arch/powerpc/kvm/book3s_hv.c

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