val);
1402                 break;
1403         case KVM_REG_PPC_SDAR:
1404                 vcpu->arch.sdar = set_reg_val(id, *val);
1405                 break;
1406         case KVM_REG_PPC_SIER:
1407                 vcpu->arch.sier = set_reg_val(id, *val);
1408                 break;
1409         case KVM_REG_PPC_IAMR:
1410                 vcpu->arch.iamr = set_reg_val(id, *val);
1411                 break;
1412         case KVM_REG_PPC_PSPB:
1413                 vcpu->arch.pspb = set_reg_val(id, *val);
1414                 break;
1415         case KVM_REG_PPC_DPDES:
1416                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1417                 break;
1418         case KVM_REG_PPC_VTB:
1419                 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1420                 break;
1421         case KVM_REG_PPC_DAWR:
1422                 vcpu->arch.dawr = set_reg_val(id, *val);
1423                 break;
1424         case KVM_REG_PPC_DAWRX:
1425                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1426                 break;
1427         case KVM_REG_PPC_CIABR:
1428                 vcpu->arch.ciabr = set_reg_val(id, *val);
1429                 /* Don't allow setting breakpoints in hypervisor code */
1430                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1431                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1432                 break;
1433         case KVM_REG_PPC_CSIGR:
1434                 vcpu->arch.csigr = set_reg_val(id, *val);
1435                 break;
1436         case KVM_REG_PPC_TACR:
1437                 vcpu->arch.tacr = set_reg_val(id, *val);
1438                 break;
1439         case KVM_REG_PPC_TCSCR:
1440                 vcpu->arch.tcscr = set_reg_val(id, *val);
1441                 break;
1442         case KVM_REG_PPC_PID:
1443                 vcpu->arch.pid = set_reg_val(id, *val);
1444                 break;
1445         case KVM_REG_PPC_ACOP:
1446                 vcpu->arch.acop = set_reg_val(id, *val);
1447                 break;
1448         case KVM_REG_PPC_WORT:
1449                 vcpu->arch.wort = set_reg_val(id, *val);
1450                 break;
1451         case KVM_REG_PPC_TIDR:
1452                 vcpu->arch.tid = set_reg_val(id, *val);
1453                 break;
1454         case KVM_REG_PPC_PSSCR:
1455                 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1456                 break;
1457         case KVM_REG_PPC_VPA_ADDR:
1458                 addr = set_reg_val(id, *val);
1459                 r = -EINVAL;
1460                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1461                               vcpu->arch.dtl.next_gpa))
1462                         break;
1463                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1464                 break;
1465         case KVM_REG_PPC_VPA_SLB:
1466                 addr = val->vpaval.addr;
1467                 len = val->vpaval.length;
1468                 r = -EINVAL;
1469                 if (addr && !vcpu->arch.vpa.next_gpa)
1470                         break;
1471                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1472                 break;
1473         case KVM_REG_PPC_VPA_DTL:
1474                 addr = val->vpaval.addr;
1475                 len = val->vpaval.length;
1476                 r = -EINVAL;
1477                 if (addr && (len < sizeof(struct dtl_entry) ||
1478                              !vcpu->arch.vpa.next_gpa))
1479                         break;
1480                 len -= len % sizeof(struct dtl_entry);
1481                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1482                 break;
1483         case KVM_REG_PPC_TB_OFFSET:
1484                 /* round up to multiple of 2^24 */
1485                 vcpu->arch.vcore->tb_offset =
1486                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1487                 break;
1488         case KVM_REG_PPC_LPCR:
1489                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1490                 break;
1491         case KVM_REG_PPC_LPCR_64:
1492                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1493                 break;
1494         case KVM_REG_PPC_PPR:
1495                 vcpu->arch.ppr = set_reg_val(id, *val);
1496                 break;
1497 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1498         case KVM_REG_PPC_TFHAR:
1499                 vcpu->arch.tfhar = set_reg_val(id, *val);
1500                 break;
1501         case KVM_REG_PPC_TFIAR:
1502                 vcpu->arch.tfiar = set_reg_val(id, *val);
1503                 break;
1504         case KVM_REG_PPC_TEXASR:
1505                 vcpu->arch.texasr = set_reg_val(id, *val);
1506                 break;
1507         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1508                 i = id - KVM_REG_PPC_TM_GPR0;
1509                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
1510                 break;
1511         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1512         {
1513                 int j;
1514                 i = id - KVM_REG_PPC_TM_VSR0;
1515                 if (i < 32)
1516                         for (j = 0; j < TS_FPRWIDTH; j++)
1517                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
1518                 else
1519                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1520                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
1521                         else
1522                                 r = -ENXIO;
1523                 break;
1524         }
1525         case KVM_REG_PPC_TM_CR:
1526                 vcpu->arch.cr_tm = set_reg_val(id, *val);
1527                 break;
1528         case KVM_REG_PPC_TM_XER:
1529                 vcpu->arch.xer_tm = set_reg_val(id, *val);
1530                 break;
1531         case KVM_REG_PPC_TM_LR:
1532                 vcpu->arch.lr_tm = set_reg_val(id, *val);
1533                 break;
1534         case KVM_REG_PPC_TM_CTR:
1535                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
1536                 break;
1537         case KVM_REG_PPC_TM_FPSCR:
1538                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
1539                 break;
1540         case KVM_REG_PPC_TM_AMR:
1541                 vcpu->arch.amr_tm = set_reg_val(id, *val);
1542                 break;
1543         case KVM_REG_PPC_TM_PPR:
1544                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
1545                 break;
1546         case KVM_REG_PPC_TM_VRSAVE:
1547                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
1548                 break;
1549         case KVM_REG_PPC_TM_VSCR:
1550                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1551                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
1552                 else
1553                         r = - ENXIO;
1554                 break;
1555         case KVM_REG_PPC_TM_DSCR:
1556                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
1557                 break;
1558         case KVM_REG_PPC_TM_TAR:
1559                 vcpu->arch.tar_tm = set_reg_val(id, *val);
1560                 break;
1561 #endif
1562         case KVM_REG_PPC_ARCH_COMPAT:
1563                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
1564                 break;
1565         default:
1566                 r = -EINVAL;
1567                 break;
1568         }
1569
1570         return r;
1571 }
1572
1573 /*
1574  * On POWER9, threads are independent and can be in different partitions.
1575  * Therefore we consider each thread to be a subcore.
1576  * There is a restriction that all threads have to be in the same
1577  * MMU mode (radix or HPT), unfortunately, but since we only support
1578  * HPT guests on a HPT host so far, that isn't an impediment yet.
1579  */
1580 static int threads_per_vcore(void)
1581 {
1582         if (cpu_has_feature(CPU_FTR_ARCH_300))
1583                 return 1;
1584         return threads_per_subcore;
1585 }
1586
1587 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
1588 {
1589         struct kvmppc_vcore *vcore;
1590
1591         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
1592
1593         if (vcore == NULL)
1594                 return NULL;
1595
1596         spin_lock_init(&vcore->lock);
1597         spin_lock_init(&vcore->stoltb_lock);
1598         init_swait_queue_head(&vcore->wq);
1599         vcore->preempt_tb = TB_NIL;
1600         vcore->lpcr = kvm->arch.lpcr;
1601         vcore->first_vcpuid = core * threads_per_vcore();
1602         vcore->kvm = kvm;
1603         INIT_LIST_HEAD(&vcore->preempt_list);
1604
1605         return vcore;
1606 }
1607
1608 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1609 static struct debugfs_timings_element {
1610         const char *name;
1611         size_t offset;
1612 } timings[] = {
1613         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
1614         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
1615         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
1616         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
1617         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
1618 };
1619
1620 #define N_TIMINGS       (sizeof(timings) / sizeof(timings[0]))
1621
1622 struct debugfs_timings_state {
1623         struct kvm_vcpu *vcpu;
1624         unsigned int    buflen;
1625         char            buf[N_TIMINGS * 100];
1626 };
1627
1628 static int debugfs_timings_open(struct inode *inode, struct file *file)
1629 {
1630         struct kvm_vcpu *vcpu = inode->i_private;
1631         struct debugfs_timings_state *p;
1632
1633         p = kzalloc(sizeof(*p), GFP_KERNEL);
1634         if (!p)
1635                 return -ENOMEM;
1636
1637         kvm_get_kvm(vcpu->kvm);
1638         p->vcpu = vcpu;
1639         file->private_data = p;
1640
1641         return nonseekable_open(inode, file);
1642 }
1643
1644 static int debugfs_timings_release(struct inode *inode, struct file *file)
1645 {
1646         struct debugfs_timings_state *p = file->private_data;
1647
1648         kvm_put_kvm(p->vcpu->kvm);
1649         kfree(p);
1650         return 0;
1651 }
1652
1653 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
1654                                     size_t len, loff_t *ppos)
1655 {
1656         struct debugfs_timings_state *p = file->private_data;
1657         struct kvm_vcpu *vcpu = p->vcpu;
1658         char *s, *buf_end;
1659         struct kvmhv_tb_accumulator tb;
1660         u64 count;
1661         loff_t pos;
1662         ssize_t n;
1663         int i, loops;
1664         bool ok;
1665
1666         if (!p->buflen) {
1667                 s = p->buf;
1668                 buf_end = s + sizeof(p->buf);
1669                 for (i = 0; i < N_TIMINGS; ++i) {
1670                         struct kvmhv_tb_accumulator *acc;
1671
1672                         acc = (struct kvmhv_tb_accumulator *)
1673                                 ((unsigned long)vcpu + timings[i].offset);
1674                         ok = false;
1675                         for (loops = 0; loops < 1000; ++loops) {
1676                                 count = acc->seqcount;
1677                                 if (!(count & 1)) {
1678                                         smp_rmb();
1679                                         tb = *acc;
1680                                         smp_rmb();
1681                                         if (count == acc->seqcount) {
1682                                                 ok = true;
1683                                                 break;
1684                                         }
1685                                 }
1686                                 udelay(1);
1687                         }
1688                         if (!ok)
1689                                 snprintf(s, buf_end - s, "%s: stuck\n",
1690                                         timings[i].name);
1691                         else
1692                                 snprintf(s, buf_end - s,
1693                                         "%s: %llu %llu %llu %llu\n",
1694                                         timings[i].name, count / 2,
1695                                         tb_to_ns(tb.tb_total),
1696                                         tb_to_ns(tb.tb_min),
1697                                         tb_to_ns(tb.tb_max));
1698                         s += strlen(s);
1699                 }
1700                 p->buflen = s - p->buf;
1701         }
1702
1703         pos = *ppos;
1704         if (pos >= p->buflen)
1705                 return 0;
1706         if (len > p->buflen - pos)
1707                 len = p->buflen - pos;
1708         n = copy_to_user(buf, p->buf + pos, len);
1709         if (n) {
1710                 if (n == len)
1711                         return -EFAULT;
1712                 len -= n;
1713         }
1714         *ppos = pos + len;
1715         return len;
1716 }
1717
1718 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
1719                                      size_t len, loff_t *ppos)
1720 {
1721         return -EACCES;
1722 }
1723
1724 static const struct file_operations debugfs_timings_ops = {
1725         .owner   = THIS_MODULE,
1726         .open    = debugfs_timings_open,
1727         .release = debugfs_timings_release,
1728         .read    = debugfs_timings_read,
1729         .write   = debugfs_timings_write,
1730         .llseek  = generic_file_llseek,
1731 };
1732
1733 /* Create a debugfs directory for the vcpu */
1734 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1735 {
1736         char buf[16];
1737         struct kvm *kvm = vcpu->kvm;
1738
1739         snprintf(buf, sizeof(buf), "vcpu%u", id);
1740         if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
1741                 return;
1742         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
1743         if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
1744                 return;
1745         vcpu->arch.debugfs_timings =
1746                 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
1747                                     vcpu, &debugfs_timings_ops);
1748 }
1749
1750 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1751 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
1752 {
1753 }
1754 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
1755
1756 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
1757                                                    unsigned int id)
1758 {
1759         struct kvm_vcpu *vcpu;
1760         int err = -EINVAL;
1761         int core;
1762         struct kvmppc_vcore *vcore;
1763
1764         core = id / threads_per_vcore();
1765         if (core >= KVM_MAX_VCORES)
1766                 goto out;
1767
1768         err = -ENOMEM;
1769         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1770         if (!vcpu)
1771                 goto out;
1772
1773         err = kvm_vcpu_init(vcpu, kvm, id);
1774         if (err)
1775                 goto free_vcpu;
1776
1777         vcpu->arch.shared = &vcpu->arch.shregs;
1778 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1779         /*
1780          * The shared struct is never shared on HV,
1781          * so we can always use host endianness
1782          */
1783 #ifdef __BIG_ENDIAN__
1784         vcpu->arch.shared_big_endian = true;
1785 #else
1786         vcpu->arch.shared_big_endian = false;
1787 #endif
1788 #endif
1789         vcpu->arch.mmcr[0] = MMCR0_FC;
1790         vcpu->arch.ctrl = CTRL_RUNLATCH;
1791         /* default to host PVR, since we can't spoof it */
1792         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
1793         spin_lock_init(&vcpu->arch.vpa_update_lock);
1794         spin_lock_init(&vcpu->arch.tbacct_lock);
1795         vcpu->arch.busy_preempt = TB_NIL;
1796         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
1797
1798         kvmppc_mmu_book3s_hv_init(vcpu);
1799
1800         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
1801
1802         init_waitqueue_head(&vcpu->arch.cpu_run);
1803
1804         mutex_lock(&kvm->lock);
1805         vcore = kvm->arch.vcores[core];
1806         if (!vcore) {
1807                 vcore = kvmppc_vcore_create(kvm, core);
1808                 kvm->arch.vcores[core] = vcore;
1809                 kvm->arch.online_vcores++;
1810         }
1811         mutex_unlock(&kvm->lock);
1812
1813         if (!vcore)
1814                 goto free_vcpu;
1815
1816         spin_lock(&vcore->lock);
1817         ++vcore->num_threads;
1818         spin_unlock(&vcore->lock);
1819         vcpu->arch.vcore = vcore;
1820         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
1821         vcpu->arch.thread_cpu = -1;
1822         vcpu->arch.prev_cpu = -1;
1823
1824         vcpu->arch.cpu_type = KVM_CPU_3S_64;
1825         kvmppc_sanity_check(vcpu);
1826
1827         debugfs_vcpu_init(vcpu, id);
1828
1829         return vcpu;
1830
1831 free_vcpu:
1832         kmem_cache_free(kvm_vcpu_cache, vcpu);
1833 out:
1834         return ERR_PTR(err);
1835 }
1836
1837 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
1838 {
1839         if (vpa->pinned_addr)
1840                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
1841                                         vpa->dirty);
1842 }
1843
1844 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
1845 {
1846         spin_lock(&vcpu->arch.vpa_update_lock);
1847         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
1848         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
1849         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
1850         spin_unlock(&vcpu->arch.vpa_update_lock);
1851         kvm_vcpu_uninit(vcpu);
1852         kmem_cache_free(kvm_vcpu_cache, vcpu);
1853 }
1854
1855 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
1856 {
1857         /* Indicate we want to get back into the guest */
1858         return 1;
1859 }
1860
1861 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
1862 {
1863         unsigned long dec_nsec, now;
1864
1865         now = get_tb();
1866         if (now > vcpu->arch.dec_expires) {
1867                 /* decrementer has already gone negative */
1868                 kvmppc_core_queue_dec(vcpu);
1869                 kvmppc_core_prepare_to_enter(vcpu);
1870                 return;
1871         }
1872         dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC
1873                    / tb_ticks_per_sec;
1874         hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
1875         vcpu->arch.timer_running = 1;
1876 }
1877
1878 static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
1879 {
1880         vcpu->arch.ceded = 0;
1881         if (vcpu->arch.timer_running) {
1882                 hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1883                 vcpu->arch.timer_running = 0;
1884         }
1885 }
1886
1887 extern void __kvmppc_vcore_entry(void);
1888
1889 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
1890                                    struct kvm_vcpu *vcpu)
1891 {
1892         u64 now;
1893
1894         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
1895                 return;
1896         spin_lock_irq(&vcpu->arch.tbacct_lock);
1897         now = mftb();
1898         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
1899                 vcpu->arch.stolen_logged;
1900         vcpu->arch.busy_preempt = now;
1901         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
1902         spin_unlock_irq(&vcpu->arch.tbacct_lock);
1903         --vc->n_runnable;
1904         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
1905 }
1906
1907 static int kvmppc_grab_hwthread(int cpu)
1908 {
1909         struct paca_struct *tpaca;
1910         long timeout = 10000;
1911
1912         tpaca = &paca[cpu];
1913
1914         /* Ensure the thread won't go into the kernel if it wakes */
1915         tpaca->kvm_hstate.kvm_vcpu = NULL;
1916         tpaca->kvm_hstate.kvm_vcore = NULL;
1917         tpaca->kvm_hstate.napping = 0;
1918         smp_wmb();
1919         tpaca->kvm_hstate.hwthread_req = 1;
1920
1921         /*
1922          * If the thread is already executing in the kernel (e.g. handling
1923          * a stray interrupt), wait for it to get back to nap mode.
1924          * The smp_mb() is to ensure that our setting of hwthread_req
1925          * is visible before we look at hwthread_state, so if this
1926          * races with the code at system_reset_pSeries and the thread
1927          * misses our setting of hwthread_req, we are sure to see its
1928          * setting of hwthread_state, and vice versa.
1929          */
1930         smp_mb();
1931         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
1932                 if (--timeout <= 0) {
1933                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
1934                         return -EBUSY;
1935                 }
1936                 udelay(1);
1937         }
1938         return 0;
1939 }
1940
1941 static void kvmppc_release_hwthread(int cpu)
1942 {
1943         struct paca_struct *tpaca;
1944
1945         tpaca = &paca[cpu];
1946         tpaca->kvm_hstate.hwthread_req = 0;
1947         tpaca->kvm_hstate.kvm_vcpu = NULL;
1948         tpaca->kvm_hstate.kvm_vcore = NULL;
1949         tpaca->kvm_hstate.kvm_split_mode = NULL;
1950 }
1951
1952 static void do_nothing(void *x)
1953 {
1954 }
1955
1956 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
1957 {
1958         int i;
1959
1960         cpu = cpu_first_thread_sibling(cpu);
1961         cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
1962         /*
1963          * Make sure setting of bit in need_tlb_flush precedes
1964          * testing of cpu_in_guest bits.  The matching barrier on
1965          * the other side is the first smp_mb() in kvmppc_run_core().
1966          */
1967         smp_mb();
1968         for (i = 0; i < threads_per_core; ++i)
1969                 if (cpumask_test_cpu(cpu + i, &kvm->arch.cpu_in_guest))
1970                         smp_call_function_single(cpu + i, do_nothing, NULL, 1);
1971 }
1972
1973 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
1974 {
1975         int cpu;
1976         struct paca_struct *tpaca;
1977         struct kvmppc_vcore *mvc = vc->master_vcore;
1978         struct kvm *kvm = vc->kvm;
1979
1980         cpu = vc->pcpu;
1981         if (vcpu) {
1982                 if (vcpu->arch.timer_running) {
1983                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
1984                         vcpu->arch.timer_running = 0;
1985                 }
1986                 cpu += vcpu->arch.ptid;
1987                 vcpu->cpu = mvc->pcpu;
1988                 vcpu->arch.thread_cpu = cpu;
1989
1990                 /*
1991                  * With radix, the guest can do TLB invalidations itself,
1992                  * and it could choose to use the local form (tlbiel) if
1993                  * it is invalidating a translation that has only ever been
1994                  * used on one vcpu.  However, that doesn't mean it has
1995                  * only ever been used on one physical cpu, since vcpus
1996                  * can move around between pcpus.  To cope with this, when
1997                  * a vcpu moves from one pcpu to another, we need to tell
1998                  * any vcpus running on the same core as this vcpu previously
1999                  * ran to flush the TLB.  The TLB is shared between threads,
2000                  * so we use a single bit in .need_tlb_flush for all 4 threads.
2001                  */
2002                 if (kvm_is_radix(kvm) && vcpu->arch.prev_cpu != cpu) {
2003                         if (vcpu->arch.prev_cpu >= 0 &&
2004                             cpu_first_thread_sibling(vcpu->arch.prev_cpu) !=
2005                             cpu_first_thread_sibling(cpu))
2006                                 radix_flush_cpu(kvm, vcpu->arch.prev_cpu, vcpu);
2007                         vcpu->arch.prev_cpu = cpu;
2008                 }
2009                 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2010         }
2011         tpaca = &paca[cpu];
2012         tpaca->kvm_hstate.kvm_vcpu = vcpu;
2013         tpaca->kvm_hstate.ptid = cpu - mvc->pcpu;
2014         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2015         smp_wmb();
2016         tpaca->kvm_hstate.kvm_vcore = mvc;
2017         if (cpu != smp_processor_id())
2018                 kvmppc_ipi_thread(cpu);
2019 }
2020
2021 static void kvmppc_wait_for_nap(void)
2022 {
2023         int cpu = smp_processor_id();
2024         int i, loops;
2025         int n_threads = threads_per_vcore();
2026
2027         if (n_threads <= 1)
2028                 return;
2029         for (loops = 0; loops < 1000000; ++loops) {
2030                 /*
2031                  * Check if all threads are finished.
2032                  * We set the vcore pointer when starting a thread
2033                  * and the thread clears it when finished, so we look
2034                  * for any threads that still have a non-NULL vcore ptr.
2035                  */
2036                 for (i = 1; i < n_threads; ++i)
2037                         if (paca[cpu + i].kvm_hstate.kvm_vcore)
2038                                 break;
2039                 if (i == n_threads) {
2040                         HMT_medium();
2041                         return;
2042                 }
2043                 HMT_low();
2044         }
2045         HMT_medium();
2046         for (i = 1; i < n_threads; ++i)
2047                 if (paca[cpu + i].kvm_hstate.kvm_vcore)
2048                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2049 }
2050
2051 /*
2052  * Check that we are on thread 0 and that any other threads in
2053  * this core are off-line.  Then grab the threads so they can't
2054  * enter the kernel.
2055  */
2056 static int on_primary_thread(void)
2057 {
2058         int cpu = smp_processor_id();
2059         int thr;
2060
2061         /* Are we on a primary subcore? */
2062         if (cpu_thread_in_subcore(cpu))
2063                 return 0;
2064
2065         thr = 0;
2066         while (++thr < threads_per_subcore)
2067                 if (cpu_online(cpu + thr))
2068                         return 0;
2069
2070         /* Grab all hw threads so they can't go into the kernel */
2071         for (thr = 1; thr < threads_per_subcore; ++thr) {
2072                 if (kvmppc_grab_hwthread(cpu + thr)) {
2073                         /* Couldn't grab one; let the others go */
2074                         do {
2075                                 kvmppc_release_hwthread(cpu + thr);
2076                         } while (--thr > 0);
2077                         return 0;
2078                 }
2079         }
2080         return 1;
2081 }
2082
2083 /*
2084  * A list of virtual cores for each physical CPU.
2085  * These are vcores that could run but their runner VCPU tasks are
2086  * (or may be) preempted.
2087  */
2088 struct preempted_vcore_list {
2089         struct list_head        list;
2090         spinlock_t              lock;
2091 };
2092
2093 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2094
2095 static void init_vcore_lists(void)
2096 {
2097         int cpu;
2098
2099         for_each_possible_cpu(cpu) {
2100                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2101                 spin_lock_init(&lp->lock);
2102                 INIT_LIST_HEAD(&lp->list);
2103         }
2104 }
2105
2106 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2107 {
2108         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2109
2110         vc->vcore_state = VCORE_PREEMPT;
2111         vc->pcpu = smp_processor_id();
2112         if (vc->num_threads < threads_per_vcore()) {
2113                 spin_lock(&lp->lock);
2114                 list_add_tail(&vc->preempt_list, &lp->list);
2115                 spin_unlock(&lp->lock);
2116         }
2117
2118         /* Start accumulating stolen time */
2119         kvmppc_core_start_stolen(vc);
2120 }
2121
2122 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2123 {
2124         struct preempted_vcore_list *lp;
2125
2126         kvmppc_core_end_stolen(vc);
2127         if (!list_empty(&vc->preempt_list)) {
2128                 lp = &per_cpu(preempted_vcores, vc->pcpu);
2129                 spin_lock(&lp->lock);
2130                 list_del_init(&vc->preempt_list);
2131                 spin_unlock(&lp->lock);
2132         }
2133         vc->vcore_state = VCORE_INACTIVE;
2134 }
2135
2136 /*
2137  * This stores information about the virtual cores currently
2138  * assigned to a physical core.
2139  */
2140 struct core_info {
2141         int             n_subcores;
2142         int             max_subcore_threads;
2143         int             total_threads;
2144         int             subcore_threads[MAX_SUBCORES];
2145         struct kvm      *subcore_vm[MAX_SUBCORES];
2146         struct list_head vcs[MAX_SUBCORES];
2147 };
2148
2149 /*
2150  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2151  * respectively in 2-way micro-threading (split-core) mode.
2152  */
2153 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2154
2155 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2156 {
2157         int sub;
2158
2159         memset(cip, 0, sizeof(*cip));
2160         cip->n_subcores = 1;
2161         cip->max_subcore_threads = vc->num_threads;
2162         cip->total_threads = vc->num_threads;
2163         cip->subcore_threads[0] = vc->num_threads;
2164         cip->subcore_vm[0] = vc->kvm;
2165         for (sub = 0; sub < MAX_SUBCORES; ++sub)
2166                 INIT_LIST_HEAD(&cip->vcs[sub]);
2167         list_add_tail(&vc->preempt_list, &cip->vcs[0]);
2168 }
2169
2170 static bool subcore_config_ok(int n_subcores, int n_threads)
2171 {
2172         /* Can only dynamically split if unsplit to begin with */
2173         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2174                 return false;
2175         if (n_subcores > MAX_SUBCORES)
2176                 return false;
2177         if (n_subcores > 1) {
2178                 if (!(dynamic_mt_modes & 2))
2179                         n_subcores = 4;
2180                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2181                         return false;
2182         }
2183
2184         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2185 }
2186
2187 static void init_master_vcore(struct kvmppc_vcore *vc)
2188 {
2189         vc->master_vcore = vc;
2190         vc->entry_exit_map = 0;
2191         vc->in_guest = 0;
2192         vc->napping_threads = 0;
2193         vc->conferring_threads = 0;
2194 }
2195
2196 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2197 {
2198         int n_threads = vc->num_threads;
2199         int sub;
2200
2201         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2202                 return false;
2203
2204         if (n_threads < cip->max_subcore_threads)
2205                 n_threads = cip->max_subcore_threads;
2206         if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2207                 return false;
2208         cip->max_subcore_threads = n_threads;
2209
2210         sub = cip->n_subcores;
2211         ++cip->n_subcores;
2212         cip->total_threads += vc->num_threads;
2213         cip->subcore_threads[sub] = vc->num_threads;
2214         cip->subcore_vm[sub] = vc->kvm;
2215         init_master_vcore(vc);
2216         list_move_tail(&vc->preempt_list, &cip->vcs[sub]);
2217
2218         return true;
2219 }
2220
2221 /*
2222  * Work out whether it is possible to piggyback the execution of
2223  * vcore *pvc onto the execution of the other vcores described in *cip.
2224  */
2225 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2226                           int target_threads)
2227 {
2228         if (cip->total_threads + pvc->num_threads > target_threads)
2229                 return false;
2230
2231         return can_dynamic_split(pvc, cip);
2232 }
2233
2234 static void prepare_threads(struct kvmppc_vcore *vc)
2235 {
2236         int i;
2237         struct kvm_vcpu *vcpu;
2238
2239         for_each_runnable_thread(i, vcpu, vc) {
2240                 if (signal_pending(vcpu->arch.run_task))
2241                         vcpu->arch.ret = -EINTR;
2242                 else if (vcpu->arch.vpa.update_pending ||
2243                          vcpu->arch.slb_shadow.update_pending ||
2244                          vcpu->arch.dtl.update_pending)
2245                         vcpu->arch.ret = RESUME_GUEST;
2246                 else
2247                         continue;
2248                 kvmppc_remove_runnable(vc, vcpu);
2249                 wake_up(&vcpu->arch.cpu_run);
2250         }
2251 }
2252
2253 static void collect_piggybacks(struct core_info *cip, int target_threads)
2254 {
2255         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2256         struct kvmppc_vcore *pvc, *vcnext;
2257
2258         spin_lock(&lp->lock);
2259         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2260                 if (!spin_trylock(&pvc->lock))
2261                         continue;
2262                 prepare_threads(pvc);
2263                 if (!pvc->n_runnable) {
2264                         list_del_init(&pvc->preempt_list);
2265                         if (pvc->runner == NULL) {
2266                                 pvc->vcore_state = VCORE_INACTIVE;
2267                                 kvmppc_core_end_stolen(pvc);
2268                         }
2269                         spin_unlock(&pvc->lock);
2270                         continue;
2271                 }
2272                 if (!can_piggyback(pvc, cip, target_threads)) {
2273                         spin_unlock(&pvc->lock);
2274                         continue;
2275                 }
2276                 kvmppc_core_end_stolen(pvc);
2277                 pvc->vcore_state = VCORE_PIGGYBACK;
2278                 if (cip->total_threads >= target_threads)
2279                         break;
2280         }
2281         spin_unlock(&lp->lock);
2282 }
2283
2284 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2285 {
2286         int still_running = 0, i;
2287         u64 now;
2288         long ret;
2289         struct kvm_vcpu *vcpu;
2290
2291         spin_lock(&vc->lock);
2292         now = get_tb();
2293         for_each_runnable_thread(i, vcpu, vc) {
2294                 /* cancel pending dec exception if dec is positive */
2295                 if (now < vcpu->arch.dec_expires &&
2296                     kvmppc_core_pending_dec(vcpu))
2297                         kvmppc_core_dequeue_dec(vcpu);
2298
2299                 trace_kvm_guest_exit(vcpu);
2300
2301                 ret = RESUME_GUEST;
2302                 if (vcpu->arch.trap)
2303                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2304                                                     vcpu->arch.run_task);
2305
2306                 vcpu->arch.ret = ret;
2307                 vcpu->arch.trap = 0;
2308
2309                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2310                         if (vcpu->arch.pending_exceptions)
2311                                 kvmppc_core_prepare_to_enter(vcpu);
2312                         if (vcpu->arch.ceded)
2313                                 kvmppc_set_timer(vcpu);
2314                         else
2315                                 ++still_running;
2316                 } else {
2317                         kvmppc_remove_runnable(vc, vcpu);
2318                         wake_up(&vcpu->arch.cpu_run);
2319                 }
2320         }
2321         list_del_init(&vc->preempt_list);
2322         if (!is_master) {
2323                 if (still_running > 0) {
2324                         kvmppc_vcore_preempt(vc);
2325                 } else if (vc->runner) {
2326                         vc->vcore_state = VCORE_PREEMPT;
2327                         kvmppc_core_start_stolen(vc);
2328                 } else {
2329                         vc->vcore_state = VCORE_INACTIVE;
2330                 }
2331                 if (vc->n_runnable > 0 && vc->runner == NULL) {
2332                         /* make sure there's a candidate runner awake */
2333                         i = -1;
2334                         vcpu = next_runnable_thread(vc, &i);
2335                         wake_up(&vcpu->arch.cpu_run);
2336                 }
2337         }
2338         spin_unlock(&vc->lock);
2339 }
2340
2341 /*
2342  * Clear core from the list of active host cores as we are about to
2343  * enter the guest. Only do this if it is the primary thread of the
2344  * core (not if a subcore) that is entering the guest.
2345  */
2346 static inline int kvmppc_clear_host_core(unsigned int cpu)
2347 {
2348         int core;
2349
2350         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2351                 return 0;
2352         /*
2353          * Memory barrier can be omitted here as we will do a smp_wmb()
2354          * later in kvmppc_start_thread and we need ensure that state is
2355          * visible to other CPUs only after we enter guest.
2356          */
2357         core = cpu >> threads_shift;
2358         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2359         return 0;
2360 }
2361
2362 /*
2363  * Advertise this core as an active host core since we exited the guest
2364  * Only need to do this if it is the primary thread of the core that is
2365  * exiting.
2366  */
2367 static inline int kvmppc_set_host_core(unsigned int cpu)
2368 {
2369         int core;
2370
2371         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2372                 return 0;
2373
2374         /*
2375          * Memory barrier can be omitted here because we do a spin_unlock
2376          * immediately after this which provides the memory barrier.
2377          */
2378         core = cpu >> threads_shift;
2379         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
2380         return 0;
2381 }
2382
2383 /*
2384  * Run a set of guest threads on a physical core.
2385  * Called with vc->lock held.
2386  */
2387 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
2388 {
2389         struct kvm_vcpu *vcpu;
2390         int i;
2391         int srcu_idx;
2392         struct core_info core_info;
2393         struct kvmppc_vcore *pvc, *vcnext;
2394         struct kvm_split_mode split_info, *sip;
2395         int split, subcore_size, active;
2396         int sub;
2397         bool thr0_done;
2398         unsigned long cmd_bit, stat_bit;
2399         int pcpu, thr;
2400         int target_threads;
2401         int controlled_threads;
2402
2403         /*
2404          * Remove from the list any threads that have a signal pending
2405          * or need a VPA update done
2406          */
2407         prepare_threads(vc);
2408
2409         /* if the runner is no longer runnable, let the caller pick a new one */
2410         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
2411                 return;
2412
2413         /*
2414          * Initialize *vc.
2415          */
2416         init_master_vcore(vc);
2417         vc->preempt_tb = TB_NIL;
2418
2419         /*
2420          * Number of threads that we will be controlling: the same as
2421          * the number of threads per subcore, except on POWER9,
2422          * where it's 1 because the threads are (mostly) independent.
2423          */
2424         controlled_threads = threads_per_vcore();
2425
2426         /*
2427          * Make sure we are running on primary threads, and that secondary
2428          * threads are offline.  Also check if the number of threads in this
2429          * guest are greater than the current system threads per guest.
2430          */
2431         if ((controlled_threads > 1) &&
2432             ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) {
2433                 for_each_runnable_thread(i, vcpu, vc) {
2434                         vcpu->arch.ret = -EBUSY;
2435                         kvmppc_remove_runnable(vc, vcpu);
2436                         wake_up(&vcpu->arch.cpu_run);
2437                 }
2438                 goto out;
2439         }
2440
2441         /*
2442          * See if we could run any other vcores on the physical core
2443          * along with this one.
2444          */
2445         init_core_info(&core_info, vc);
2446         pcpu = smp_processor_id();
2447         target_threads = controlled_threads;
2448         if (target_smt_mode && target_smt_mode < target_threads)
2449                 target_threads = target_smt_mode;
2450         if (vc->num_threads < target_threads)
2451                 collect_piggybacks(&core_info, target_threads);
2452
2453         /* Decide on micro-threading (split-core) mode */
2454         subcore_size = threads_per_subcore;
2455         cmd_bit = stat_bit = 0;
2456         split = core_info.n_subcores;
2457         sip = NULL;
2458         if (split > 1) {
2459                 /* threads_per_subcore must be MAX_SMT_THREADS (8) here */
2460                 if (split == 2 && (dynamic_mt_modes & 2)) {
2461                         cmd_bit = HID0_POWER8_1TO2LPAR;
2462                         stat_bit = HID0_POWER8_2LPARMODE;
2463                 } else {
2464                         split = 4;
2465                         cmd_bit = HID0_POWER8_1TO4LPAR;
2466                         stat_bit = HID0_POWER8_4LPARMODE;
2467                 }
2468                 subcore_size = MAX_SMT_THREADS / split;
2469                 sip = &split_info;
2470                 memset(&split_info, 0, sizeof(split_info));
2471                 split_info.rpr = mfspr(SPRN_RPR);
2472                 split_info.pmmar = mfspr(SPRN_PMMAR);
2473                 split_info.ldbar = mfspr(SPRN_LDBAR);
2474                 split_info.subcore_size = subcore_size;
2475                 for (sub = 0; sub < core_info.n_subcores; ++sub)
2476                         split_info.master_vcs[sub] =
2477                                 list_first_entry(&core_info.vcs[sub],
2478                                         struct kvmppc_vcore, preempt_list);
2479                 /* order writes to split_info before kvm_split_mode pointer */
2480                 smp_wmb();
2481         }
2482         pcpu = smp_processor_id();
2483         for (thr = 0; thr < controlled_threads; ++thr)
2484                 paca[pcpu + thr].kvm_hstate.kvm_split_mode = sip;
2485
2486         /* Initiate micro-threading (split-core) if required */
2487         if (cmd_bit) {
2488                 unsigned long hid0 = mfspr(SPRN_HID0);
2489
2490                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
2491                 mb();
2492                 mtspr(SPRN_HID0, hid0);
2493                 isync();
2494                 for (;;) {
2495                         hid0 = mfspr(SPRN_HID0);
2496                         if (hid0 & stat_bit)
2497                                 break;
2498                         cpu_relax();
2499                 }
2500         }
2501
2502         kvmppc_clear_host_core(pcpu);
2503
2504         /* Start all the threads */
2505         active = 0;
2506         for (sub = 0; sub < core_info.n_subcores; ++sub) {
2507                 thr = subcore_thread_map[sub];
2508                 thr0_done = false;
2509                 active |= 1 << thr;
2510                 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list) {
2511                         pvc->pcpu = pcpu + thr;
2512                         for_each_runnable_thread(i, vcpu, pvc) {
2513                                 kvmppc_start_thread(vcpu, pvc);
2514                                 kvmppc_create_dtl_entry(vcpu, pvc);
2515                                 trace_kvm_guest_enter(vcpu);
2516                                 if (!vcpu->arch.ptid)
2517                                         thr0_done = true;
2518                                 active |= 1 << (thr + vcpu->arch.ptid);
2519                         }
2520                         /*
2521                          * We need to start the first thread of each subcore
2522                          * even if it doesn't have a vcpu.
2523                          */
2524                         if (pvc->master_vcore == pvc && !thr0_done)
2525                                 kvmppc_start_thread(NULL, pvc);
2526                         thr += pvc->num_threads;
2527                 }
2528         }
2529
2530         /*
2531          * Ensure that split_info.do_nap is set after setting
2532          * the vcore pointer in the PACA of the secondaries.
2533          */
2534         smp_mb();
2535         if (cmd_bit)
2536                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
2537
2538         /*
2539          * When doing micro-threading, poke the inactive threads as well.
2540          * This gets them to the nap instruction after kvm_do_nap,
2541          * which reduces the time taken to unsplit later.
2542          */
2543         if (split > 1)
2544                 for (thr = 1; thr < threads_per_subcore; ++thr)
2545                         if (!(active & (1 << thr)))
2546                                 kvmppc_ipi_thread(pcpu + thr);
2547
2548         vc->vcore_state = VCORE_RUNNING;
2549         preempt_disable();
2550
2551         trace_kvmppc_run_core(vc, 0);
2552
2553         for (sub = 0; sub < core_info.n_subcores; ++sub)
2554                 list_for_each_entry(pvc, &core_info.vcs[sub], preempt_list)
2555                         spin_unlock(&pvc->lock);
2556
2557         guest_enter();
2558
2559         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
2560
2561         __kvmppc_vcore_entry();
2562
2563         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
2564
2565         spin_lock(&vc->lock);
2566         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
2567         vc->vcore_state = VCORE_EXITING;
2568
2569         /* wait for secondary threads to finish writing their state to memory */
2570         kvmppc_wait_for_nap();
2571
2572         /* Return to whole-core mode if we split the core earlier */
2573         if (split > 1) {
2574                 unsigned long hid0 = mfspr(SPRN_HID0);
2575                 unsigned long loops = 0;
2576
2577                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
2578                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
2579                 mb();
2580                 mtspr(SPRN_HID0, hid0);
2581                 isync();
2582                 for (;;) {
2583                         hid0 = mfspr(SPRN_HID0);
2584                         if (!(hid0 & stat_bit))
2585                                 break;
2586                         cpu_relax();
2587                         ++loops;
2588                 }
2589                 split_info.do_nap = 0;
2590         }
2591
2592         /* Let secondaries go back to the offline loop */
2593         for (i = 0; i < controlled_threads; ++i) {
2594                 kvmppc_release_hwthread(pcpu + i);
2595                 if (sip && sip->napped[i])
2596                         kvmppc_ipi_thread(pcpu + i);
2597                 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
2598         }
2599
2600         kvmppc_set_host_core(pcpu);
2601
2602         spin_unlock(&vc->lock);
2603
2604         /* make sure updates to secondary vcpu structs are visible now */
2605         smp_mb();
2606         guest_exit();
2607
2608         for (sub = 0; sub < core_info.n_subcores; ++sub)
2609                 list_for_each_entry_safe(pvc, vcnext, &core_info.vcs[sub],
2610                                          preempt_list)
2611                         post_guest_process(pvc, pvc == vc);
2612
2613         spin_lock(&vc->lock);
2614         preempt_enable();
2615
2616  out:
2617         vc->vcore_state = VCORE_INACTIVE;
2618         trace_kvmppc_run_core(vc, 1);
2619 }
2620
2621 /*
2622  * Wait for some other vcpu thread to execute us, and
2623  * wake us up when we need to handle something in the host.
2624  */
2625 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
2626                                  struct kvm_vcpu *vcpu, int wait_state)
2627 {
2628         DEFINE_WAIT(wait);
2629
2630         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
2631         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2632                 spin_unlock(&vc->lock);
2633                 schedule();
2634                 spin_lock(&vc->lock);
2635         }
2636         finish_wait(&vcpu->arch.cpu_run, &wait);
2637 }
2638
2639 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
2640 {
2641         /* 10us base */
2642         if (vc->halt_poll_ns == 0 && halt_poll_ns_grow)
2643                 vc->halt_poll_ns = 10000;
2644         else
2645                 vc->halt_poll_ns *= halt_poll_ns_grow;
2646 }
2647
2648 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
2649 {
2650         if (halt_poll_ns_shrink == 0)
2651                 vc->halt_poll_ns = 0;
2652         else
2653                 vc->halt_poll_ns /= halt_poll_ns_shrink;
2654 }
2655
2656 /*
2657  * Check to see if any of the runnable vcpus on the vcore have pending
2658  * exceptions or are no longer ceded
2659  */
2660 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
2661 {
2662         struct kvm_vcpu *vcpu;
2663         int i;
2664
2665         for_each_runnable_thread(i, vcpu, vc) {
2666                 if (vcpu->arch.pending_exceptions || !vcpu->arch.ceded ||
2667                     vcpu->arch.prodded)
2668                         return 1;
2669         }
2670
2671         return 0;
2672 }
2673
2674 /*
2675  * All the vcpus in this vcore are idle, so wait for a decrementer
2676  * or external interrupt to one of the vcpus.  vc->lock is held.
2677  */
2678 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
2679 {
2680         ktime_t cur, start_poll, start_wait;
2681         int do_sleep = 1;
2682         u64 block_ns;
2683         DECLARE_SWAITQUEUE(wait);
2684
2685         /* Poll for pending exceptions and ceded state */
2686         cur = start_poll = ktime_get();
2687         if (vc->halt_poll_ns) {
2688                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
2689                 ++vc->runner->stat.halt_attempted_poll;
2690
2691                 vc->vcore_state = VCORE_POLLING;
2692                 spin_unlock(&vc->lock);
2693
2694                 do {
2695                         if (kvmppc_vcore_check_block(vc)) {
2696                                 do_sleep = 0;
2697                                 break;
2698                         }
2699                         cur = ktime_get();
2700                 } while (single_task_running() && ktime_before(cur, stop));
2701
2702                 spin_lock(&vc->lock);
2703                 vc->vcore_state = VCORE_INACTIVE;
2704
2705                 if (!do_sleep) {
2706                         ++vc->runner->stat.halt_successful_poll;
2707                         goto out;
2708                 }
2709         }
2710
2711         prepare_to_swait(&vc->wq, &wait, TASK_INTERRUPTIBLE);
2712
2713         if (kvmppc_vcore_check_block(vc)) {
2714                 finish_swait(&vc->wq, &wait);
2715                 do_sleep = 0;
2716                 /* If we polled, count this as a successful poll */
2717                 if (vc->halt_poll_ns)
2718                         ++vc->runner->stat.halt_successful_poll;
2719                 goto out;
2720         }
2721
2722         start_wait = ktime_get();
2723
2724         vc->vcore_state = VCORE_SLEEPING;
2725         trace_kvmppc_vcore_blocked(vc, 0);
2726         spin_unlock(&vc->lock);
2727         schedule();
2728         finish_swait(&vc->wq, &wait);
2729         spin_lock(&vc->lock);
2730         vc->vcore_state = VCORE_INACTIVE;
2731         trace_kvmppc_vcore_blocked(vc, 1);
2732         ++vc->runner->stat.halt_successful_wait;
2733
2734         cur = ktime_get();
2735
2736 out:
2737         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
2738
2739         /* Attribute wait time */
2740         if (do_sleep) {
2741                 vc->runner->stat.halt_wait_ns +=
2742                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
2743                 /* Attribute failed poll time */
2744                 if (vc->halt_poll_ns)
2745                         vc->runner->stat.halt_poll_fail_ns +=
2746                                 ktime_to_ns(start_wait) -
2747                                 ktime_to_ns(start_poll);
2748         } else {
2749                 /* Attribute successful poll time */
2750                 if (vc->halt_poll_ns)
2751                         vc->runner->stat.halt_poll_success_ns +=
2752                                 ktime_to_ns(cur) -
2753                                 ktime_to_ns(start_poll);
2754         }
2755
2756         /* Adjust poll time */
2757         if (halt_poll_ns) {
2758                 if (block_ns <= vc->halt_poll_ns)
2759                         ;
2760                 /* We slept and blocked for longer than the max halt time */
2761                 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
2762                         shrink_halt_poll_ns(vc);
2763                 /* We slept and our poll time is too small */
2764                 else if (vc->halt_poll_ns < halt_poll_ns &&
2765                                 block_ns < halt_poll_ns)
2766                         grow_halt_poll_ns(vc);
2767                 if (vc->halt_poll_ns > halt_poll_ns)
2768                         vc->halt_poll_ns = halt_poll_ns;
2769         } else
2770                 vc->halt_poll_ns = 0;
2771
2772         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
2773 }
2774
2775 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2776 {
2777         int n_ceded, i;
2778         struct kvmppc_vcore *vc;
2779         struct kvm_vcpu *v;
2780
2781         trace_kvmppc_run_vcpu_enter(vcpu);
2782
2783         kvm_run->exit_reason = 0;
2784         vcpu->arch.ret = RESUME_GUEST;
2785         vcpu->arch.trap = 0;
2786         kvmppc_update_vpas(vcpu);
2787
2788         /*
2789          * Synchronize with other threads in this virtual core
2790          */
2791         vc = vcpu->arch.vcore;
2792         spin_lock(&vc->lock);
2793         vcpu->arch.ceded = 0;
2794         vcpu->arch.run_task = current;
2795         vcpu->arch.kvm_run = kvm_run;
2796         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
2797         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
2798         vcpu->arch.busy_preempt = TB_NIL;
2799         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
2800         ++vc->n_runnable;
2801
2802         /*
2803          * This happens the first time this is called for a vcpu.
2804          * If the vcore is already running, we may be able to start
2805          * this thread straight away and have it join in.
2806          */
2807         if (!signal_pending(current)) {
2808                 if (vc->vcore_state == VCORE_PIGGYBACK) {
2809                         struct kvmppc_vcore *mvc = vc->master_vcore;
2810                         if (spin_trylock(&mvc->lock)) {
2811                                 if (mvc->vcore_state == VCORE_RUNNING &&
2812                                     !VCORE_IS_EXITING(mvc)) {
2813                                         kvmppc_create_dtl_entry(vcpu, vc);
2814                                         kvmppc_start_thread(vcpu, vc);
2815                                         trace_kvm_guest_enter(vcpu);
2816                                 }
2817                                 spin_unlock(&mvc->lock);
2818                         }
2819                 } else if (vc->vcore_state == VCORE_RUNNING &&
2820                            !VCORE_IS_EXITING(vc)) {
2821                         kvmppc_create_dtl_entry(vcpu, vc);
2822                         kvmppc_start_thread(vcpu, vc);
2823                         trace_kvm_guest_enter(vcpu);
2824                 } else if (vc->vcore_state == VCORE_SLEEPING) {
2825                         swake_up(&vc->wq);
2826                 }
2827
2828         }
2829
2830         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2831                !signal_pending(current)) {
2832                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2833                         kvmppc_vcore_end_preempt(vc);
2834
2835                 if (vc->vcore_state != VCORE_INACTIVE) {
2836                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
2837                         continue;
2838                 }
2839                 for_each_runnable_thread(i, v, vc) {
2840                         kvmppc_core_prepare_to_enter(v);
2841                         if (signal_pending(v->arch.run_task)) {
2842                                 kvmppc_remove_runnable(vc, v);
2843                                 v->stat.signal_exits++;
2844                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
2845                                 v->arch.ret = -EINTR;
2846                                 wake_up(&v->arch.cpu_run);
2847                         }
2848                 }
2849                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2850                         break;
2851                 n_ceded = 0;
2852                 for_each_runnable_thread(i, v, vc) {
2853                         if (!v->arch.pending_exceptions && !v->arch.prodded)
2854                                 n_ceded += v->arch.ceded;
2855                         else
2856                                 v->arch.ceded = 0;
2857                 }
2858                 vc->runner = vcpu;
2859                 if (n_ceded == vc->n_runnable) {
2860                         kvmppc_vcore_blocked(vc);
2861                 } else if (need_resched()) {
2862                         kvmppc_vcore_preempt(vc);
2863                         /* Let something else run */
2864                         cond_resched_lock(&vc->lock);
2865                         if (vc->vcore_state == VCORE_PREEMPT)
2866                                 kvmppc_vcore_end_preempt(vc);
2867                 } else {
2868                         kvmppc_run_core(vc);
2869                 }
2870                 vc->runner = NULL;
2871         }
2872
2873         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
2874                (vc->vcore_state == VCORE_RUNNING ||
2875                 vc->vcore_state == VCORE_EXITING ||
2876                 vc->vcore_state == VCORE_PIGGYBACK))
2877                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
2878
2879         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
2880                 kvmppc_vcore_end_preempt(vc);
2881
2882         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
2883                 kvmppc_remove_runnable(vc, vcpu);
2884                 vcpu->stat.signal_exits++;
2885                 kvm_run->exit_reason = KVM_EXIT_INTR;
2886                 vcpu->arch.ret = -EINTR;
2887         }
2888
2889         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
2890                 /* Wake up some vcpu to run the core */
2891                 i = -1;
2892                 v = next_runnable_thread(vc, &i);
2893                 wake_up(&v->arch.cpu_run);
2894         }
2895
2896         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
2897         spin_unlock(&vc->lock);
2898         return vcpu->arch.ret;
2899 }
2900
2901 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
2902 {
2903         int r;
2904         int srcu_idx;
2905
2906         if (!vcpu->arch.sane) {
2907                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2908                 return -EINVAL;
2909         }
2910
2911         kvmppc_core_prepare_to_enter(vcpu);
2912
2913         /* No need to go into the guest when all we'll do is come back out */
2914         if (signal_pending(current)) {
2915                 run->exit_reason = KVM_EXIT_INTR;
2916                 return -EINTR;
2917         }
2918
2919         atomic_inc(&vcpu->kvm->arch.vcpus_running);
2920         /* Order vcpus_running vs. hpte_setup_done, see kvmppc_alloc_reset_hpt */
2921         smp_mb();
2922
2923         /* On the first time here, set up HTAB and VRMA */
2924         if (!kvm_is_radix(vcpu->kvm) && !vcpu->kvm->arch.hpte_setup_done) {
2925                 r = kvmppc_hv_setup_htab_rma(vcpu);
2926                 if (r)
2927                         goto out;
2928         }
2929
2930         flush_all_to_thread(current);
2931
2932         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
2933         vcpu->arch.pgdir = current->mm->pgd;
2934         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2935
2936         do {
2937                 r = kvmppc_run_vcpu(run, vcpu);
2938
2939                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
2940                     !(vcpu->arch.shregs.msr & MSR_PR)) {
2941                         trace_kvm_hcall_enter(vcpu);
2942                         r = kvmppc_pseries_do_hcall(vcpu);
2943                         trace_kvm_hcall_exit(vcpu, r);
2944                         kvmppc_core_prepare_to_enter(vcpu);
2945                 } else if (r == RESUME_PAGE_FAULT) {
2946                         srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2947                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
2948                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
2949                         srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2950                 } else if (r == RESUME_PASSTHROUGH)
2951                         r = kvmppc_xics_rm_complete(vcpu, 0);
2952         } while (is_kvmppc_resume_guest(r));
2953
2954  out:
2955         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2956         atomic_dec(&vcpu->kvm->arch.vcpus_running);
2957         return r;
2958 }
2959
2960 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
2961                                      int linux_psize)
2962 {
2963         struct mmu_psize_def *def = &mmu_psize_defs[linux_psize];
2964
2965         if (!def->shift)
2966                 return;
2967         (*sps)->page_shift = def->shift;
2968         (*sps)->slb_enc = def->sllp;
2969         (*sps)->enc[0].page_shift = def->shift;
2970         (*sps)->enc[0].pte_enc = def->penc[linux_psize];
2971         /*
2972          * Add 16MB MPSS support if host supports it
2973          */
2974         if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) {
2975                 (*sps)->enc[1].page_shift = 24;
2976                 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M];
2977         }
2978         (*sps)++;
2979 }
2980
2981 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
2982                                          struct kvm_ppc_smmu_info *info)
2983 {
2984         struct kvm_ppc_one_seg_page_size *sps;
2985
2986         /*
2987          * Since we don't yet support HPT guests on a radix host,
2988          * return an error if the host uses radix.
2989          */
2990         if (radix_enabled())
2991                 return -EINVAL;
2992
2993         info->flags = KVM_PPC_PAGE_SIZES_REAL;
2994         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
2995                 info->flags |= KVM_PPC_1T_SEGMENTS;
2996         info->slb_size = mmu_slb_size;
2997
2998         /* We only support these sizes for now, and no muti-size segments */
2999         sps = &info->sps[0];
3000         kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K);
3001         kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K);
3002         kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M);
3003
3004         return 0;
3005 }
3006
3007 /*
3008  * Get (and clear) the dirty memory log for a memory slot.
3009  */
3010 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
3011                                          struct kvm_dirty_log *log)
3012 {
3013         struct kvm_memslots *slots;
3014         struct kvm_memory_slot *memslot;
3015         int i, r;
3016         unsigned long n;
3017         unsigned long *buf;
3018         struct kvm_vcpu *vcpu;
3019
3020         mutex_lock(&kvm->slots_lock);
3021
3022         r = -EINVAL;
3023         if (log->slot >= KVM_USER_MEM_SLOTS)
3024                 goto out;
3025
3026         slots = kvm_memslots(kvm);
3027         memslot = id_to_memslot(slots, log->slot);
3028         r = -ENOENT;
3029         if (!memslot->dirty_bitmap)
3030                 goto out;
3031
3032         /*
3033          * Use second half of bitmap area because radix accumulates
3034          * bits in the first half.
3035          */
3036         n = kvm_dirty_bitmap_bytes(memslot);
3037         buf = memslot->dirty_bitmap + n / sizeof(long);
3038         memset(buf, 0, n);
3039
3040         if (kvm_is_radix(kvm))
3041                 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
3042         else
3043                 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
3044         if (r)
3045                 goto out;
3046
3047         /* Harvest dirty bits from VPA and DTL updates */
3048         /* Note: we never modify the SLB shadow buffer areas */
3049         kvm_for_each_vcpu(i, vcpu, kvm) {
3050                 spin_lock(&vcpu->arch.vpa_update_lock);
3051                 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
3052                 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
3053                 spin_unlock(&vcpu->arch.vpa_update_lock);
3054         }
3055
3056         r = -EFAULT;
3057         if (copy_to_user(log->dirty_bitmap, buf, n))
3058                 goto out;
3059
3060         r = 0;
3061 out:
3062         mutex_unlock(&kvm->slots_lock);
3063         return r;
3064 }
3065
3066 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
3067                                         struct kvm_memory_slot *dont)
3068 {
3069         if (!dont || free->arch.rmap != dont->arch.rmap) {
3070                 vfree(free->arch.rmap);
3071                 free->arch.rmap = NULL;
3072         }
3073 }
3074
3075 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
3076                                          unsigned long npages)
3077 {
3078         /*
3079          * For now, if radix_enabled() then we only support radix guests,
3080          * and in that case we don't need the rmap array.
3081          */
3082         if (radix_enabled()) {
3083                 slot->arch.rmap = NULL;
3084                 return 0;
3085         }
3086
3087         slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap));
3088         if (!slot->arch.rmap)
3089                 return -ENOMEM;
3090
3091         return 0;
3092 }
3093
3094 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
3095                                         struct kvm_memory_slot *memslot,
3096                                         const struct kvm_userspace_memory_region *mem)
3097 {
3098         return 0;
3099 }
3100
3101 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
3102                                 const struct kvm_userspace_memory_region *mem,
3103                                 const struct kvm_memory_slot *old,
3104                                 const struct kvm_memory_slot *new)
3105 {
3106         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
3107         struct kvm_memslots *slots;
3108         struct kvm_memory_slot *memslot;
3109
3110         /*
3111          * If we are making a new memslot, it might make
3112          * some address that was previously cached as emulated
3113          * MMIO be no longer emulated MMIO, so invalidate
3114          * all the caches of emulated MMIO translations.
3115          */
3116         if (npages)
3117                 atomic64_inc(&kvm->arch.mmio_update);
3118
3119         if (npages && old->npages && !kvm_is_radix(kvm)) {
3120                 /*
3121                  * If modifying a memslot, reset all the rmap dirty bits.
3122                  * If this is a new memslot, we don't need to do anything
3123                  * since the rmap array starts out as all zeroes,
3124                  * i.e. no pages are dirty.
3125                  */
3126                 slots = kvm_memslots(kvm);
3127                 memslot = id_to_memslot(slots, mem->slot);
3128                 kvmppc_hv_get_dirty_log_hpt(kvm, memslot, NULL);
3129         }
3130 }
3131
3132 /*
3133  * Update LPCR values in kvm->arch and in vcores.
3134  * Caller must hold kvm->lock.
3135  */
3136 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
3137 {
3138         long int i;
3139         u32 cores_done = 0;
3140
3141         if ((kvm->arch.lpcr & mask) == lpcr)
3142                 return;
3143
3144         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
3145
3146         for (i = 0; i < KVM_MAX_VCORES; ++i) {
3147                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
3148                 if (!vc)
3149                         continue;
3150                 spin_lock(&vc->lock);
3151                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
3152                 spin_unlock(&vc->lock);
3153                 if (++cores_done >= kvm->arch.online_vcores)
3154                         break;
3155         }
3156 }
3157
3158 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
3159 {
3160         return;
3161 }
3162
3163 static void kvmppc_setup_partition_table(struct kvm *kvm)
3164 {
3165         unsigned long dw0, dw1;
3166
3167         if (!kvm_is_radix(kvm)) {
3168                 /* PS field - page size for VRMA */
3169                 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
3170                         ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
3171                 /* HTABSIZE and HTABORG fields */
3172                 dw0 |= kvm->arch.sdr1;
3173
3174                 /* Second dword as set by userspace */
3175                 dw1 = kvm->arch.process_table;
3176         } else {
3177                 dw0 = PATB_HR | radix__get_tree_size() |
3178                         __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
3179                 dw1 = PATB_GR | kvm->arch.process_table;
3180         }
3181
3182         mmu_partition_table_set_entry(kvm->arch.lpid, dw0, dw1);
3183 }
3184
3185 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
3186 {
3187         int err = 0;
3188         struct kvm *kvm = vcpu->kvm;
3189         unsigned long hva;
3190         struct kvm_memory_slot *memslot;
3191         struct vm_area_struct *vma;
3192         unsigned long lpcr = 0, senc;
3193         unsigned long psize, porder;
3194         int srcu_idx;
3195
3196         mutex_lock(&kvm->lock);
3197         if (kvm->arch.hpte_setup_done)
3198                 goto out;       /* another vcpu beat us to it */
3199
3200         /* Allocate hashed page table (if not done already) and reset it */
3201         if (!kvm->arch.hpt.virt) {
3202                 int order = KVM_DEFAULT_HPT_ORDER;
3203                 struct kvm_hpt_info info;
3204
3205                 err = kvmppc_allocate_hpt(&info, order);
3206                 /* If we get here, it means userspace didn't specify a
3207                  * size explicitly.  So, try successively smaller
3208                  * sizes if the default failed. */
3209                 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
3210                         err  = kvmppc_allocate_hpt(&info, order);
3211
3212                 if (err < 0) {
3213                         pr_err("KVM: Couldn't alloc HPT\n");
3214                         goto out;
3215                 }
3216
3217                 kvmppc_set_hpt(kvm, &info);
3218         }
3219
3220         /* Look up the memslot for guest physical address 0 */
3221         srcu_idx = srcu_read_lock(&kvm->srcu);
3222         memslot = gfn_to_memslot(kvm, 0);
3223
3224         /* We must have some memory at 0 by now */
3225         err = -EINVAL;
3226         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
3227                 goto out_srcu;
3228
3229         /* Look up the VMA for the start of this memory slot */
3230         hva = memslot->userspace_addr;
3231         down_read(&current->mm->mmap_sem);
3232         vma = find_vma(current->mm, hva);
3233         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
3234                 goto up_out;
3235
3236         psize = vma_kernel_pagesize(vma);
3237         porder = __ilog2(psize);
3238
3239         up_read(&current->mm->mmap_sem);
3240
3241         /* We can handle 4k, 64k or 16M pages in the VRMA */
3242         err = -EINVAL;
3243         if (!(psize == 0x1000 || psize == 0x10000 ||
3244               psize == 0x1000000))
3245                 goto out_srcu;
3246
3247         senc = slb_pgsize_encoding(psize);
3248         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
3249                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3250         /* Create HPTEs in the hash page table for the VRMA */
3251         kvmppc_map_vrma(vcpu, memslot, porder);
3252
3253         /* Update VRMASD field in the LPCR */
3254         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
3255                 /* the -4 is to account for senc values starting at 0x10 */
3256                 lpcr = senc << (LPCR_VRMASD_SH - 4);
3257                 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
3258         } else {
3259                 kvmppc_setup_partition_table(kvm);
3260         }
3261
3262         /* Order updates to kvm->arch.lpcr etc. vs. hpte_setup_done */
3263         smp_wmb();
3264         kvm->arch.hpte_setup_done = 1;
3265         err = 0;
3266  out_srcu:
3267         srcu_read_unlock(&kvm->srcu, srcu_idx);
3268  out:
3269         mutex_unlock(&kvm->lock);
3270         return err;
3271
3272  up_out:
3273         up_read(&current->mm->mmap_sem);
3274         goto out_srcu;
3275 }
3276
3277 #ifdef CONFIG_KVM_XICS
3278 /*
3279  * Allocate a per-core structure for managing state about which cores are
3280  * running in the host versus the guest and for exchanging data between
3281  * real mode KVM and CPU running in the host.
3282  * This is only done for the first VM.
3283  * The allocated structure stays even if all VMs have stopped.
3284  * It is only freed when the kvm-hv module is unloaded.
3285  * It's OK for this routine to fail, we just don't support host
3286  * core operations like redirecting H_IPI wakeups.
3287  */
3288 void kvmppc_alloc_host_rm_ops(void)
3289 {
3290         struct kvmppc_host_rm_ops *ops;
3291         unsigned long l_ops;
3292         int cpu, core;
3293         int size;
3294
3295         /* Not the first time here ? */
3296         if (kvmppc_host_rm_ops_hv != NULL)
3297                 return;
3298
3299         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
3300         if (!ops)
3301                 return;
3302
3303         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
3304         ops->rm_core = kzalloc(size, GFP_KERNEL);
3305
3306         if (!ops->rm_core) {
3307                 kfree(ops);
3308                 return;
3309         }
3310
3311         get_online_cpus();
3312
3313         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
3314                 if (!cpu_online(cpu))
3315                         continue;
3316
3317                 core = cpu >> threads_shift;
3318                 ops->rm_core[core].rm_state.in_host = 1;
3319         }
3320
3321         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
3322
3323         /*
3324          * Make the contents of the kvmppc_host_rm_ops structure visible
3325          * to other CPUs before we assign it to the global variable.
3326          * Do an atomic assignment (no locks used here), but if someone
3327          * beats us to it, just free our copy and return.
3328          */
3329         smp_wmb();
3330         l_ops = (unsigned long) ops;
3331
3332         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
3333                 put_online_cpus();
3334                 kfree(ops->rm_core);
3335                 kfree(ops);
3336                 return;
3337         }
3338
3339         cpuhp_setup_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE,
3340                                   "ppc/kvm_book3s:prepare",
3341                                   kvmppc_set_host_core,
3342                                   kvmppc_clear_host_core);
3343         put_online_cpus();
3344 }
3345
3346 void kvmppc_free_host_rm_ops(void)
3347 {
3348         if (kvmppc_host_rm_ops_hv) {
3349                 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
3350                 kfree(kvmppc_host_rm_ops_hv->rm_core);
3351                 kfree(kvmppc_host_rm_ops_hv);
3352                 kvmppc_host_rm_ops_hv = NULL;
3353         }
3354 }
3355 #endif
3356
3357 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
3358 {
3359         unsigned long lpcr, lpid;
3360         char buf[32];
3361         int ret;
3362
3363         /* Allocate the guest's logical partition ID */
3364
3365         lpid = kvmppc_alloc_lpid();
3366         if ((long)lpid < 0)
3367                 return -ENOMEM;
3368         kvm->arch.lpid = lpid;
3369
3370         kvmppc_alloc_host_rm_ops();
3371
3372         /*
3373          * Since we don't flush the TLB when tearing down a VM,
3374          * and this lpid might have previously been used,
3375          * make sure we flush on each core before running the new VM.
3376          * On POWER9, the tlbie in mmu_partition_table_set_entry()
3377          * does this flush for us.
3378          */
3379         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3380                 cpumask_setall(&kvm->arch.need_tlb_flush);
3381
3382         /* Start out with the default set of hcalls enabled */
3383         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
3384                sizeof(kvm->arch.enabled_hcalls));
3385
3386         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3387                 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
3388
3389         /* Init LPCR for virtual RMA mode */
3390         kvm->arch.host_lpid = mfspr(SPRN_LPID);
3391         kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
3392         lpcr &= LPCR_PECE | LPCR_LPES;
3393         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
3394                 LPCR_VPM0 | LPCR_VPM1;
3395         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
3396                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
3397         /* On POWER8 turn on online bit to enable PURR/SPURR */
3398         if (cpu_has_feature(CPU_FTR_ARCH_207S))
3399                 lpcr |= LPCR_ONL;
3400         /*
3401          * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
3402          * Set HVICE bit to enable hypervisor virtualization interrupts.
3403          */
3404         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
3405                 lpcr &= ~LPCR_VPM0;
3406                 lpcr |= LPCR_HVICE;
3407         }
3408
3409         /*
3410          * For now, if the host uses radix, the guest must be radix.
3411          */
3412         if (radix_enabled()) {
3413                 kvm->arch.radix = 1;
3414                 lpcr &= ~LPCR_VPM1;
3415                 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
3416                 ret = kvmppc_init_vm_radix(kvm);
3417                 if (ret) {
3418                         kvmppc_free_lpid(kvm->arch.lpid);
3419                         return ret;
3420                 }
3421                 kvmppc_setup_partition_table(kvm);
3422         }
3423
3424         kvm->arch.lpcr = lpcr;
3425
3426         /* Initialization for future HPT resizes */
3427         kvm->arch.resize_hpt = NULL;
3428
3429         /*
3430          * Work out how many sets the TLB has, for the use of
3431          * the TLB invalidation loop in book3s_hv_rmhandlers.S.
3432          */
3433         if (kvm_is_radix(kvm))
3434                 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;     /* 128 */
3435         else if (cpu_has_feature(CPU_FTR_ARCH_300))
3436                 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;      /* 256 */
3437         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
3438                 kvm->arch.tlb_sets = POWER8_TLB_SETS;           /* 512 */
3439         else
3440                 kvm->arch.tlb_sets = POWER7_TLB_SETS;           /* 128 */
3441
3442         /*
3443          * Track that we now have a HV mode VM active. This blocks secondary
3444          * CPU threads from coming online.
3445          * On POWER9, we only need to do this for HPT guests on a radix
3446          * host, which is not yet supported.
3447          */
3448         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3449                 kvm_hv_vm_activated();
3450
3451         /*
3452          * Create a debugfs directory for the VM
3453          */
3454         snprintf(buf, sizeof(buf), "vm%d", current->pid);
3455         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
3456         if (!IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
3457                 kvmppc_mmu_debugfs_init(kvm);
3458
3459         return 0;
3460 }
3461
3462 static void kvmppc_free_vcores(struct kvm *kvm)
3463 {
3464         long int i;
3465
3466         for (i = 0; i < KVM_MAX_VCORES; ++i)
3467                 kfree(kvm->arch.vcores[i]);
3468         kvm->arch.online_vcores = 0;
3469 }
3470
3471 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
3472 {
3473         debugfs_remove_recursive(kvm->arch.debugfs_dir);
3474
3475         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3476                 kvm_hv_vm_deactivated();
3477
3478         kvmppc_free_vcores(kvm);
3479
3480         kvmppc_free_lpid(kvm->arch.lpid);
3481
3482         if (kvm_is_radix(kvm))
3483                 kvmppc_free_radix(kvm);
3484         else
3485                 kvmppc_free_hpt(&kvm->arch.hpt);
3486
3487         kvmppc_free_pimap(kvm);
3488 }
3489
3490 /* We don't need to emulate any privileged instructions or dcbz */
3491 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
3492                                      unsigned int inst, int *advance)
3493 {
3494         return EMULATE_FAIL;
3495 }
3496
3497 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
3498                                         ulong spr_val)
3499 {
3500         return EMULATE_FAIL;
3501 }
3502
3503 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
3504                                         ulong *spr_val)
3505 {
3506         return EMULATE_FAIL;
3507 }
3508
3509 static int kvmppc_core_check_processor_compat_hv(void)
3510 {
3511         if (!cpu_has_feature(CPU_FTR_HVMODE) ||
3512             !cpu_has_feature(CPU_FTR_ARCH_206))
3513                 return -EIO;
3514
3515         return 0;
3516 }
3517
3518 #ifdef CONFIG_KVM_XICS
3519
3520 void kvmppc_free_pimap(struct kvm *kvm)
3521 {
3522         kfree(kvm->arch.pimap);
3523 }
3524
3525 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
3526 {
3527         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
3528 }
3529
3530 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3531 {
3532         struct irq_desc *desc;
3533         struct kvmppc_irq_map *irq_map;
3534         struct kvmppc_passthru_irqmap *pimap;
3535         struct irq_chip *chip;
3536         int i;
3537
3538         if (!kvm_irq_bypass)
3539                 return 1;
3540
3541         desc = irq_to_desc(host_irq);
3542         if (!desc)
3543                 return -EIO;
3544
3545         mutex_lock(&kvm->lock);
3546
3547         pimap = kvm->arch.pimap;
3548         if (pimap == NULL) {
3549                 /* First call, allocate structure to hold IRQ map */
3550                 pimap = kvmppc_alloc_pimap();
3551                 if (pimap == NULL) {
3552                         mutex_unlock(&kvm->lock);
3553                         return -ENOMEM;
3554                 }
3555                 kvm->arch.pimap = pimap;
3556         }
3557
3558         /*
3559          * For now, we only support interrupts for which the EOI operation
3560          * is an OPAL call followed by a write to XIRR, since that's
3561          * what our real-mode EOI code does.
3562          */
3563         chip = irq_data_get_irq_chip(&desc->irq_data);
3564         if (!chip || !is_pnv_opal_msi(chip)) {
3565                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
3566                         host_irq, guest_gsi);
3567                 mutex_unlock(&kvm->lock);
3568                 return -ENOENT;
3569         }
3570
3571         /*
3572          * See if we already have an entry for this guest IRQ number.
3573          * If it's mapped to a hardware IRQ number, that's an error,
3574          * otherwise re-use this entry.
3575          */
3576         for (i = 0; i < pimap->n_mapped; i++) {
3577                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
3578                         if (pimap->mapped[i].r_hwirq) {
3579                                 mutex_unlock(&kvm->lock);
3580                                 return -EINVAL;
3581                         }
3582                         break;
3583                 }
3584         }
3585
3586         if (i == KVMPPC_PIRQ_MAPPED) {
3587                 mutex_unlock(&kvm->lock);
3588                 return -EAGAIN;         /* table is full */
3589         }
3590
3591         irq_map = &pimap->mapped[i];
3592
3593         irq_map->v_hwirq = guest_gsi;
3594         irq_map->desc = desc;
3595
3596         /*
3597          * Order the above two stores before the next to serialize with
3598          * the KVM real mode handler.
3599          */
3600         smp_wmb();
3601         irq_map->r_hwirq = desc->irq_data.hwirq;
3602
3603         if (i == pimap->n_mapped)
3604                 pimap->n_mapped++;
3605
3606         kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
3607
3608         mutex_unlock(&kvm->lock);
3609
3610         return 0;
3611 }
3612
3613 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
3614 {
3615         struct irq_desc *desc;
3616         struct kvmppc_passthru_irqmap *pimap;
3617         int i;
3618
3619         if (!kvm_irq_bypass)
3620                 return 0;
3621
3622         desc = irq_to_desc(host_irq);
3623         if (!desc)
3624                 return -EIO;
3625
3626         mutex_lock(&kvm->lock);
3627
3628         if (kvm->arch.pimap == NULL) {
3629                 mutex_unlock(&kvm->lock);
3630                 return 0;
3631         }
3632         pimap = kvm->arch.pimap;
3633
3634         for (i = 0; i < pimap->n_mapped; i++) {
3635                 if (guest_gsi == pimap->mapped[i].v_hwirq)
3636                         break;
3637         }
3638
3639         if (i == pimap->n_mapped) {
3640                 mutex_unlock(&kvm->lock);
3641                 return -ENODEV;
3642         }
3643
3644         kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
3645
3646         /* invalidate the entry */
3647         pimap->mapped[i].r_hwirq = 0;
3648
3649         /*
3650          * We don't free this structure even when the count goes to
3651          * zero. The structure is freed when we destroy the VM.
3652          */
3653
3654         mutex_unlock(&kvm->lock);
3655         return 0;
3656 }
3657
3658 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
3659                                              struct irq_bypass_producer *prod)
3660 {
3661         int ret = 0;
3662         struct kvm_kernel_irqfd *irqfd =
3663                 container_of(cons, struct kvm_kernel_irqfd, consumer);
3664
3665         irqfd->producer = prod;
3666
3667         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
3668         if (ret)
3669                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
3670                         prod->irq, irqfd->gsi, ret);
3671
3672         return ret;
3673 }
3674
3675 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
3676                                               struct irq_bypass_producer *prod)
3677 {
3678         int ret;
3679         struct kvm_kernel_irqfd *irqfd =
3680                 container_of(cons, struct kvm_kernel_irqfd, consumer);
3681
3682         irqfd->producer = NULL;
3683
3684         /*
3685          * When producer of consumer is unregistered, we change back to
3686          * default external interrupt handling mode - KVM real mode
3687          * will switch back to host.
3688          */
3689         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
3690         if (ret)
3691                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
3692                         prod->irq, irqfd->gsi, ret);
3693 }
3694 #endif
3695
3696 static long kvm_arch_vm_ioctl_hv(struct file *filp,
3697                                  unsigned int ioctl, unsigned long arg)
3698 {
3699         struct kvm *kvm __maybe_unused = filp->private_data;
3700         void __user *argp = (void __user *)arg;
3701         long r;
3702
3703         switch (ioctl) {
3704
3705         case KVM_PPC_ALLOCATE_HTAB: {
3706                 u32 htab_order;
3707
3708                 r = -EFAULT;
3709                 if (get_user(htab_order, (u32 __user *)argp))
3710                         break;
3711                 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
3712                 if (r)
3713                         break;
3714                 r = 0;
3715                 break;
3716         }
3717
3718         case KVM_PPC_GET_HTAB_FD: {
3719                 struct kvm_get_htab_fd ghf;
3720
3721                 r = -EFAULT;
3722                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
3723                         break;
3724                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
3725                 break;
3726         }
3727
3728         case KVM_PPC_RESIZE_HPT_PREPARE: {
3729                 struct kvm_ppc_resize_hpt rhpt;
3730
3731                 r = -EFAULT;
3732                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
3733                         break;
3734
3735                 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
3736                 break;
3737         }
3738
3739         case KVM_PPC_RESIZE_HPT_COMMIT: {
3740                 struct kvm_ppc_resize_hpt rhpt;
3741
3742                 r = -EFAULT;
3743                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
3744                         break;
3745
3746                 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
3747                 break;
3748         }
3749
3750         default:
3751                 r = -ENOTTY;
3752         }
3753
3754         return r;
3755 }
3756
3757 /*
3758  * List of hcall numbers to enable by default.
3759  * For compatibility with old userspace, we enable by default
3760  * all hcalls that were implemented before the hcall-enabling
3761  * facility was added.  Note this list should not include H_RTAS.
3762  */
3763 static unsigned int default_hcall_list[] = {
3764         H_REMOVE,
3765         H_ENTER,
3766         H_READ,
3767         H_PROTECT,
3768         H_BULK_REMOVE,
3769         H_GET_TCE,
3770         H_PUT_TCE,
3771         H_SET_DABR,
3772         H_SET_XDABR,
3773         H_CEDE,
3774         H_PROD,
3775         H_CONFER,
3776         H_REGISTER_VPA,
3777 #ifdef CONFIG_KVM_XICS
3778         H_EOI,
3779         H_CPPR,
3780         H_IPI,
3781         H_IPOLL,
3782         H_XIRR,
3783         H_XIRR_X,
3784 #endif
3785         0
3786 };
3787
3788 static void init_default_hcalls(void)
3789 {
3790         int i;
3791         unsigned int hcall;
3792
3793         for (i = 0; default_hcall_list[i]; ++i) {
3794                 hcall = default_hcall_list[i];
3795                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
3796                 __set_bit(hcall / 4, default_enabled_hcalls);
3797         }
3798 }
3799
3800 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
3801 {
3802         unsigned long lpcr;
3803         int radix;
3804
3805         /* If not on a POWER9, reject it */
3806         if (!cpu_has_feature(CPU_FTR_ARCH_300))
3807                 return -ENODEV;
3808
3809         /* If any unknown flags set, reject it */
3810         if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
3811                 return -EINVAL;
3812
3813         /* We can't change a guest to/from radix yet */
3814         radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
3815         if (radix != kvm_is_radix(kvm))
3816                 return -EINVAL;
3817
3818         /* GR (guest radix) bit in process_table field must match */
3819         if (!!(cfg->process_table & PATB_GR) != radix)
3820                 return -EINVAL;
3821
3822         /* Process table size field must be reasonable, i.e. <= 24 */
3823         if ((cfg->process_table & PRTS_MASK) > 24)
3824                 return -EINVAL;
3825
3826         kvm->arch.process_table = cfg->process_table;
3827         kvmppc_setup_partition_table(kvm);
3828
3829         lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
3830         kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
3831
3832         return 0;
3833 }
3834
3835 static struct kvmppc_ops kvm_ops_hv = {
3836         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
3837         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
3838         .get_one_reg = kvmppc_get_one_reg_hv,
3839         .set_one_reg = kvmppc_set_one_reg_hv,
3840         .vcpu_load   = kvmppc_core_vcpu_load_hv,
3841         .vcpu_put    = kvmppc_core_vcpu_put_hv,
3842         .set_msr     = kvmppc_set_msr_hv,
3843         .vcpu_run    = kvmppc_vcpu_run_hv,
3844         .vcpu_create = kvmppc_core_vcpu_create_hv,
3845         .vcpu_free   = kvmppc_core_vcpu_free_hv,
3846         .check_requests = kvmppc_core_check_requests_hv,
3847         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
3848         .flush_memslot  = kvmppc_core_flush_memslot_hv,
3849         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
3850         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
3851         .unmap_hva = kvm_unmap_hva_hv,
3852         .unmap_hva_range = kvm_unmap_hva_range_hv,
3853         .age_hva  = kvm_age_hva_hv,
3854         .test_age_hva = kvm_test_age_hva_hv,
3855         .set_spte_hva = kvm_set_spte_hva_hv,
3856         .mmu_destroy  = kvmppc_mmu_destroy_hv,
3857         .free_memslot = kvmppc_core_free_memslot_hv,
3858         .create_memslot = kvmppc_core_create_memslot_hv,
3859         .init_vm =  kvmppc_core_init_vm_hv,
3860         .destroy_vm = kvmppc_core_destroy_vm_hv,
3861         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
3862         .emulate_op = kvmppc_core_emulate_op_hv,
3863         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
3864         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
3865         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
3866         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
3867         .hcall_implemented = kvmppc_hcall_impl_hv,
3868 #ifdef CONFIG_KVM_XICS
3869         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
3870         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
3871 #endif
3872         .configure_mmu = kvmhv_configure_mmu,
3873         .get_rmmu_info = kvmhv_get_rmmu_info,
3874 };
3875
3876 static int kvm_init_subcore_bitmap(void)
3877 {
3878         int i, j;
3879         int nr_cores = cpu_nr_cores();
3880         struct sibling_subcore_state *sibling_subcore_state;
3881
3882         for (i = 0; i < nr_cores; i++) {
3883                 int first_cpu = i * threads_per_core;
3884                 int node = cpu_to_node(first_cpu);
3885
3886                 /* Ignore if it is already allocated. */
3887                 if (paca[first_cpu].sibling_subcore_state)
3888                         continue;
3889
3890                 sibling_subcore_state =
3891                         kmalloc_node(sizeof(struct sibling_subcore_state),
3892                                                         GFP_KERNEL, node);
3893                 if (!sibling_subcore_state)
3894                         return -ENOMEM;
3895
3896                 memset(sibling_subcore_state, 0,
3897                                 sizeof(struct sibling_subcore_state));
3898
3899                 for (j = 0; j < threads_per_core; j++) {
3900                         int cpu = first_cpu + j;
3901
3902                         paca[cpu].sibling_subcore_state = sibling_subcore_state;
3903                 }
3904         }
3905         return 0;
3906 }
3907
3908 static int kvmppc_radix_possible(void)
3909 {
3910         return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
3911 }
3912
3913 static int kvmppc_book3s_init_hv(void)
3914 {
3915         int r;
3916         /*
3917          * FIXME!! Do we need to check on all cpus ?
3918          */
3919         r = kvmppc_core_check_processor_compat_hv();
3920         if (r < 0)
3921                 return -ENODEV;
3922
3923         r = kvm_init_subcore_bitmap();
3924         if (r)
3925                 return r;
3926
3927         /*
3928          * We need a way of accessing the XICS interrupt controller,
3929          * either directly, via paca[cpu].kvm_hstate.xics_phys, or
3930          * indirectly, via OPAL.
3931          */
3932 #ifdef CONFIG_SMP
3933         if (!get_paca()->kvm_hstate.xics_phys) {
3934                 struct device_node *np;
3935
3936                 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
3937                 if (!np) {
3938                         pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
3939                         return -ENODEV;
3940                 }
3941         }
3942 #endif
3943
3944         kvm_ops_hv.owner = THIS_MODULE;
3945         kvmppc_hv_ops = &kvm_ops_hv;
3946
3947         init_default_hcalls();
3948
3949         init_vcore_lists();
3950
3951         r = kvmppc_mmu_hv_init();
3952         if (r)
3953                 return r;
3954
3955         if (kvmppc_radix_possible())
3956                 r = kvmppc_radix_init();
3957         return r;
3958 }
3959
3960 static void kvmppc_book3s_exit_hv(void)
3961 {
3962         kvmppc_free_host_rm_ops();
3963         if (kvmppc_radix_possible())
3964                 kvmppc_radix_exit();
3965         kvmppc_hv_ops = NULL;
3966 }
3967
3968 module_init(kvmppc_book3s_init_hv);
3969 module_exit(kvmppc_book3s_exit_hv);
3970 MODULE_LICENSE("GPL");
3971 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3972 MODULE_ALIAS("devname:kvm");
3973