arch/powerpc/kvm/book3s_hv.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
   4  * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved.
   5  *
   6  * Authors:
   7  *    Paul Mackerras <paulus@au1.ibm.com>
   8  *    Alexander Graf <agraf@suse.de>
   9  *    Kevin Wolf <mail@kevin-wolf.de>
  10  *
  11  * Description: KVM functions specific to running on Book 3S
  12  * processors in hypervisor mode (specifically POWER7 and later).
  13  *
  14  * This file is derived from arch/powerpc/kvm/book3s.c,
  15  * by Alexander Graf <agraf@suse.de>.
  16  */
  17
  18 #include <linux/kvm_host.h>
  19 #include <linux/kernel.h>
  20 #include <linux/err.h>
  21 #include <linux/slab.h>
  22 #include <linux/preempt.h>
  23 #include <linux/sched/signal.h>
  24 #include <linux/sched/stat.h>
  25 #include <linux/delay.h>
  26 #include <linux/export.h>
  27 #include <linux/fs.h>
  28 #include <linux/anon_inodes.h>
  29 #include <linux/cpu.h>
  30 #include <linux/cpumask.h>
  31 #include <linux/spinlock.h>
  32 #include <linux/page-flags.h>
  33 #include <linux/srcu.h>
  34 #include <linux/miscdevice.h>
  35 #include <linux/debugfs.h>
  36 #include <linux/gfp.h>
  37 #include <linux/vmalloc.h>
  38 #include <linux/highmem.h>
  39 #include <linux/hugetlb.h>
  40 #include <linux/kvm_irqfd.h>
  41 #include <linux/irqbypass.h>
  42 #include <linux/module.h>
  43 #include <linux/compiler.h>
  44 #include <linux/of.h>
  45
  46 #include <asm/ftrace.h>
  47 #include <asm/reg.h>
  48 #include <asm/ppc-opcode.h>
  49 #include <asm/asm-prototypes.h>
  50 #include <asm/archrandom.h>
  51 #include <asm/debug.h>
  52 #include <asm/disassemble.h>
  53 #include <asm/cputable.h>
  54 #include <asm/cacheflush.h>
  55 #include <linux/uaccess.h>
  56 #include <asm/io.h>
  57 #include <asm/kvm_ppc.h>
  58 #include <asm/kvm_book3s.h>
  59 #include <asm/mmu_context.h>
  60 #include <asm/lppaca.h>
  61 #include <asm/processor.h>
  62 #include <asm/cputhreads.h>
  63 #include <asm/page.h>
  64 #include <asm/hvcall.h>
  65 #include <asm/switch_to.h>
  66 #include <asm/smp.h>
  67 #include <asm/dbell.h>
  68 #include <asm/hmi.h>
  69 #include <asm/pnv-pci.h>
  70 #include <asm/mmu.h>
  71 #include <asm/opal.h>
  72 #include <asm/xics.h>
  73 #include <asm/xive.h>
  74 #include <asm/hw_breakpoint.h>
  75 #include <asm/kvm_book3s_uvmem.h>
  76 #include <asm/ultravisor.h>
  77
  78 #include "book3s.h"
  79
  80 #define CREATE_TRACE_POINTS
  81 #include "trace_hv.h"
  82
  83 /* #define EXIT_DEBUG */
  84 /* #define EXIT_DEBUG_SIMPLE */
  85 /* #define EXIT_DEBUG_INT */
  86
  87 /* Used to indicate that a guest page fault needs to be handled */
  88 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  89 /* Used to indicate that a guest passthrough interrupt needs to be handled */
  90 #define RESUME_PASSTHROUGH      (RESUME_GUEST | RESUME_FLAG_ARCH2)
  91
  92 /* Used as a "null" value for timebase values */
  93 #define TB_NIL  (~(u64)0)
  94
  95 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
  96
  97 static int dynamic_mt_modes = 6;
  98 module_param(dynamic_mt_modes, int, 0644);
  99 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
 100 static int target_smt_mode;
 101 module_param(target_smt_mode, int, 0644);
 102 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
 103
 104 static bool indep_threads_mode = true;
 105 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
 106 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
 107
 108 static bool one_vm_per_core;
 109 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
 110 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires indep_threads_mode=N)");
 111
 112 #ifdef CONFIG_KVM_XICS
 113 static struct kernel_param_ops module_param_ops = {
 114         .set = param_set_int,
 115         .get = param_get_int,
 116 };
 117
 118 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
 119 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
 120
 121 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
 122 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
 123 #endif
 124
 125 /* If set, guests are allowed to create and control nested guests */
 126 static bool nested = true;
 127 module_param(nested, bool, S_IRUGO | S_IWUSR);
 128 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
 129
 130 static inline bool nesting_enabled(struct kvm *kvm)
 131 {
 132         return kvm->arch.nested_enable && kvm_is_radix(kvm);
 133 }
 134
 135 /* If set, the threads on each CPU core have to be in the same MMU mode */
 136 static bool no_mixing_hpt_and_radix;
 137
 138 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
 139
 140 /*
 141  * RWMR values for POWER8.  These control the rate at which PURR
 142  * and SPURR count and should be set according to the number of
 143  * online threads in the vcore being run.
 144  */
 145 #define RWMR_RPA_P8_1THREAD     0x164520C62609AECAUL
 146 #define RWMR_RPA_P8_2THREAD     0x7FFF2908450D8DA9UL
 147 #define RWMR_RPA_P8_3THREAD     0x164520C62609AECAUL
 148 #define RWMR_RPA_P8_4THREAD     0x199A421245058DA9UL
 149 #define RWMR_RPA_P8_5THREAD     0x164520C62609AECAUL
 150 #define RWMR_RPA_P8_6THREAD     0x164520C62609AECAUL
 151 #define RWMR_RPA_P8_7THREAD     0x164520C62609AECAUL
 152 #define RWMR_RPA_P8_8THREAD     0x164520C62609AECAUL
 153
 154 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
 155         RWMR_RPA_P8_1THREAD,
 156         RWMR_RPA_P8_1THREAD,
 157         RWMR_RPA_P8_2THREAD,
 158         RWMR_RPA_P8_3THREAD,
 159         RWMR_RPA_P8_4THREAD,
 160         RWMR_RPA_P8_5THREAD,
 161         RWMR_RPA_P8_6THREAD,
 162         RWMR_RPA_P8_7THREAD,
 163         RWMR_RPA_P8_8THREAD,
 164 };
 165
 166 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
 167                 int *ip)
 168 {
 169         int i = *ip;
 170         struct kvm_vcpu *vcpu;
 171
 172         while (++i < MAX_SMT_THREADS) {
 173                 vcpu = READ_ONCE(vc->runnable_threads[i]);
 174                 if (vcpu) {
 175                         *ip = i;
 176                         return vcpu;
 177                 }
 178         }
 179         return NULL;
 180 }
 181
 182 /* Used to traverse the list of runnable threads for a given vcore */
 183 #define for_each_runnable_thread(i, vcpu, vc) \
 184         for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
 185
 186 static bool kvmppc_ipi_thread(int cpu)
 187 {
 188         unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
 189
 190         /* If we're a nested hypervisor, fall back to ordinary IPIs for now */
 191         if (kvmhv_on_pseries())
 192                 return false;
 193
 194         /* On POWER9 we can use msgsnd to IPI any cpu */
 195         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
 196                 msg |= get_hard_smp_processor_id(cpu);
 197                 smp_mb();
 198                 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 199                 return true;
 200         }
 201
 202         /* On POWER8 for IPIs to threads in the same core, use msgsnd */
 203         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
 204                 preempt_disable();
 205                 if (cpu_first_thread_sibling(cpu) ==
 206                     cpu_first_thread_sibling(smp_processor_id())) {
 207                         msg |= cpu_thread_in_core(cpu);
 208                         smp_mb();
 209                         __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 210                         preempt_enable();
 211                         return true;
 212                 }
 213                 preempt_enable();
 214         }
 215
 216 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
 217         if (cpu >= 0 && cpu < nr_cpu_ids) {
 218                 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
 219                         xics_wake_cpu(cpu);
 220                         return true;
 221                 }
 222                 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
 223                 return true;
 224         }
 225 #endif
 226
 227         return false;
 228 }
 229
 230 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
 231 {
 232         int cpu;
 233         struct rcuwait *waitp;
 234
 235         waitp = kvm_arch_vcpu_get_wait(vcpu);
 236         if (rcuwait_wake_up(waitp))
 237                 ++vcpu->stat.halt_wakeup;
 238
 239         cpu = READ_ONCE(vcpu->arch.thread_cpu);
 240         if (cpu >= 0 && kvmppc_ipi_thread(cpu))
 241                 return;
 242
 243         /* CPU points to the first thread of the core */
 244         cpu = vcpu->cpu;
 245         if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
 246                 smp_send_reschedule(cpu);
 247 }
 248
 249 /*
 250  * We use the vcpu_load/put functions to measure stolen time.
 251  * Stolen time is counted as time when either the vcpu is able to
 252  * run as part of a virtual core, but the task running the vcore
 253  * is preempted or sleeping, or when the vcpu needs something done
 254  * in the kernel by the task running the vcpu, but that task is
 255  * preempted or sleeping.  Those two things have to be counted
 256  * separately, since one of the vcpu tasks will take on the job
 257  * of running the core, and the other vcpu tasks in the vcore will
 258  * sleep waiting for it to do that, but that sleep shouldn't count
 259  * as stolen time.
 260  *
 261  * Hence we accumulate stolen time when the vcpu can run as part of
 262  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 263  * needs its task to do other things in the kernel (for example,
 264  * service a page fault) in busy_stolen.  We don't accumulate
 265  * stolen time for a vcore when it is inactive, or for a vcpu
 266  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 267  * a misnomer; it means that the vcpu task is not executing in
 268  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 269  * the kernel.  We don't have any way of dividing up that time
 270  * between time that the vcpu is genuinely stopped, time that
 271  * the task is actively working on behalf of the vcpu, and time
 272  * that the task is preempted, so we don't count any of it as
 273  * stolen.
 274  *
 275  * Updates to busy_stolen are protected by arch.tbacct_lock;
 276  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
 277  * lock.  The stolen times are measured in units of timebase ticks.
 278  * (Note that the != TB_NIL checks below are purely defensive;
 279  * they should never fail.)
 280  */
 281
 282 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
 283 {
 284         unsigned long flags;
 285
 286         spin_lock_irqsave(&vc->stoltb_lock, flags);
 287         vc->preempt_tb = mftb();
 288         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 289 }
 290
 291 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
 292 {
 293         unsigned long flags;
 294
 295         spin_lock_irqsave(&vc->stoltb_lock, flags);
 296         if (vc->preempt_tb != TB_NIL) {
 297                 vc->stolen_tb += mftb() - vc->preempt_tb;
 298                 vc->preempt_tb = TB_NIL;
 299         }
 300         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 301 }
 302
 303 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 304 {
 305         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 306         unsigned long flags;
 307
 308         /*
 309          * We can test vc->runner without taking the vcore lock,
 310          * because only this task ever sets vc->runner to this
 311          * vcpu, and once it is set to this vcpu, only this task
 312          * ever sets it to NULL.
 313          */
 314         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 315                 kvmppc_core_end_stolen(vc);
 316
 317         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 318         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 319             vcpu->arch.busy_preempt != TB_NIL) {
 320                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 321                 vcpu->arch.busy_preempt = TB_NIL;
 322         }
 323         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 324 }
 325
 326 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 327 {
 328         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 329         unsigned long flags;
 330
 331         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 332                 kvmppc_core_start_stolen(vc);
 333
 334         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 335         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 336                 vcpu->arch.busy_preempt = mftb();
 337         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 338 }
 339
 340 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 341 {
 342         vcpu->arch.pvr = pvr;
 343 }
 344
 345 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 346 {
 347         unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
 348         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 349
 350         /* We can (emulate) our own architecture version and anything older */
 351         if (cpu_has_feature(CPU_FTR_ARCH_300))
 352                 host_pcr_bit = PCR_ARCH_300;
 353         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
 354                 host_pcr_bit = PCR_ARCH_207;
 355         else if (cpu_has_feature(CPU_FTR_ARCH_206))
 356                 host_pcr_bit = PCR_ARCH_206;
 357         else
 358                 host_pcr_bit = PCR_ARCH_205;
 359
 360         /* Determine lowest PCR bit needed to run guest in given PVR level */
 361         guest_pcr_bit = host_pcr_bit;
 362         if (arch_compat) {
 363                 switch (arch_compat) {
 364                 case PVR_ARCH_205:
 365                         guest_pcr_bit = PCR_ARCH_205;
 366                         break;
 367                 case PVR_ARCH_206:
 368                 case PVR_ARCH_206p:
 369                         guest_pcr_bit = PCR_ARCH_206;
 370                         break;
 371                 case PVR_ARCH_207:
 372                         guest_pcr_bit = PCR_ARCH_207;
 373                         break;
 374                 case PVR_ARCH_300:
 375                         guest_pcr_bit = PCR_ARCH_300;
 376                         break;
 377                 default:
 378                         return -EINVAL;
 379                 }
 380         }
 381
 382         /* Check requested PCR bits don't exceed our capabilities */
 383         if (guest_pcr_bit > host_pcr_bit)
 384                 return -EINVAL;
 385
 386         spin_lock(&vc->lock);
 387         vc->arch_compat = arch_compat;
 388         /*
 389          * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
 390          * Also set all reserved PCR bits
 391          */
 392         vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
 393         spin_unlock(&vc->lock);
 394
 395         return 0;
 396 }
 397
 398 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 399 {
 400         int r;
 401
 402         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 403         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 404                vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
 405         for (r = 0; r < 16; ++r)
 406                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 407                        r, kvmppc_get_gpr(vcpu, r),
 408                        r+16, kvmppc_get_gpr(vcpu, r+16));
 409         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 410                vcpu->arch.regs.ctr, vcpu->arch.regs.link);
 411         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 412                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 413         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 414                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 415         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 416                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 417         pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
 418                vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
 419         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 420         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 421                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 422         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 423         for (r = 0; r < vcpu->arch.slb_max; ++r)
 424                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 425                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 426         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 427                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 428                vcpu->arch.last_inst);
 429 }
 430
 431 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 432 {
 433         return kvm_get_vcpu_by_id(kvm, id);
 434 }
 435
 436 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 437 {
 438         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 439         vpa->yield_count = cpu_to_be32(1);
 440 }
 441
 442 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 443                    unsigned long addr, unsigned long len)
 444 {
 445         /* check address is cacheline aligned */
 446         if (addr & (L1_CACHE_BYTES - 1))
 447                 return -EINVAL;
 448         spin_lock(&vcpu->arch.vpa_update_lock);
 449         if (v->next_gpa != addr || v->len != len) {
 450                 v->next_gpa = addr;
 451                 v->len = addr ? len : 0;
 452                 v->update_pending = 1;
 453         }
 454         spin_unlock(&vcpu->arch.vpa_update_lock);
 455         return 0;
 456 }
 457
 458 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 459 struct reg_vpa {
 460         u32 dummy;
 461         union {
 462                 __be16 hword;
 463                 __be32 word;
 464         } length;
 465 };
 466
 467 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 468 {
 469         if (vpap->update_pending)
 470                 return vpap->next_gpa != 0;
 471         return vpap->pinned_addr != NULL;
 472 }
 473
 474 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 475                                        unsigned long flags,
 476                                        unsigned long vcpuid, unsigned long vpa)
 477 {
 478         struct kvm *kvm = vcpu->kvm;
 479         unsigned long len, nb;
 480         void *va;
 481         struct kvm_vcpu *tvcpu;
 482         int err;
 483         int subfunc;
 484         struct kvmppc_vpa *vpap;
 485
 486         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 487         if (!tvcpu)
 488                 return H_PARAMETER;
 489
 490         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 491         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 492             subfunc == H_VPA_REG_SLB) {
 493                 /* Registering new area - address must be cache-line aligned */
 494                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 495                         return H_PARAMETER;
 496
 497                 /* convert logical addr to kernel addr and read length */
 498                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 499                 if (va == NULL)
 500                         return H_PARAMETER;
 501                 if (subfunc == H_VPA_REG_VPA)
 502                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 503                 else
 504                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 505                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 506
 507                 /* Check length */
 508                 if (len > nb || len < sizeof(struct reg_vpa))
 509                         return H_PARAMETER;
 510         } else {
 511                 vpa = 0;
 512                 len = 0;
 513         }
 514
 515         err = H_PARAMETER;
 516         vpap = NULL;
 517         spin_lock(&tvcpu->arch.vpa_update_lock);
 518
 519         switch (subfunc) {
 520         case H_VPA_REG_VPA:             /* register VPA */
 521                 /*
 522                  * The size of our lppaca is 1kB because of the way we align
 523                  * it for the guest to avoid crossing a 4kB boundary. We only
 524                  * use 640 bytes of the structure though, so we should accept
 525                  * clients that set a size of 640.
 526                  */
 527                 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
 528                 if (len < sizeof(struct lppaca))
 529                         break;
 530                 vpap = &tvcpu->arch.vpa;
 531                 err = 0;
 532                 break;
 533
 534         case H_VPA_REG_DTL:             /* register DTL */
 535                 if (len < sizeof(struct dtl_entry))
 536                         break;
 537                 len -= len % sizeof(struct dtl_entry);
 538
 539                 /* Check that they have previously registered a VPA */
 540                 err = H_RESOURCE;
 541                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 542                         break;
 543
 544                 vpap = &tvcpu->arch.dtl;
 545                 err = 0;
 546                 break;
 547
 548         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 549                 /* Check that they have previously registered a VPA */
 550                 err = H_RESOURCE;
 551                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 552                         break;
 553
 554                 vpap = &tvcpu->arch.slb_shadow;
 555                 err = 0;
 556                 break;
 557
 558         case H_VPA_DEREG_VPA:           /* deregister VPA */
 559                 /* Check they don't still have a DTL or SLB buf registered */
 560                 err = H_RESOURCE;
 561                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 562                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 563                         break;
 564
 565                 vpap = &tvcpu->arch.vpa;
 566                 err = 0;
 567                 break;
 568
 569         case H_VPA_DEREG_DTL:           /* deregister DTL */
 570                 vpap = &tvcpu->arch.dtl;
 571                 err = 0;
 572                 break;
 573
 574         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 575                 vpap = &tvcpu->arch.slb_shadow;
 576                 err = 0;
 577                 break;
 578         }
 579
 580         if (vpap) {
 581                 vpap->next_gpa = vpa;
 582                 vpap->len = len;
 583                 vpap->update_pending = 1;
 584         }
 585
 586         spin_unlock(&tvcpu->arch.vpa_update_lock);
 587
 588         return err;
 589 }
 590
 591 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 592 {
 593         struct kvm *kvm = vcpu->kvm;
 594         void *va;
 595         unsigned long nb;
 596         unsigned long gpa;
 597
 598         /*
 599          * We need to pin the page pointed to by vpap->next_gpa,
 600          * but we can't call kvmppc_pin_guest_page under the lock
 601          * as it does get_user_pages() and down_read().  So we
 602          * have to drop the lock, pin the page, then get the lock
 603          * again and check that a new area didn't get registered
 604          * in the meantime.
 605          */
 606         for (;;) {
 607                 gpa = vpap->next_gpa;
 608                 spin_unlock(&vcpu->arch.vpa_update_lock);
 609                 va = NULL;
 610                 nb = 0;
 611                 if (gpa)
 612                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 613                 spin_lock(&vcpu->arch.vpa_update_lock);
 614                 if (gpa == vpap->next_gpa)
 615                         break;
 616                 /* sigh... unpin that one and try again */
 617                 if (va)
 618                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 619         }
 620
 621         vpap->update_pending = 0;
 622         if (va && nb < vpap->len) {
 623                 /*
 624                  * If it's now too short, it must be that userspace
 625                  * has changed the mappings underlying guest memory,
 626                  * so unregister the region.
 627                  */
 628                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 629                 va = NULL;
 630         }
 631         if (vpap->pinned_addr)
 632                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 633                                         vpap->dirty);
 634         vpap->gpa = gpa;
 635         vpap->pinned_addr = va;
 636         vpap->dirty = false;
 637         if (va)
 638                 vpap->pinned_end = va + vpap->len;
 639 }
 640
 641 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 642 {
 643         if (!(vcpu->arch.vpa.update_pending ||
 644               vcpu->arch.slb_shadow.update_pending ||
 645               vcpu->arch.dtl.update_pending))
 646                 return;
 647
 648         spin_lock(&vcpu->arch.vpa_update_lock);
 649         if (vcpu->arch.vpa.update_pending) {
 650                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 651                 if (vcpu->arch.vpa.pinned_addr)
 652                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 653         }
 654         if (vcpu->arch.dtl.update_pending) {
 655                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 656                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 657                 vcpu->arch.dtl_index = 0;
 658         }
 659         if (vcpu->arch.slb_shadow.update_pending)
 660                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 661         spin_unlock(&vcpu->arch.vpa_update_lock);
 662 }
 663
 664 /*
 665  * Return the accumulated stolen time for the vcore up until `now'.
 666  * The caller should hold the vcore lock.
 667  */
 668 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 669 {
 670         u64 p;
 671         unsigned long flags;
 672
 673         spin_lock_irqsave(&vc->stoltb_lock, flags);
 674         p = vc->stolen_tb;
 675         if (vc->vcore_state != VCORE_INACTIVE &&
 676             vc->preempt_tb != TB_NIL)
 677                 p += now - vc->preempt_tb;
 678         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 679         return p;
 680 }
 681
 682 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 683                                     struct kvmppc_vcore *vc)
 684 {
 685         struct dtl_entry *dt;
 686         struct lppaca *vpa;
 687         unsigned long stolen;
 688         unsigned long core_stolen;
 689         u64 now;
 690         unsigned long flags;
 691
 692         dt = vcpu->arch.dtl_ptr;
 693         vpa = vcpu->arch.vpa.pinned_addr;
 694         now = mftb();
 695         core_stolen = vcore_stolen_time(vc, now);
 696         stolen = core_stolen - vcpu->arch.stolen_logged;
 697         vcpu->arch.stolen_logged = core_stolen;
 698         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 699         stolen += vcpu->arch.busy_stolen;
 700         vcpu->arch.busy_stolen = 0;
 701         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 702         if (!dt || !vpa)
 703                 return;
 704         memset(dt, 0, sizeof(struct dtl_entry));
 705         dt->dispatch_reason = 7;
 706         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 707         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 708         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 709         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 710         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 711         ++dt;
 712         if (dt == vcpu->arch.dtl.pinned_end)
 713                 dt = vcpu->arch.dtl.pinned_addr;
 714         vcpu->arch.dtl_ptr = dt;
 715         /* order writing *dt vs. writing vpa->dtl_idx */
 716         smp_wmb();
 717         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 718         vcpu->arch.dtl.dirty = true;
 719 }
 720
 721 /* See if there is a doorbell interrupt pending for a vcpu */
 722 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
 723 {
 724         int thr;
 725         struct kvmppc_vcore *vc;
 726
 727         if (vcpu->arch.doorbell_request)
 728                 return true;
 729         /*
 730          * Ensure that the read of vcore->dpdes comes after the read
 731          * of vcpu->doorbell_request.  This barrier matches the
 732          * smp_wmb() in kvmppc_guest_entry_inject().
 733          */
 734         smp_rmb();
 735         vc = vcpu->arch.vcore;
 736         thr = vcpu->vcpu_id - vc->first_vcpuid;
 737         return !!(vc->dpdes & (1 << thr));
 738 }
 739
 740 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 741 {
 742         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 743                 return true;
 744         if ((!vcpu->arch.vcore->arch_compat) &&
 745             cpu_has_feature(CPU_FTR_ARCH_207S))
 746                 return true;
 747         return false;
 748 }
 749
 750 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 751                              unsigned long resource, unsigned long value1,
 752                              unsigned long value2)
 753 {
 754         switch (resource) {
 755         case H_SET_MODE_RESOURCE_SET_CIABR:
 756                 if (!kvmppc_power8_compatible(vcpu))
 757                         return H_P2;
 758                 if (value2)
 759                         return H_P4;
 760                 if (mflags)
 761                         return H_UNSUPPORTED_FLAG_START;
 762                 /* Guests can't breakpoint the hypervisor */
 763                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 764                         return H_P3;
 765                 vcpu->arch.ciabr  = value1;
 766                 return H_SUCCESS;
 767         case H_SET_MODE_RESOURCE_SET_DAWR:
 768                 if (!kvmppc_power8_compatible(vcpu))
 769                         return H_P2;
 770                 if (!ppc_breakpoint_available())
 771                         return H_P2;
 772                 if (mflags)
 773                         return H_UNSUPPORTED_FLAG_START;
 774                 if (value2 & DABRX_HYP)
 775                         return H_P4;
 776                 vcpu->arch.dawr  = value1;
 777                 vcpu->arch.dawrx = value2;
 778                 return H_SUCCESS;
 779         case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
 780                 /* KVM does not support mflags=2 (AIL=2) */
 781                 if (mflags != 0 && mflags != 3)
 782                         return H_UNSUPPORTED_FLAG_START;
 783                 return H_TOO_HARD;
 784         default:
 785                 return H_TOO_HARD;
 786         }
 787 }
 788
 789 /* Copy guest memory in place - must reside within a single memslot */
 790 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
 791                                   unsigned long len)
 792 {
 793         struct kvm_memory_slot *to_memslot = NULL;
 794         struct kvm_memory_slot *from_memslot = NULL;
 795         unsigned long to_addr, from_addr;
 796         int r;
 797
 798         /* Get HPA for from address */
 799         from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
 800         if (!from_memslot)
 801                 return -EFAULT;
 802         if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
 803                              << PAGE_SHIFT))
 804                 return -EINVAL;
 805         from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
 806         if (kvm_is_error_hva(from_addr))
 807                 return -EFAULT;
 808         from_addr |= (from & (PAGE_SIZE - 1));
 809
 810         /* Get HPA for to address */
 811         to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
 812         if (!to_memslot)
 813                 return -EFAULT;
 814         if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
 815                            << PAGE_SHIFT))
 816                 return -EINVAL;
 817         to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
 818         if (kvm_is_error_hva(to_addr))
 819                 return -EFAULT;
 820         to_addr |= (to & (PAGE_SIZE - 1));
 821
 822         /* Perform copy */
 823         r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
 824                              len);
 825         if (r)
 826                 return -EFAULT;
 827         mark_page_dirty(kvm, to >> PAGE_SHIFT);
 828         return 0;
 829 }
 830
 831 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
 832                                unsigned long dest, unsigned long src)
 833 {
 834         u64 pg_sz = SZ_4K;              /* 4K page size */
 835         u64 pg_mask = SZ_4K - 1;
 836         int ret;
 837
 838         /* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
 839         if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
 840                       H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
 841                 return H_PARAMETER;
 842
 843         /* dest (and src if copy_page flag set) must be page aligned */
 844         if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
 845                 return H_PARAMETER;
 846
 847         /* zero and/or copy the page as determined by the flags */
 848         if (flags & H_COPY_PAGE) {
 849                 ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
 850                 if (ret < 0)
 851                         return H_PARAMETER;
 852         } else if (flags & H_ZERO_PAGE) {
 853                 ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
 854                 if (ret < 0)
 855                         return H_PARAMETER;
 856         }
 857
 858         /* We can ignore the remaining flags */
 859
 860         return H_SUCCESS;
 861 }
 862
 863 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 864 {
 865         struct kvmppc_vcore *vcore = target->arch.vcore;
 866
 867         /*
 868          * We expect to have been called by the real mode handler
 869          * (kvmppc_rm_h_confer()) which would have directly returned
 870          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 871          * have useful work to do and should not confer) so we don't
 872          * recheck that here.
 873          */
 874
 875         spin_lock(&vcore->lock);
 876         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 877             vcore->vcore_state != VCORE_INACTIVE &&
 878             vcore->runner)
 879                 target = vcore->runner;
 880         spin_unlock(&vcore->lock);
 881
 882         return kvm_vcpu_yield_to(target);
 883 }
 884
 885 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 886 {
 887         int yield_count = 0;
 888         struct lppaca *lppaca;
 889
 890         spin_lock(&vcpu->arch.vpa_update_lock);
 891         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 892         if (lppaca)
 893                 yield_count = be32_to_cpu(lppaca->yield_count);
 894         spin_unlock(&vcpu->arch.vpa_update_lock);
 895         return yield_count;
 896 }
 897
 898 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 899 {
 900         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 901         unsigned long target, ret = H_SUCCESS;
 902         int yield_count;
 903         struct kvm_vcpu *tvcpu;
 904         int idx, rc;
 905
 906         if (req <= MAX_HCALL_OPCODE &&
 907             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 908                 return RESUME_HOST;
 909
 910         switch (req) {
 911         case H_CEDE:
 912                 break;
 913         case H_PROD:
 914                 target = kvmppc_get_gpr(vcpu, 4);
 915                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 916                 if (!tvcpu) {
 917                         ret = H_PARAMETER;
 918                         break;
 919                 }
 920                 tvcpu->arch.prodded = 1;
 921                 smp_mb();
 922                 if (tvcpu->arch.ceded)
 923                         kvmppc_fast_vcpu_kick_hv(tvcpu);
 924                 break;
 925         case H_CONFER:
 926                 target = kvmppc_get_gpr(vcpu, 4);
 927                 if (target == -1)
 928                         break;
 929                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 930                 if (!tvcpu) {
 931                         ret = H_PARAMETER;
 932                         break;
 933                 }
 934                 yield_count = kvmppc_get_gpr(vcpu, 5);
 935                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 936                         break;
 937                 kvm_arch_vcpu_yield_to(tvcpu);
 938                 break;
 939         case H_REGISTER_VPA:
 940                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 941                                         kvmppc_get_gpr(vcpu, 5),
 942                                         kvmppc_get_gpr(vcpu, 6));
 943                 break;
 944         case H_RTAS:
 945                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 946                         return RESUME_HOST;
 947
 948                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 949                 rc = kvmppc_rtas_hcall(vcpu);
 950                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 951
 952                 if (rc == -ENOENT)
 953                         return RESUME_HOST;
 954                 else if (rc == 0)
 955                         break;
 956
 957                 /* Send the error out to userspace via KVM_RUN */
 958                 return rc;
 959         case H_LOGICAL_CI_LOAD:
 960                 ret = kvmppc_h_logical_ci_load(vcpu);
 961                 if (ret == H_TOO_HARD)
 962                         return RESUME_HOST;
 963                 break;
 964         case H_LOGICAL_CI_STORE:
 965                 ret = kvmppc_h_logical_ci_store(vcpu);
 966                 if (ret == H_TOO_HARD)
 967                         return RESUME_HOST;
 968                 break;
 969         case H_SET_MODE:
 970                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 971                                         kvmppc_get_gpr(vcpu, 5),
 972                                         kvmppc_get_gpr(vcpu, 6),
 973                                         kvmppc_get_gpr(vcpu, 7));
 974                 if (ret == H_TOO_HARD)
 975                         return RESUME_HOST;
 976                 break;
 977         case H_XIRR:
 978         case H_CPPR:
 979         case H_EOI:
 980         case H_IPI:
 981         case H_IPOLL:
 982         case H_XIRR_X:
 983                 if (kvmppc_xics_enabled(vcpu)) {
 984                         if (xics_on_xive()) {
 985                                 ret = H_NOT_AVAILABLE;
 986                                 return RESUME_GUEST;
 987                         }
 988                         ret = kvmppc_xics_hcall(vcpu, req);
 989                         break;
 990                 }
 991                 return RESUME_HOST;
 992         case H_SET_DABR:
 993                 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
 994                 break;
 995         case H_SET_XDABR:
 996                 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
 997                                                 kvmppc_get_gpr(vcpu, 5));
 998                 break;
 999 #ifdef CONFIG_SPAPR_TCE_IOMMU
1000         case H_GET_TCE:
1001                 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1002                                                 kvmppc_get_gpr(vcpu, 5));
1003                 if (ret == H_TOO_HARD)
1004                         return RESUME_HOST;
1005                 break;
1006         case H_PUT_TCE:
1007                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1008                                                 kvmppc_get_gpr(vcpu, 5),
1009                                                 kvmppc_get_gpr(vcpu, 6));
1010                 if (ret == H_TOO_HARD)
1011                         return RESUME_HOST;
1012                 break;
1013         case H_PUT_TCE_INDIRECT:
1014                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1015                                                 kvmppc_get_gpr(vcpu, 5),
1016                                                 kvmppc_get_gpr(vcpu, 6),
1017                                                 kvmppc_get_gpr(vcpu, 7));
1018                 if (ret == H_TOO_HARD)
1019                         return RESUME_HOST;
1020                 break;
1021         case H_STUFF_TCE:
1022                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1023                                                 kvmppc_get_gpr(vcpu, 5),
1024                                                 kvmppc_get_gpr(vcpu, 6),
1025                                                 kvmppc_get_gpr(vcpu, 7));
1026                 if (ret == H_TOO_HARD)
1027                         return RESUME_HOST;
1028                 break;
1029 #endif
1030         case H_RANDOM:
1031                 if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1032                         ret = H_HARDWARE;
1033                 break;
1034
1035         case H_SET_PARTITION_TABLE:
1036                 ret = H_FUNCTION;
1037                 if (nesting_enabled(vcpu->kvm))
1038                         ret = kvmhv_set_partition_table(vcpu);
1039                 break;
1040         case H_ENTER_NESTED:
1041                 ret = H_FUNCTION;
1042                 if (!nesting_enabled(vcpu->kvm))
1043                         break;
1044                 ret = kvmhv_enter_nested_guest(vcpu);
1045                 if (ret == H_INTERRUPT) {
1046                         kvmppc_set_gpr(vcpu, 3, 0);
1047                         vcpu->arch.hcall_needed = 0;
1048                         return -EINTR;
1049                 } else if (ret == H_TOO_HARD) {
1050                         kvmppc_set_gpr(vcpu, 3, 0);
1051                         vcpu->arch.hcall_needed = 0;
1052                         return RESUME_HOST;
1053                 }
1054                 break;
1055         case H_TLB_INVALIDATE:
1056                 ret = H_FUNCTION;
1057                 if (nesting_enabled(vcpu->kvm))
1058                         ret = kvmhv_do_nested_tlbie(vcpu);
1059                 break;
1060         case H_COPY_TOFROM_GUEST:
1061                 ret = H_FUNCTION;
1062                 if (nesting_enabled(vcpu->kvm))
1063                         ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1064                 break;
1065         case H_PAGE_INIT:
1066                 ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1067                                          kvmppc_get_gpr(vcpu, 5),
1068                                          kvmppc_get_gpr(vcpu, 6));
1069                 break;
1070         case H_SVM_PAGE_IN:
1071                 ret = H_UNSUPPORTED;
1072                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1073                         ret = kvmppc_h_svm_page_in(vcpu->kvm,
1074                                                    kvmppc_get_gpr(vcpu, 4),
1075                                                    kvmppc_get_gpr(vcpu, 5),
1076                                                    kvmppc_get_gpr(vcpu, 6));
1077                 break;
1078         case H_SVM_PAGE_OUT:
1079                 ret = H_UNSUPPORTED;
1080                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1081                         ret = kvmppc_h_svm_page_out(vcpu->kvm,
1082                                                     kvmppc_get_gpr(vcpu, 4),
1083                                                     kvmppc_get_gpr(vcpu, 5),
1084                                                     kvmppc_get_gpr(vcpu, 6));
1085                 break;
1086         case H_SVM_INIT_START:
1087                 ret = H_UNSUPPORTED;
1088                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1089                         ret = kvmppc_h_svm_init_start(vcpu->kvm);
1090                 break;
1091         case H_SVM_INIT_DONE:
1092                 ret = H_UNSUPPORTED;
1093                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1094                         ret = kvmppc_h_svm_init_done(vcpu->kvm);
1095                 break;
1096         case H_SVM_INIT_ABORT:
1097                 ret = H_UNSUPPORTED;
1098                 if (kvmppc_get_srr1(vcpu) & MSR_S)
1099                         ret = kvmppc_h_svm_init_abort(vcpu->kvm);
1100                 break;
1101
1102         default:
1103                 return RESUME_HOST;
1104         }
1105         kvmppc_set_gpr(vcpu, 3, ret);
1106         vcpu->arch.hcall_needed = 0;
1107         return RESUME_GUEST;
1108 }
1109
1110 /*
1111  * Handle H_CEDE in the nested virtualization case where we haven't
1112  * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1113  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1114  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1115  */
1116 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1117 {
1118         vcpu->arch.shregs.msr |= MSR_EE;
1119         vcpu->arch.ceded = 1;
1120         smp_mb();
1121         if (vcpu->arch.prodded) {
1122                 vcpu->arch.prodded = 0;
1123                 smp_mb();
1124                 vcpu->arch.ceded = 0;
1125         }
1126 }
1127
1128 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1129 {
1130         switch (cmd) {
1131         case H_CEDE:
1132         case H_PROD:
1133         case H_CONFER:
1134         case H_REGISTER_VPA:
1135         case H_SET_MODE:
1136         case H_LOGICAL_CI_LOAD:
1137         case H_LOGICAL_CI_STORE:
1138 #ifdef CONFIG_KVM_XICS
1139         case H_XIRR:
1140         case H_CPPR:
1141         case H_EOI:
1142         case H_IPI:
1143         case H_IPOLL:
1144         case H_XIRR_X:
1145 #endif
1146         case H_PAGE_INIT:
1147                 return 1;
1148         }
1149
1150         /* See if it's in the real-mode table */
1151         return kvmppc_hcall_impl_hv_realmode(cmd);
1152 }
1153
1154 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
1155                                         struct kvm_vcpu *vcpu)
1156 {
1157         u32 last_inst;
1158
1159         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1160                                         EMULATE_DONE) {
1161                 /*
1162                  * Fetch failed, so return to guest and
1163                  * try executing it again.
1164                  */
1165                 return RESUME_GUEST;
1166         }
1167
1168         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1169                 run->exit_reason = KVM_EXIT_DEBUG;
1170                 run->debug.arch.address = kvmppc_get_pc(vcpu);
1171                 return RESUME_HOST;
1172         } else {
1173                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1174                 return RESUME_GUEST;
1175         }
1176 }
1177
1178 static void do_nothing(void *x)
1179 {
1180 }
1181
1182 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1183 {
1184         int thr, cpu, pcpu, nthreads;
1185         struct kvm_vcpu *v;
1186         unsigned long dpdes;
1187
1188         nthreads = vcpu->kvm->arch.emul_smt_mode;
1189         dpdes = 0;
1190         cpu = vcpu->vcpu_id & ~(nthreads - 1);
1191         for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1192                 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1193                 if (!v)
1194                         continue;
1195                 /*
1196                  * If the vcpu is currently running on a physical cpu thread,
1197                  * interrupt it in order to pull it out of the guest briefly,
1198                  * which will update its vcore->dpdes value.
1199                  */
1200                 pcpu = READ_ONCE(v->cpu);
1201                 if (pcpu >= 0)
1202                         smp_call_function_single(pcpu, do_nothing, NULL, 1);
1203                 if (kvmppc_doorbell_pending(v))
1204                         dpdes |= 1 << thr;
1205         }
1206         return dpdes;
1207 }
1208
1209 /*
1210  * On POWER9, emulate doorbell-related instructions in order to
1211  * give the guest the illusion of running on a multi-threaded core.
1212  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1213  * and mfspr DPDES.
1214  */
1215 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1216 {
1217         u32 inst, rb, thr;
1218         unsigned long arg;
1219         struct kvm *kvm = vcpu->kvm;
1220         struct kvm_vcpu *tvcpu;
1221
1222         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1223                 return RESUME_GUEST;
1224         if (get_op(inst) != 31)
1225                 return EMULATE_FAIL;
1226         rb = get_rb(inst);
1227         thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1228         switch (get_xop(inst)) {
1229         case OP_31_XOP_MSGSNDP:
1230                 arg = kvmppc_get_gpr(vcpu, rb);
1231                 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1232                         break;
1233                 arg &= 0x3f;
1234                 if (arg >= kvm->arch.emul_smt_mode)
1235                         break;
1236                 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1237                 if (!tvcpu)
1238                         break;
1239                 if (!tvcpu->arch.doorbell_request) {
1240                         tvcpu->arch.doorbell_request = 1;
1241                         kvmppc_fast_vcpu_kick_hv(tvcpu);
1242                 }
1243                 break;
1244         case OP_31_XOP_MSGCLRP:
1245                 arg = kvmppc_get_gpr(vcpu, rb);
1246                 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1247                         break;
1248                 vcpu->arch.vcore->dpdes = 0;
1249                 vcpu->arch.doorbell_request = 0;
1250                 break;
1251         case OP_31_XOP_MFSPR:
1252                 switch (get_sprn(inst)) {
1253                 case SPRN_TIR:
1254                         arg = thr;
1255                         break;
1256                 case SPRN_DPDES:
1257                         arg = kvmppc_read_dpdes(vcpu);
1258                         break;
1259                 default:
1260                         return EMULATE_FAIL;
1261                 }
1262                 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1263                 break;
1264         default:
1265                 return EMULATE_FAIL;
1266         }
1267         kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1268         return RESUME_GUEST;
1269 }
1270
1271 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1272                                  struct task_struct *tsk)
1273 {
1274         int r = RESUME_HOST;
1275
1276         vcpu->stat.sum_exits++;
1277
1278         /*
1279          * This can happen if an interrupt occurs in the last stages
1280          * of guest entry or the first stages of guest exit (i.e. after
1281          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1282          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1283          * That can happen due to a bug, or due to a machine check
1284          * occurring at just the wrong time.
1285          */
1286         if (vcpu->arch.shregs.msr & MSR_HV) {
1287                 printk(KERN_EMERG "KVM trap in HV mode!\n");
1288                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1289                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1290                         vcpu->arch.shregs.msr);
1291                 kvmppc_dump_regs(vcpu);
1292                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1293                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1294                 return RESUME_HOST;
1295         }
1296         run->exit_reason = KVM_EXIT_UNKNOWN;
1297         run->ready_for_interrupt_injection = 1;
1298         switch (vcpu->arch.trap) {
1299         /* We're good on these - the host merely wanted to get our attention */
1300         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1301                 vcpu->stat.dec_exits++;
1302                 r = RESUME_GUEST;
1303                 break;
1304         case BOOK3S_INTERRUPT_EXTERNAL:
1305         case BOOK3S_INTERRUPT_H_DOORBELL:
1306         case BOOK3S_INTERRUPT_H_VIRT:
1307                 vcpu->stat.ext_intr_exits++;
1308                 r = RESUME_GUEST;
1309                 break;
1310         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1311         case BOOK3S_INTERRUPT_HMI:
1312         case BOOK3S_INTERRUPT_PERFMON:
1313         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1314                 r = RESUME_GUEST;
1315                 break;
1316         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1317                 /* Print the MCE event to host console. */
1318                 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1319
1320                 /*
1321                  * If the guest can do FWNMI, exit to userspace so it can
1322                  * deliver a FWNMI to the guest.
1323                  * Otherwise we synthesize a machine check for the guest
1324                  * so that it knows that the machine check occurred.
1325                  */
1326                 if (!vcpu->kvm->arch.fwnmi_enabled) {
1327                         ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1328                         kvmppc_core_queue_machine_check(vcpu, flags);
1329                         r = RESUME_GUEST;
1330                         break;
1331                 }
1332
1333                 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1334                 run->exit_reason = KVM_EXIT_NMI;
1335                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1336                 /* Clear out the old NMI status from run->flags */
1337                 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1338                 /* Now set the NMI status */
1339                 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1340                         run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1341                 else
1342                         run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1343
1344                 r = RESUME_HOST;
1345                 break;
1346         case BOOK3S_INTERRUPT_PROGRAM:
1347         {
1348                 ulong flags;
1349                 /*
1350                  * Normally program interrupts are delivered directly
1351                  * to the guest by the hardware, but we can get here
1352                  * as a result of a hypervisor emulation interrupt
1353                  * (e40) getting turned into a 700 by BML RTAS.
1354                  */
1355                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1356                 kvmppc_core_queue_program(vcpu, flags);
1357                 r = RESUME_GUEST;
1358                 break;
1359         }
1360         case BOOK3S_INTERRUPT_SYSCALL:
1361         {
1362                 /* hcall - punt to userspace */
1363                 int i;
1364
1365                 /* hypercall with MSR_PR has already been handled in rmode,
1366                  * and never reaches here.
1367                  */
1368
1369                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1370                 for (i = 0; i < 9; ++i)
1371                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1372                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1373                 vcpu->arch.hcall_needed = 1;
1374                 r = RESUME_HOST;
1375                 break;
1376         }
1377         /*
1378          * We get these next two if the guest accesses a page which it thinks
1379          * it has mapped but which is not actually present, either because
1380          * it is for an emulated I/O device or because the corresonding
1381          * host page has been paged out.  Any other HDSI/HISI interrupts
1382          * have been handled already.
1383          */
1384         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1385                 r = RESUME_PAGE_FAULT;
1386                 break;
1387         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1388                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1389                 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1390                         DSISR_SRR1_MATCH_64S;
1391                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1392                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1393                 r = RESUME_PAGE_FAULT;
1394                 break;
1395         /*
1396          * This occurs if the guest executes an illegal instruction.
1397          * If the guest debug is disabled, generate a program interrupt
1398          * to the guest. If guest debug is enabled, we need to check
1399          * whether the instruction is a software breakpoint instruction.
1400          * Accordingly return to Guest or Host.
1401          */
1402         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1403                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1404                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1405                                 swab32(vcpu->arch.emul_inst) :
1406                                 vcpu->arch.emul_inst;
1407                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1408                         r = kvmppc_emulate_debug_inst(run, vcpu);
1409                 } else {
1410                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1411                         r = RESUME_GUEST;
1412                 }
1413                 break;
1414         /*
1415          * This occurs if the guest (kernel or userspace), does something that
1416          * is prohibited by HFSCR.
1417          * On POWER9, this could be a doorbell instruction that we need
1418          * to emulate.
1419          * Otherwise, we just generate a program interrupt to the guest.
1420          */
1421         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1422                 r = EMULATE_FAIL;
1423                 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1424                     cpu_has_feature(CPU_FTR_ARCH_300))
1425                         r = kvmppc_emulate_doorbell_instr(vcpu);
1426                 if (r == EMULATE_FAIL) {
1427                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1428                         r = RESUME_GUEST;
1429                 }
1430                 break;
1431
1432 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1433         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1434                 /*
1435                  * This occurs for various TM-related instructions that
1436                  * we need to emulate on POWER9 DD2.2.  We have already
1437                  * handled the cases where the guest was in real-suspend
1438                  * mode and was transitioning to transactional state.
1439                  */
1440                 r = kvmhv_p9_tm_emulation(vcpu);
1441                 break;
1442 #endif
1443
1444         case BOOK3S_INTERRUPT_HV_RM_HARD:
1445                 r = RESUME_PASSTHROUGH;
1446                 break;
1447         default:
1448                 kvmppc_dump_regs(vcpu);
1449                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1450                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1451                         vcpu->arch.shregs.msr);
1452                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1453                 r = RESUME_HOST;
1454                 break;
1455         }
1456
1457         return r;
1458 }
1459
1460 static int kvmppc_handle_nested_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
1461 {
1462         int r;
1463         int srcu_idx;
1464
1465         vcpu->stat.sum_exits++;
1466
1467         /*
1468          * This can happen if an interrupt occurs in the last stages
1469          * of guest entry or the first stages of guest exit (i.e. after
1470          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1471          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1472          * That can happen due to a bug, or due to a machine check
1473          * occurring at just the wrong time.
1474          */
1475         if (vcpu->arch.shregs.msr & MSR_HV) {
1476                 pr_emerg("KVM trap in HV mode while nested!\n");
1477                 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1478                          vcpu->arch.trap, kvmppc_get_pc(vcpu),
1479                          vcpu->arch.shregs.msr);
1480                 kvmppc_dump_regs(vcpu);
1481                 return RESUME_HOST;
1482         }
1483         switch (vcpu->arch.trap) {
1484         /* We're good on these - the host merely wanted to get our attention */
1485         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1486                 vcpu->stat.dec_exits++;
1487                 r = RESUME_GUEST;
1488                 break;
1489         case BOOK3S_INTERRUPT_EXTERNAL:
1490                 vcpu->stat.ext_intr_exits++;
1491                 r = RESUME_HOST;
1492                 break;
1493         case BOOK3S_INTERRUPT_H_DOORBELL:
1494         case BOOK3S_INTERRUPT_H_VIRT:
1495                 vcpu->stat.ext_intr_exits++;
1496                 r = RESUME_GUEST;
1497                 break;
1498         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1499         case BOOK3S_INTERRUPT_HMI:
1500         case BOOK3S_INTERRUPT_PERFMON:
1501         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1502                 r = RESUME_GUEST;
1503                 break;
1504         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1505                 /* Pass the machine check to the L1 guest */
1506                 r = RESUME_HOST;
1507                 /* Print the MCE event to host console. */
1508                 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1509                 break;
1510         /*
1511          * We get these next two if the guest accesses a page which it thinks
1512          * it has mapped but which is not actually present, either because
1513          * it is for an emulated I/O device or because the corresonding
1514          * host page has been paged out.
1515          */
1516         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1517                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1518                 r = kvmhv_nested_page_fault(run, vcpu);
1519                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1520                 break;
1521         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1522                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1523                 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1524                                          DSISR_SRR1_MATCH_64S;
1525                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1526                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1527                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1528                 r = kvmhv_nested_page_fault(run, vcpu);
1529                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1530                 break;
1531
1532 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1533         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1534                 /*
1535                  * This occurs for various TM-related instructions that
1536                  * we need to emulate on POWER9 DD2.2.  We have already
1537                  * handled the cases where the guest was in real-suspend
1538                  * mode and was transitioning to transactional state.
1539                  */
1540                 r = kvmhv_p9_tm_emulation(vcpu);
1541                 break;
1542 #endif
1543
1544         case BOOK3S_INTERRUPT_HV_RM_HARD:
1545                 vcpu->arch.trap = 0;
1546                 r = RESUME_GUEST;
1547                 if (!xics_on_xive())
1548                         kvmppc_xics_rm_complete(vcpu, 0);
1549                 break;
1550         default:
1551                 r = RESUME_HOST;
1552                 break;
1553         }
1554
1555         return r;
1556 }
1557
1558 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1559                                             struct kvm_sregs *sregs)
1560 {
1561         int i;
1562
1563         memset(sregs, 0, sizeof(struct kvm_sregs));
1564         sregs->pvr = vcpu->arch.pvr;
1565         for (i = 0; i < vcpu->arch.slb_max; i++) {
1566                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1567                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1568         }
1569
1570         return 0;
1571 }
1572
1573 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1574                                             struct kvm_sregs *sregs)
1575 {
1576         int i, j;
1577
1578         /* Only accept the same PVR as the host's, since we can't spoof it */
1579         if (sregs->pvr != vcpu->arch.pvr)
1580                 return -EINVAL;
1581
1582         j = 0;
1583         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1584                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1585                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1586                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1587                         ++j;
1588                 }
1589         }
1590         vcpu->arch.slb_max = j;
1591
1592         return 0;
1593 }
1594
1595 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1596                 bool preserve_top32)
1597 {
1598         struct kvm *kvm = vcpu->kvm;
1599         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1600         u64 mask;
1601
1602         spin_lock(&vc->lock);
1603         /*
1604          * If ILE (interrupt little-endian) has changed, update the
1605          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1606          */
1607         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1608                 struct kvm_vcpu *vcpu;
1609                 int i;
1610
1611                 kvm_for_each_vcpu(i, vcpu, kvm) {
1612                         if (vcpu->arch.vcore != vc)
1613                                 continue;
1614                         if (new_lpcr & LPCR_ILE)
1615                                 vcpu->arch.intr_msr |= MSR_LE;
1616                         else
1617                                 vcpu->arch.intr_msr &= ~MSR_LE;
1618                 }
1619         }
1620
1621         /*
1622          * Userspace can only modify DPFD (default prefetch depth),
1623          * ILE (interrupt little-endian) and TC (translation control).
1624          * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1625          */
1626         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1627         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1628                 mask |= LPCR_AIL;
1629         /*
1630          * On POWER9, allow userspace to enable large decrementer for the
1631          * guest, whether or not the host has it enabled.
1632          */
1633         if (cpu_has_feature(CPU_FTR_ARCH_300))
1634                 mask |= LPCR_LD;
1635
1636         /* Broken 32-bit version of LPCR must not clear top bits */
1637         if (preserve_top32)
1638                 mask &= 0xFFFFFFFF;
1639         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1640         spin_unlock(&vc->lock);
1641 }
1642
1643 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1644                                  union kvmppc_one_reg *val)
1645 {
1646         int r = 0;
1647         long int i;
1648
1649         switch (id) {
1650         case KVM_REG_PPC_DEBUG_INST:
1651                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1652                 break;
1653         case KVM_REG_PPC_HIOR:
1654                 *val = get_reg_val(id, 0);
1655                 break;
1656         case KVM_REG_PPC_DABR:
1657                 *val = get_reg_val(id, vcpu->arch.dabr);
1658                 break;
1659         case KVM_REG_PPC_DABRX:
1660                 *val = get_reg_val(id, vcpu->arch.dabrx);
1661                 break;
1662         case KVM_REG_PPC_DSCR:
1663                 *val = get_reg_val(id, vcpu->arch.dscr);
1664                 break;
1665         case KVM_REG_PPC_PURR:
1666                 *val = get_reg_val(id, vcpu->arch.purr);
1667                 break;
1668         case KVM_REG_PPC_SPURR:
1669                 *val = get_reg_val(id, vcpu->arch.spurr);
1670                 break;
1671         case KVM_REG_PPC_AMR:
1672                 *val = get_reg_val(id, vcpu->arch.amr);
1673                 break;
1674         case KVM_REG_PPC_UAMOR:
1675                 *val = get_reg_val(id, vcpu->arch.uamor);
1676                 break;
1677         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1678                 i = id - KVM_REG_PPC_MMCR0;
1679                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1680                 break;
1681         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1682                 i = id - KVM_REG_PPC_PMC1;
1683                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1684                 break;
1685         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1686                 i = id - KVM_REG_PPC_SPMC1;
1687                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1688                 break;
1689         case KVM_REG_PPC_SIAR:
1690                 *val = get_reg_val(id, vcpu->arch.siar);
1691                 break;
1692         case KVM_REG_PPC_SDAR:
1693                 *val = get_reg_val(id, vcpu->arch.sdar);
1694                 break;
1695         case KVM_REG_PPC_SIER:
1696                 *val = get_reg_val(id, vcpu->arch.sier);
1697                 break;
1698         case KVM_REG_PPC_IAMR:
1699                 *val = get_reg_val(id, vcpu->arch.iamr);
1700                 break;
1701         case KVM_REG_PPC_PSPB:
1702                 *val = get_reg_val(id, vcpu->arch.pspb);
1703                 break;
1704         case KVM_REG_PPC_DPDES:
1705                 /*
1706                  * On POWER9, where we are emulating msgsndp etc.,
1707                  * we return 1 bit for each vcpu, which can come from
1708                  * either vcore->dpdes or doorbell_request.
1709                  * On POWER8, doorbell_request is 0.
1710                  */
1711                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes |
1712                                    vcpu->arch.doorbell_request);
1713                 break;
1714         case KVM_REG_PPC_VTB:
1715                 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1716                 break;
1717         case KVM_REG_PPC_DAWR:
1718                 *val = get_reg_val(id, vcpu->arch.dawr);
1719                 break;
1720         case KVM_REG_PPC_DAWRX:
1721                 *val = get_reg_val(id, vcpu->arch.dawrx);
1722                 break;
1723         case KVM_REG_PPC_CIABR:
1724                 *val = get_reg_val(id, vcpu->arch.ciabr);
1725                 break;
1726         case KVM_REG_PPC_CSIGR:
1727                 *val = get_reg_val(id, vcpu->arch.csigr);
1728                 break;
1729         case KVM_REG_PPC_TACR:
1730                 *val = get_reg_val(id, vcpu->arch.tacr);
1731                 break;
1732         case KVM_REG_PPC_TCSCR:
1733                 *val = get_reg_val(id, vcpu->arch.tcscr);
1734                 break;
1735         case KVM_REG_PPC_PID:
1736                 *val = get_reg_val(id, vcpu->arch.pid);
1737                 break;
1738         case KVM_REG_PPC_ACOP:
1739                 *val = get_reg_val(id, vcpu->arch.acop);
1740                 break;
1741         case KVM_REG_PPC_WORT:
1742                 *val = get_reg_val(id, vcpu->arch.wort);
1743                 break;
1744         case KVM_REG_PPC_TIDR:
1745                 *val = get_reg_val(id, vcpu->arch.tid);
1746                 break;
1747         case KVM_REG_PPC_PSSCR:
1748                 *val = get_reg_val(id, vcpu->arch.psscr);
1749                 break;
1750         case KVM_REG_PPC_VPA_ADDR:
1751                 spin_lock(&vcpu->arch.vpa_update_lock);
1752                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1753                 spin_unlock(&vcpu->arch.vpa_update_lock);
1754                 break;
1755         case KVM_REG_PPC_VPA_SLB:
1756                 spin_lock(&vcpu->arch.vpa_update_lock);
1757                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1758                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1759                 spin_unlock(&vcpu->arch.vpa_update_lock);
1760                 break;
1761         case KVM_REG_PPC_VPA_DTL:
1762                 spin_lock(&vcpu->arch.vpa_update_lock);
1763                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1764                 val->vpaval.length = vcpu->arch.dtl.len;
1765                 spin_unlock(&vcpu->arch.vpa_update_lock);
1766                 break;
1767         case KVM_REG_PPC_TB_OFFSET:
1768                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1769                 break;
1770         case KVM_REG_PPC_LPCR:
1771         case KVM_REG_PPC_LPCR_64:
1772                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1773                 break;
1774         case KVM_REG_PPC_PPR:
1775                 *val = get_reg_val(id, vcpu->arch.ppr);
1776                 break;
1777 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1778         case KVM_REG_PPC_TFHAR:
1779                 *val = get_reg_val(id, vcpu->arch.tfhar);
1780                 break;
1781         case KVM_REG_PPC_TFIAR:
1782                 *val = get_reg_val(id, vcpu->arch.tfiar);
1783                 break;
1784         case KVM_REG_PPC_TEXASR:
1785                 *val = get_reg_val(id, vcpu->arch.texasr);
1786                 break;
1787         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1788                 i = id - KVM_REG_PPC_TM_GPR0;
1789                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1790                 break;
1791         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1792         {
1793                 int j;
1794                 i = id - KVM_REG_PPC_TM_VSR0;
1795                 if (i < 32)
1796                         for (j = 0; j < TS_FPRWIDTH; j++)
1797                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1798                 else {
1799                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1800                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1801                         else
1802                                 r = -ENXIO;
1803                 }
1804                 break;
1805         }
1806         case KVM_REG_PPC_TM_CR:
1807                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1808                 break;
1809         case KVM_REG_PPC_TM_XER:
1810                 *val = get_reg_val(id, vcpu->arch.xer_tm);
1811                 break;
1812         case KVM_REG_PPC_TM_LR:
1813                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1814                 break;
1815         case KVM_REG_PPC_TM_CTR:
1816                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1817                 break;
1818         case KVM_REG_PPC_TM_FPSCR:
1819                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1820                 break;
1821         case KVM_REG_PPC_TM_AMR:
1822                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1823                 break;
1824         case KVM_REG_PPC_TM_PPR:
1825                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1826                 break;
1827         case KVM_REG_PPC_TM_VRSAVE:
1828                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1829                 break;
1830         case KVM_REG_PPC_TM_VSCR:
1831                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1832                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1833                 else
1834                         r = -ENXIO;
1835                 break;
1836         case KVM_REG_PPC_TM_DSCR:
1837                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1838                 break;
1839         case KVM_REG_PPC_TM_TAR:
1840                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1841                 break;
1842 #endif
1843         case KVM_REG_PPC_ARCH_COMPAT:
1844                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1845                 break;
1846         case KVM_REG_PPC_DEC_EXPIRY:
1847                 *val = get_reg_val(id, vcpu->arch.dec_expires +
1848                                    vcpu->arch.vcore->tb_offset);
1849                 break;
1850         case KVM_REG_PPC_ONLINE:
1851                 *val = get_reg_val(id, vcpu->arch.online);
1852                 break;
1853         case KVM_REG_PPC_PTCR:
1854                 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1855                 break;
1856         default:
1857                 r = -EINVAL;
1858                 break;
1859         }
1860
1861         return r;
1862 }
1863
1864 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1865                                  union kvmppc_one_reg *val)
1866 {
1867         int r = 0;
1868         long int i;
1869         unsigned long addr, len;
1870
1871         switch (id) {
1872         case KVM_REG_PPC_HIOR:
1873                 /* Only allow this to be set to zero */
1874                 if (set_reg_val(id, *val))
1875                         r = -EINVAL;
1876                 break;
1877         case KVM_REG_PPC_DABR:
1878                 vcpu->arch.dabr = set_reg_val(id, *val);
1879                 break;
1880         case KVM_REG_PPC_DABRX:
1881                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1882                 break;
1883         case KVM_REG_PPC_DSCR:
1884                 vcpu->arch.dscr = set_reg_val(id, *val);
1885                 break;
1886         case KVM_REG_PPC_PURR:
1887                 vcpu->arch.purr = set_reg_val(id, *val);
1888                 break;
1889         case KVM_REG_PPC_SPURR:
1890                 vcpu->arch.spurr = set_reg_val(id, *val);
1891                 break;
1892         case KVM_REG_PPC_AMR:
1893                 vcpu->arch.amr = set_reg_val(id, *val);
1894                 break;
1895         case KVM_REG_PPC_UAMOR:
1896                 vcpu->arch.uamor = set_reg_val(id, *val);
1897                 break;
1898         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1899                 i = id - KVM_REG_PPC_MMCR0;
1900                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1901                 break;
1902         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1903                 i = id - KVM_REG_PPC_PMC1;
1904                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1905                 break;
1906         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1907                 i = id - KVM_REG_PPC_SPMC1;
1908                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1909                 break;
1910         case KVM_REG_PPC_SIAR:
1911                 vcpu->arch.siar = set_reg_val(id, *val);
1912                 break;
1913         case KVM_REG_PPC_SDAR:
1914                 vcpu->arch.sdar = set_reg_val(id, *val);
1915                 break;
1916         case KVM_REG_PPC_SIER:
1917                 vcpu->arch.sier = set_reg_val(id, *val);
1918                 break;
1919         case KVM_REG_PPC_IAMR:
1920                 vcpu->arch.iamr = set_reg_val(id, *val);
1921                 break;
1922         case KVM_REG_PPC_PSPB:
1923                 vcpu->arch.pspb = set_reg_val(id, *val);
1924                 break;
1925         case KVM_REG_PPC_DPDES:
1926                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1927                 break;
1928         case KVM_REG_PPC_VTB:
1929                 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1930                 break;
1931         case KVM_REG_PPC_DAWR:
1932                 vcpu->arch.dawr = set_reg_val(id, *val);
1933                 break;
1934         case KVM_REG_PPC_DAWRX:
1935                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1936                 break;
1937         case KVM_REG_PPC_CIABR:
1938                 vcpu->arch.ciabr = set_reg_val(id, *val);
1939                 /* Don't allow setting breakpoints in hypervisor code */
1940                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1941                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1942                 break;
1943         case KVM_REG_PPC_CSIGR:
1944                 vcpu->arch.csigr = set_reg_val(id, *val);
1945                 break;
1946         case KVM_REG_PPC_TACR:
1947                 vcpu->arch.tacr = set_reg_val(id, *val);
1948                 break;
1949         case KVM_REG_PPC_TCSCR:
1950                 vcpu->arch.tcscr = set_reg_val(id, *val);
1951                 break;
1952         case KVM_REG_PPC_PID:
1953                 vcpu->arch.pid = set_reg_val(id, *val);
1954                 break;
1955         case KVM_REG_PPC_ACOP:
1956                 vcpu->arch.acop = set_reg_val(id, *val);
1957                 break;
1958         case KVM_REG_PPC_WORT:
1959                 vcpu->arch.wort = set_reg_val(id, *val);
1960                 break;
1961         case KVM_REG_PPC_TIDR:
1962                 vcpu->arch.tid = set_reg_val(id, *val);
1963                 break;
1964         case KVM_REG_PPC_PSSCR:
1965                 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1966                 break;
1967         case KVM_REG_PPC_VPA_ADDR:
1968                 addr = set_reg_val(id, *val);
1969                 r = -EINVAL;
1970                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1971                               vcpu->arch.dtl.next_gpa))
1972                         break;
1973                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1974                 break;
1975         case KVM_REG_PPC_VPA_SLB:
1976                 addr = val->vpaval.addr;
1977                 len = val->vpaval.length;
1978                 r = -EINVAL;
1979                 if (addr && !vcpu->arch.vpa.next_gpa)
1980                         break;
1981                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1982                 break;
1983         case KVM_REG_PPC_VPA_DTL:
1984                 addr = val->vpaval.addr;
1985                 len = val->vpaval.length;
1986                 r = -EINVAL;
1987                 if (addr && (len < sizeof(struct dtl_entry) ||
1988                              !vcpu->arch.vpa.next_gpa))
1989                         break;
1990                 len -= len % sizeof(struct dtl_entry);
1991                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1992                 break;
1993         case KVM_REG_PPC_TB_OFFSET:
1994                 /* round up to multiple of 2^24 */
1995                 vcpu->arch.vcore->tb_offset =
1996                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1997                 break;
1998         case KVM_REG_PPC_LPCR:
1999                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
2000                 break;
2001         case KVM_REG_PPC_LPCR_64:
2002                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
2003                 break;
2004         case KVM_REG_PPC_PPR:
2005                 vcpu->arch.ppr = set_reg_val(id, *val);
2006                 break;
2007 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2008         case KVM_REG_PPC_TFHAR:
2009                 vcpu->arch.tfhar = set_reg_val(id, *val);
2010                 break;
2011         case KVM_REG_PPC_TFIAR:
2012                 vcpu->arch.tfiar = set_reg_val(id, *val);
2013                 break;
2014         case KVM_REG_PPC_TEXASR:
2015                 vcpu->arch.texasr = set_reg_val(id, *val);
2016                 break;
2017         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2018                 i = id - KVM_REG_PPC_TM_GPR0;
2019                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2020                 break;
2021         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2022         {
2023                 int j;
2024                 i = id - KVM_REG_PPC_TM_VSR0;
2025                 if (i < 32)
2026                         for (j = 0; j < TS_FPRWIDTH; j++)
2027                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2028                 else
2029                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
2030                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
2031                         else
2032                                 r = -ENXIO;
2033                 break;
2034         }
2035         case KVM_REG_PPC_TM_CR:
2036                 vcpu->arch.cr_tm = set_reg_val(id, *val);
2037                 break;
2038         case KVM_REG_PPC_TM_XER:
2039                 vcpu->arch.xer_tm = set_reg_val(id, *val);
2040                 break;
2041         case KVM_REG_PPC_TM_LR:
2042                 vcpu->arch.lr_tm = set_reg_val(id, *val);
2043                 break;
2044         case KVM_REG_PPC_TM_CTR:
2045                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
2046                 break;
2047         case KVM_REG_PPC_TM_FPSCR:
2048                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2049                 break;
2050         case KVM_REG_PPC_TM_AMR:
2051                 vcpu->arch.amr_tm = set_reg_val(id, *val);
2052                 break;
2053         case KVM_REG_PPC_TM_PPR:
2054                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
2055                 break;
2056         case KVM_REG_PPC_TM_VRSAVE:
2057                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2058                 break;
2059         case KVM_REG_PPC_TM_VSCR:
2060                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2061                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2062                 else
2063                         r = - ENXIO;
2064                 break;
2065         case KVM_REG_PPC_TM_DSCR:
2066                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
2067                 break;
2068         case KVM_REG_PPC_TM_TAR:
2069                 vcpu->arch.tar_tm = set_reg_val(id, *val);
2070                 break;
2071 #endif
2072         case KVM_REG_PPC_ARCH_COMPAT:
2073                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2074                 break;
2075         case KVM_REG_PPC_DEC_EXPIRY:
2076                 vcpu->arch.dec_expires = set_reg_val(id, *val) -
2077                         vcpu->arch.vcore->tb_offset;
2078                 break;
2079         case KVM_REG_PPC_ONLINE:
2080                 i = set_reg_val(id, *val);
2081                 if (i && !vcpu->arch.online)
2082                         atomic_inc(&vcpu->arch.vcore->online_count);
2083                 else if (!i && vcpu->arch.online)
2084                         atomic_dec(&vcpu->arch.vcore->online_count);
2085                 vcpu->arch.online = i;
2086                 break;
2087         case KVM_REG_PPC_PTCR:
2088                 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2089                 break;
2090         default:
2091                 r = -EINVAL;
2092                 break;
2093         }
2094
2095         return r;
2096 }
2097
2098 /*
2099  * On POWER9, threads are independent and can be in different partitions.
2100  * Therefore we consider each thread to be a subcore.
2101  * There is a restriction that all threads have to be in the same
2102  * MMU mode (radix or HPT), unfortunately, but since we only support
2103  * HPT guests on a HPT host so far, that isn't an impediment yet.
2104  */
2105 static int threads_per_vcore(struct kvm *kvm)
2106 {
2107         if (kvm->arch.threads_indep)
2108                 return 1;
2109         return threads_per_subcore;
2110 }
2111
2112 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2113 {
2114         struct kvmppc_vcore *vcore;
2115
2116         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2117
2118         if (vcore == NULL)
2119                 return NULL;
2120
2121         spin_lock_init(&vcore->lock);
2122         spin_lock_init(&vcore->stoltb_lock);
2123         rcuwait_init(&vcore->wait);
2124         vcore->preempt_tb = TB_NIL;
2125         vcore->lpcr = kvm->arch.lpcr;
2126         vcore->first_vcpuid = id;
2127         vcore->kvm = kvm;
2128         INIT_LIST_HEAD(&vcore->preempt_list);
2129
2130         return vcore;
2131 }
2132
2133 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2134 static struct debugfs_timings_element {
2135         const char *name;
2136         size_t offset;
2137 } timings[] = {
2138         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
2139         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
2140         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
2141         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
2142         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
2143 };
2144
2145 #define N_TIMINGS       (ARRAY_SIZE(timings))
2146
2147 struct debugfs_timings_state {
2148         struct kvm_vcpu *vcpu;
2149         unsigned int    buflen;
2150         char            buf[N_TIMINGS * 100];
2151 };
2152
2153 static int debugfs_timings_open(struct inode *inode, struct file *file)
2154 {
2155         struct kvm_vcpu *vcpu = inode->i_private;
2156         struct debugfs_timings_state *p;
2157
2158         p = kzalloc(sizeof(*p), GFP_KERNEL);
2159         if (!p)
2160                 return -ENOMEM;
2161
2162         kvm_get_kvm(vcpu->kvm);
2163         p->vcpu = vcpu;
2164         file->private_data = p;
2165
2166         return nonseekable_open(inode, file);
2167 }
2168
2169 static int debugfs_timings_release(struct inode *inode, struct file *file)
2170 {
2171         struct debugfs_timings_state *p = file->private_data;
2172
2173         kvm_put_kvm(p->vcpu->kvm);
2174         kfree(p);
2175         return 0;
2176 }
2177
2178 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2179                                     size_t len, loff_t *ppos)
2180 {
2181         struct debugfs_timings_state *p = file->private_data;
2182         struct kvm_vcpu *vcpu = p->vcpu;
2183         char *s, *buf_end;
2184         struct kvmhv_tb_accumulator tb;
2185         u64 count;
2186         loff_t pos;
2187         ssize_t n;
2188         int i, loops;
2189         bool ok;
2190
2191         if (!p->buflen) {
2192                 s = p->buf;
2193                 buf_end = s + sizeof(p->buf);
2194                 for (i = 0; i < N_TIMINGS; ++i) {
2195                         struct kvmhv_tb_accumulator *acc;
2196
2197                         acc = (struct kvmhv_tb_accumulator *)
2198                                 ((unsigned long)vcpu + timings[i].offset);
2199                         ok = false;
2200                         for (loops = 0; loops < 1000; ++loops) {
2201                                 count = acc->seqcount;
2202                                 if (!(count & 1)) {
2203                                         smp_rmb();
2204                                         tb = *acc;
2205                                         smp_rmb();
2206                                         if (count == acc->seqcount) {
2207                                                 ok = true;
2208                                                 break;
2209                                         }
2210                                 }
2211                                 udelay(1);
2212                         }
2213                         if (!ok)
2214                                 snprintf(s, buf_end - s, "%s: stuck\n",
2215                                         timings[i].name);
2216                         else
2217                                 snprintf(s, buf_end - s,
2218                                         "%s: %llu %llu %llu %llu\n",
2219                                         timings[i].name, count / 2,
2220                                         tb_to_ns(tb.tb_total),
2221                                         tb_to_ns(tb.tb_min),
2222                                         tb_to_ns(tb.tb_max));
2223                         s += strlen(s);
2224                 }
2225                 p->buflen = s - p->buf;
2226         }
2227
2228         pos = *ppos;
2229         if (pos >= p->buflen)
2230                 return 0;
2231         if (len > p->buflen - pos)
2232                 len = p->buflen - pos;
2233         n = copy_to_user(buf, p->buf + pos, len);
2234         if (n) {
2235                 if (n == len)
2236                         return -EFAULT;
2237                 len -= n;
2238         }
2239         *ppos = pos + len;
2240         return len;
2241 }
2242
2243 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2244                                      size_t len, loff_t *ppos)
2245 {
2246         return -EACCES;
2247 }
2248
2249 static const struct file_operations debugfs_timings_ops = {
2250         .owner   = THIS_MODULE,
2251         .open    = debugfs_timings_open,
2252         .release = debugfs_timings_release,
2253         .read    = debugfs_timings_read,
2254         .write   = debugfs_timings_write,
2255         .llseek  = generic_file_llseek,
2256 };
2257
2258 /* Create a debugfs directory for the vcpu */
2259 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2260 {
2261         char buf[16];
2262         struct kvm *kvm = vcpu->kvm;
2263
2264         snprintf(buf, sizeof(buf), "vcpu%u", id);
2265         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2266         debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir, vcpu,
2267                             &debugfs_timings_ops);
2268 }
2269
2270 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2271 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2272 {
2273 }
2274 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2275
2276 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu)
2277 {
2278         int err;
2279         int core;
2280         struct kvmppc_vcore *vcore;
2281         struct kvm *kvm;
2282         unsigned int id;
2283
2284         kvm = vcpu->kvm;
2285         id = vcpu->vcpu_id;
2286
2287         vcpu->arch.shared = &vcpu->arch.shregs;
2288 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2289         /*
2290          * The shared struct is never shared on HV,
2291          * so we can always use host endianness
2292          */
2293 #ifdef __BIG_ENDIAN__
2294         vcpu->arch.shared_big_endian = true;
2295 #else
2296         vcpu->arch.shared_big_endian = false;
2297 #endif
2298 #endif
2299         vcpu->arch.mmcr[0] = MMCR0_FC;
2300         vcpu->arch.ctrl = CTRL_RUNLATCH;
2301         /* default to host PVR, since we can't spoof it */
2302         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2303         spin_lock_init(&vcpu->arch.vpa_update_lock);
2304         spin_lock_init(&vcpu->arch.tbacct_lock);
2305         vcpu->arch.busy_preempt = TB_NIL;
2306         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2307
2308         /*
2309          * Set the default HFSCR for the guest from the host value.
2310          * This value is only used on POWER9.
2311          * On POWER9, we want to virtualize the doorbell facility, so we
2312          * don't set the HFSCR_MSGP bit, and that causes those instructions
2313          * to trap and then we emulate them.
2314          */
2315         vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2316                 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP;
2317         if (cpu_has_feature(CPU_FTR_HVMODE)) {
2318                 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2319                 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2320                         vcpu->arch.hfscr |= HFSCR_TM;
2321         }
2322         if (cpu_has_feature(CPU_FTR_TM_COMP))
2323                 vcpu->arch.hfscr |= HFSCR_TM;
2324
2325         kvmppc_mmu_book3s_hv_init(vcpu);
2326
2327         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2328
2329         init_waitqueue_head(&vcpu->arch.cpu_run);
2330
2331         mutex_lock(&kvm->lock);
2332         vcore = NULL;
2333         err = -EINVAL;
2334         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2335                 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2336                         pr_devel("KVM: VCPU ID too high\n");
2337                         core = KVM_MAX_VCORES;
2338                 } else {
2339                         BUG_ON(kvm->arch.smt_mode != 1);
2340                         core = kvmppc_pack_vcpu_id(kvm, id);
2341                 }
2342         } else {
2343                 core = id / kvm->arch.smt_mode;
2344         }
2345         if (core < KVM_MAX_VCORES) {
2346                 vcore = kvm->arch.vcores[core];
2347                 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2348                         pr_devel("KVM: collision on id %u", id);
2349                         vcore = NULL;
2350                 } else if (!vcore) {
2351                         /*
2352                          * Take mmu_setup_lock for mutual exclusion
2353                          * with kvmppc_update_lpcr().
2354                          */
2355                         err = -ENOMEM;
2356                         vcore = kvmppc_vcore_create(kvm,
2357                                         id & ~(kvm->arch.smt_mode - 1));
2358                         mutex_lock(&kvm->arch.mmu_setup_lock);
2359                         kvm->arch.vcores[core] = vcore;
2360                         kvm->arch.online_vcores++;
2361                         mutex_unlock(&kvm->arch.mmu_setup_lock);
2362                 }
2363         }
2364         mutex_unlock(&kvm->lock);
2365
2366         if (!vcore)
2367                 return err;
2368
2369         spin_lock(&vcore->lock);
2370         ++vcore->num_threads;
2371         spin_unlock(&vcore->lock);
2372         vcpu->arch.vcore = vcore;
2373         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2374         vcpu->arch.thread_cpu = -1;
2375         vcpu->arch.prev_cpu = -1;
2376
2377         vcpu->arch.cpu_type = KVM_CPU_3S_64;
2378         kvmppc_sanity_check(vcpu);
2379
2380         debugfs_vcpu_init(vcpu, id);
2381
2382         return 0;
2383 }
2384
2385 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2386                               unsigned long flags)
2387 {
2388         int err;
2389         int esmt = 0;
2390
2391         if (flags)
2392                 return -EINVAL;
2393         if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2394                 return -EINVAL;
2395         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2396                 /*
2397                  * On POWER8 (or POWER7), the threading mode is "strict",
2398                  * so we pack smt_mode vcpus per vcore.
2399                  */
2400                 if (smt_mode > threads_per_subcore)
2401                         return -EINVAL;
2402         } else {
2403                 /*
2404                  * On POWER9, the threading mode is "loose",
2405                  * so each vcpu gets its own vcore.
2406                  */
2407                 esmt = smt_mode;
2408                 smt_mode = 1;
2409         }
2410         mutex_lock(&kvm->lock);
2411         err = -EBUSY;
2412         if (!kvm->arch.online_vcores) {
2413                 kvm->arch.smt_mode = smt_mode;
2414                 kvm->arch.emul_smt_mode = esmt;
2415                 err = 0;
2416         }
2417         mutex_unlock(&kvm->lock);
2418
2419         return err;
2420 }
2421
2422 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2423 {
2424         if (vpa->pinned_addr)
2425                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2426                                         vpa->dirty);
2427 }
2428
2429 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2430 {
2431         spin_lock(&vcpu->arch.vpa_update_lock);
2432         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2433         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2434         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2435         spin_unlock(&vcpu->arch.vpa_update_lock);
2436 }
2437
2438 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2439 {
2440         /* Indicate we want to get back into the guest */
2441         return 1;
2442 }
2443
2444 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2445 {
2446         unsigned long dec_nsec, now;
2447
2448         now = get_tb();
2449         if (now > vcpu->arch.dec_expires) {
2450                 /* decrementer has already gone negative */
2451                 kvmppc_core_queue_dec(vcpu);
2452                 kvmppc_core_prepare_to_enter(vcpu);
2453                 return;
2454         }
2455         dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2456         hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2457         vcpu->arch.timer_running = 1;
2458 }
2459
2460 extern int __kvmppc_vcore_entry(void);
2461
2462 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2463                                    struct kvm_vcpu *vcpu)
2464 {
2465         u64 now;
2466
2467         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2468                 return;
2469         spin_lock_irq(&vcpu->arch.tbacct_lock);
2470         now = mftb();
2471         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2472                 vcpu->arch.stolen_logged;
2473         vcpu->arch.busy_preempt = now;
2474         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2475         spin_unlock_irq(&vcpu->arch.tbacct_lock);
2476         --vc->n_runnable;
2477         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2478 }
2479
2480 static int kvmppc_grab_hwthread(int cpu)
2481 {
2482         struct paca_struct *tpaca;
2483         long timeout = 10000;
2484
2485         tpaca = paca_ptrs[cpu];
2486
2487         /* Ensure the thread won't go into the kernel if it wakes */
2488         tpaca->kvm_hstate.kvm_vcpu = NULL;
2489         tpaca->kvm_hstate.kvm_vcore = NULL;
2490         tpaca->kvm_hstate.napping = 0;
2491         smp_wmb();
2492         tpaca->kvm_hstate.hwthread_req = 1;
2493
2494         /*
2495          * If the thread is already executing in the kernel (e.g. handling
2496          * a stray interrupt), wait for it to get back to nap mode.
2497          * The smp_mb() is to ensure that our setting of hwthread_req
2498          * is visible before we look at hwthread_state, so if this
2499          * races with the code at system_reset_pSeries and the thread
2500          * misses our setting of hwthread_req, we are sure to see its
2501          * setting of hwthread_state, and vice versa.
2502          */
2503         smp_mb();
2504         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2505                 if (--timeout <= 0) {
2506                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
2507                         return -EBUSY;
2508                 }
2509                 udelay(1);
2510         }
2511         return 0;
2512 }
2513
2514 static void kvmppc_release_hwthread(int cpu)
2515 {
2516         struct paca_struct *tpaca;
2517
2518         tpaca = paca_ptrs[cpu];
2519         tpaca->kvm_hstate.hwthread_req = 0;
2520         tpaca->kvm_hstate.kvm_vcpu = NULL;
2521         tpaca->kvm_hstate.kvm_vcore = NULL;
2522         tpaca->kvm_hstate.kvm_split_mode = NULL;
2523 }
2524
2525 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2526 {
2527         struct kvm_nested_guest *nested = vcpu->arch.nested;
2528         cpumask_t *cpu_in_guest;
2529         int i;
2530
2531         cpu = cpu_first_thread_sibling(cpu);
2532         if (nested) {
2533                 cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2534                 cpu_in_guest = &nested->cpu_in_guest;
2535         } else {
2536                 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2537                 cpu_in_guest = &kvm->arch.cpu_in_guest;
2538         }
2539         /*
2540          * Make sure setting of bit in need_tlb_flush precedes
2541          * testing of cpu_in_guest bits.  The matching barrier on
2542          * the other side is the first smp_mb() in kvmppc_run_core().
2543          */
2544         smp_mb();
2545         for (i = 0; i < threads_per_core; ++i)
2546                 if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2547                         smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2548 }
2549
2550 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2551 {
2552         struct kvm_nested_guest *nested = vcpu->arch.nested;
2553         struct kvm *kvm = vcpu->kvm;
2554         int prev_cpu;
2555
2556         if (!cpu_has_feature(CPU_FTR_HVMODE))
2557                 return;
2558
2559         if (nested)
2560                 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2561         else
2562                 prev_cpu = vcpu->arch.prev_cpu;
2563
2564         /*
2565          * With radix, the guest can do TLB invalidations itself,
2566          * and it could choose to use the local form (tlbiel) if
2567          * it is invalidating a translation that has only ever been
2568          * used on one vcpu.  However, that doesn't mean it has
2569          * only ever been used on one physical cpu, since vcpus
2570          * can move around between pcpus.  To cope with this, when
2571          * a vcpu moves from one pcpu to another, we need to tell
2572          * any vcpus running on the same core as this vcpu previously
2573          * ran to flush the TLB.  The TLB is shared between threads,
2574          * so we use a single bit in .need_tlb_flush for all 4 threads.
2575          */
2576         if (prev_cpu != pcpu) {
2577                 if (prev_cpu >= 0 &&
2578                     cpu_first_thread_sibling(prev_cpu) !=
2579                     cpu_first_thread_sibling(pcpu))
2580                         radix_flush_cpu(kvm, prev_cpu, vcpu);
2581                 if (nested)
2582                         nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2583                 else
2584                         vcpu->arch.prev_cpu = pcpu;
2585         }
2586 }
2587
2588 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2589 {
2590         int cpu;
2591         struct paca_struct *tpaca;
2592         struct kvm *kvm = vc->kvm;
2593
2594         cpu = vc->pcpu;
2595         if (vcpu) {
2596                 if (vcpu->arch.timer_running) {
2597                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2598                         vcpu->arch.timer_running = 0;
2599                 }
2600                 cpu += vcpu->arch.ptid;
2601                 vcpu->cpu = vc->pcpu;
2602                 vcpu->arch.thread_cpu = cpu;
2603                 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2604         }
2605         tpaca = paca_ptrs[cpu];
2606         tpaca->kvm_hstate.kvm_vcpu = vcpu;
2607         tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2608         tpaca->kvm_hstate.fake_suspend = 0;
2609         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2610         smp_wmb();
2611         tpaca->kvm_hstate.kvm_vcore = vc;
2612         if (cpu != smp_processor_id())
2613                 kvmppc_ipi_thread(cpu);
2614 }
2615
2616 static void kvmppc_wait_for_nap(int n_threads)
2617 {
2618         int cpu = smp_processor_id();
2619         int i, loops;
2620
2621         if (n_threads <= 1)
2622                 return;
2623         for (loops = 0; loops < 1000000; ++loops) {
2624                 /*
2625                  * Check if all threads are finished.
2626                  * We set the vcore pointer when starting a thread
2627                  * and the thread clears it when finished, so we look
2628                  * for any threads that still have a non-NULL vcore ptr.
2629                  */
2630                 for (i = 1; i < n_threads; ++i)
2631                         if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2632                                 break;
2633                 if (i == n_threads) {
2634                         HMT_medium();
2635                         return;
2636                 }
2637                 HMT_low();
2638         }
2639         HMT_medium();
2640         for (i = 1; i < n_threads; ++i)
2641                 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2642                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2643 }
2644
2645 /*
2646  * Check that we are on thread 0 and that any other threads in
2647  * this core are off-line.  Then grab the threads so they can't
2648  * enter the kernel.
2649  */
2650 static int on_primary_thread(void)
2651 {
2652         int cpu = smp_processor_id();
2653         int thr;
2654
2655         /* Are we on a primary subcore? */
2656         if (cpu_thread_in_subcore(cpu))
2657                 return 0;
2658
2659         thr = 0;
2660         while (++thr < threads_per_subcore)
2661                 if (cpu_online(cpu + thr))
2662                         return 0;
2663
2664         /* Grab all hw threads so they can't go into the kernel */
2665         for (thr = 1; thr < threads_per_subcore; ++thr) {
2666                 if (kvmppc_grab_hwthread(cpu + thr)) {
2667                         /* Couldn't grab one; let the others go */
2668                         do {
2669                                 kvmppc_release_hwthread(cpu + thr);
2670                         } while (--thr > 0);
2671                         return 0;
2672                 }
2673         }
2674         return 1;
2675 }
2676
2677 /*
2678  * A list of virtual cores for each physical CPU.
2679  * These are vcores that could run but their runner VCPU tasks are
2680  * (or may be) preempted.
2681  */
2682 struct preempted_vcore_list {
2683         struct list_head        list;
2684         spinlock_t              lock;
2685 };
2686
2687 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2688
2689 static void init_vcore_lists(void)
2690 {
2691         int cpu;
2692
2693         for_each_possible_cpu(cpu) {
2694                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2695                 spin_lock_init(&lp->lock);
2696                 INIT_LIST_HEAD(&lp->list);
2697         }
2698 }
2699
2700 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2701 {
2702         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2703
2704         vc->vcore_state = VCORE_PREEMPT;
2705         vc->pcpu = smp_processor_id();
2706         if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2707                 spin_lock(&lp->lock);
2708                 list_add_tail(&vc->preempt_list, &lp->list);
2709                 spin_unlock(&lp->lock);
2710         }
2711
2712         /* Start accumulating stolen time */
2713         kvmppc_core_start_stolen(vc);
2714 }
2715
2716 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2717 {
2718         struct preempted_vcore_list *lp;
2719
2720         kvmppc_core_end_stolen(vc);
2721         if (!list_empty(&vc->preempt_list)) {
2722                 lp = &per_cpu(preempted_vcores, vc->pcpu);
2723                 spin_lock(&lp->lock);
2724                 list_del_init(&vc->preempt_list);
2725                 spin_unlock(&lp->lock);
2726         }
2727         vc->vcore_state = VCORE_INACTIVE;
2728 }
2729
2730 /*
2731  * This stores information about the virtual cores currently
2732  * assigned to a physical core.
2733  */
2734 struct core_info {
2735         int             n_subcores;
2736         int             max_subcore_threads;
2737         int             total_threads;
2738         int             subcore_threads[MAX_SUBCORES];
2739         struct kvmppc_vcore *vc[MAX_SUBCORES];
2740 };
2741
2742 /*
2743  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2744  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2745  */
2746 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2747
2748 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2749 {
2750         memset(cip, 0, sizeof(*cip));
2751         cip->n_subcores = 1;
2752         cip->max_subcore_threads = vc->num_threads;
2753         cip->total_threads = vc->num_threads;
2754         cip->subcore_threads[0] = vc->num_threads;
2755         cip->vc[0] = vc;
2756 }
2757
2758 static bool subcore_config_ok(int n_subcores, int n_threads)
2759 {
2760         /*
2761          * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2762          * split-core mode, with one thread per subcore.
2763          */
2764         if (cpu_has_feature(CPU_FTR_ARCH_300))
2765                 return n_subcores <= 4 && n_threads == 1;
2766
2767         /* On POWER8, can only dynamically split if unsplit to begin with */
2768         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2769                 return false;
2770         if (n_subcores > MAX_SUBCORES)
2771                 return false;
2772         if (n_subcores > 1) {
2773                 if (!(dynamic_mt_modes & 2))
2774                         n_subcores = 4;
2775                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2776                         return false;
2777         }
2778
2779         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2780 }
2781
2782 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2783 {
2784         vc->entry_exit_map = 0;
2785         vc->in_guest = 0;
2786         vc->napping_threads = 0;
2787         vc->conferring_threads = 0;
2788         vc->tb_offset_applied = 0;
2789 }
2790
2791 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2792 {
2793         int n_threads = vc->num_threads;
2794         int sub;
2795
2796         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2797                 return false;
2798
2799         /* In one_vm_per_core mode, require all vcores to be from the same vm */
2800         if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2801                 return false;
2802
2803         /* Some POWER9 chips require all threads to be in the same MMU mode */
2804         if (no_mixing_hpt_and_radix &&
2805             kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2806                 return false;
2807
2808         if (n_threads < cip->max_subcore_threads)
2809                 n_threads = cip->max_subcore_threads;
2810         if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2811                 return false;
2812         cip->max_subcore_threads = n_threads;
2813
2814         sub = cip->n_subcores;
2815         ++cip->n_subcores;
2816         cip->total_threads += vc->num_threads;
2817         cip->subcore_threads[sub] = vc->num_threads;
2818         cip->vc[sub] = vc;
2819         init_vcore_to_run(vc);
2820         list_del_init(&vc->preempt_list);
2821
2822         return true;
2823 }
2824
2825 /*
2826  * Work out whether it is possible to piggyback the execution of
2827  * vcore *pvc onto the execution of the other vcores described in *cip.
2828  */
2829 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2830                           int target_threads)
2831 {
2832         if (cip->total_threads + pvc->num_threads > target_threads)
2833                 return false;
2834
2835         return can_dynamic_split(pvc, cip);
2836 }
2837
2838 static void prepare_threads(struct kvmppc_vcore *vc)
2839 {
2840         int i;
2841         struct kvm_vcpu *vcpu;
2842
2843         for_each_runnable_thread(i, vcpu, vc) {
2844                 if (signal_pending(vcpu->arch.run_task))
2845                         vcpu->arch.ret = -EINTR;
2846                 else if (vcpu->arch.vpa.update_pending ||
2847                          vcpu->arch.slb_shadow.update_pending ||
2848                          vcpu->arch.dtl.update_pending)
2849                         vcpu->arch.ret = RESUME_GUEST;
2850                 else
2851                         continue;
2852                 kvmppc_remove_runnable(vc, vcpu);
2853                 wake_up(&vcpu->arch.cpu_run);
2854         }
2855 }
2856
2857 static void collect_piggybacks(struct core_info *cip, int target_threads)
2858 {
2859         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2860         struct kvmppc_vcore *pvc, *vcnext;
2861
2862         spin_lock(&lp->lock);
2863         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2864                 if (!spin_trylock(&pvc->lock))
2865                         continue;
2866                 prepare_threads(pvc);
2867                 if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
2868                         list_del_init(&pvc->preempt_list);
2869                         if (pvc->runner == NULL) {
2870                                 pvc->vcore_state = VCORE_INACTIVE;
2871                                 kvmppc_core_end_stolen(pvc);
2872                         }
2873                         spin_unlock(&pvc->lock);
2874                         continue;
2875                 }
2876                 if (!can_piggyback(pvc, cip, target_threads)) {
2877                         spin_unlock(&pvc->lock);
2878                         continue;
2879                 }
2880                 kvmppc_core_end_stolen(pvc);
2881                 pvc->vcore_state = VCORE_PIGGYBACK;
2882                 if (cip->total_threads >= target_threads)
2883                         break;
2884         }
2885         spin_unlock(&lp->lock);
2886 }
2887
2888 static bool recheck_signals_and_mmu(struct core_info *cip)
2889 {
2890         int sub, i;
2891         struct kvm_vcpu *vcpu;
2892         struct kvmppc_vcore *vc;
2893
2894         for (sub = 0; sub < cip->n_subcores; ++sub) {
2895                 vc = cip->vc[sub];
2896                 if (!vc->kvm->arch.mmu_ready)
2897                         return true;
2898                 for_each_runnable_thread(i, vcpu, vc)
2899                         if (signal_pending(vcpu->arch.run_task))
2900                                 return true;
2901         }
2902         return false;
2903 }
2904
2905 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2906 {
2907         int still_running = 0, i;
2908         u64 now;
2909         long ret;
2910         struct kvm_vcpu *vcpu;
2911
2912         spin_lock(&vc->lock);
2913         now = get_tb();
2914         for_each_runnable_thread(i, vcpu, vc) {
2915                 /*
2916                  * It's safe to unlock the vcore in the loop here, because
2917                  * for_each_runnable_thread() is safe against removal of
2918                  * the vcpu, and the vcore state is VCORE_EXITING here,
2919                  * so any vcpus becoming runnable will have their arch.trap
2920                  * set to zero and can't actually run in the guest.
2921                  */
2922                 spin_unlock(&vc->lock);
2923                 /* cancel pending dec exception if dec is positive */
2924                 if (now < vcpu->arch.dec_expires &&
2925                     kvmppc_core_pending_dec(vcpu))
2926                         kvmppc_core_dequeue_dec(vcpu);
2927
2928                 trace_kvm_guest_exit(vcpu);
2929
2930                 ret = RESUME_GUEST;
2931                 if (vcpu->arch.trap)
2932                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2933                                                     vcpu->arch.run_task);
2934
2935                 vcpu->arch.ret = ret;
2936                 vcpu->arch.trap = 0;
2937
2938                 spin_lock(&vc->lock);
2939                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2940                         if (vcpu->arch.pending_exceptions)
2941                                 kvmppc_core_prepare_to_enter(vcpu);
2942                         if (vcpu->arch.ceded)
2943                                 kvmppc_set_timer(vcpu);
2944                         else
2945                                 ++still_running;
2946                 } else {
2947                         kvmppc_remove_runnable(vc, vcpu);
2948                         wake_up(&vcpu->arch.cpu_run);
2949                 }
2950         }
2951         if (!is_master) {
2952                 if (still_running > 0) {
2953                         kvmppc_vcore_preempt(vc);
2954                 } else if (vc->runner) {
2955                         vc->vcore_state = VCORE_PREEMPT;
2956                         kvmppc_core_start_stolen(vc);
2957                 } else {
2958                         vc->vcore_state = VCORE_INACTIVE;
2959                 }
2960                 if (vc->n_runnable > 0 && vc->runner == NULL) {
2961                         /* make sure there's a candidate runner awake */
2962                         i = -1;
2963                         vcpu = next_runnable_thread(vc, &i);
2964                         wake_up(&vcpu->arch.cpu_run);
2965                 }
2966         }
2967         spin_unlock(&vc->lock);
2968 }
2969
2970 /*
2971  * Clear core from the list of active host cores as we are about to
2972  * enter the guest. Only do this if it is the primary thread of the
2973  * core (not if a subcore) that is entering the guest.
2974  */
2975 static inline int kvmppc_clear_host_core(unsigned int cpu)
2976 {
2977         int core;
2978
2979         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2980                 return 0;
2981         /*
2982          * Memory barrier can be omitted here as we will do a smp_wmb()
2983          * later in kvmppc_start_thread and we need ensure that state is
2984          * visible to other CPUs only after we enter guest.
2985          */
2986         core = cpu >> threads_shift;
2987         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
2988         return 0;
2989 }
2990
2991 /*
2992  * Advertise this core as an active host core since we exited the guest
2993  * Only need to do this if it is the primary thread of the core that is
2994  * exiting.
2995  */
2996 static inline int kvmppc_set_host_core(unsigned int cpu)
2997 {
2998         int core;
2999
3000         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3001                 return 0;
3002
3003         /*
3004          * Memory barrier can be omitted here because we do a spin_unlock
3005          * immediately after this which provides the memory barrier.
3006          */
3007         core = cpu >> threads_shift;
3008         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3009         return 0;
3010 }
3011
3012 static void set_irq_happened(int trap)
3013 {
3014         switch (trap) {
3015         case BOOK3S_INTERRUPT_EXTERNAL:
3016                 local_paca->irq_happened |= PACA_IRQ_EE;
3017                 break;
3018         case BOOK3S_INTERRUPT_H_DOORBELL:
3019                 local_paca->irq_happened |= PACA_IRQ_DBELL;
3020                 break;
3021         case BOOK3S_INTERRUPT_HMI:
3022                 local_paca->irq_happened |= PACA_IRQ_HMI;
3023                 break;
3024         case BOOK3S_INTERRUPT_SYSTEM_RESET:
3025                 replay_system_reset();
3026                 break;
3027         }
3028 }
3029
3030 /*
3031  * Run a set of guest threads on a physical core.
3032  * Called with vc->lock held.
3033  */
3034 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3035 {
3036         struct kvm_vcpu *vcpu;
3037         int i;
3038         int srcu_idx;
3039         struct core_info core_info;
3040         struct kvmppc_vcore *pvc;
3041         struct kvm_split_mode split_info, *sip;
3042         int split, subcore_size, active;
3043         int sub;
3044         bool thr0_done;
3045         unsigned long cmd_bit, stat_bit;
3046         int pcpu, thr;
3047         int target_threads;
3048         int controlled_threads;
3049         int trap;
3050         bool is_power8;
3051         bool hpt_on_radix;
3052
3053         /*
3054          * Remove from the list any threads that have a signal pending
3055          * or need a VPA update done
3056          */
3057         prepare_threads(vc);
3058
3059         /* if the runner is no longer runnable, let the caller pick a new one */
3060         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3061                 return;
3062
3063         /*
3064          * Initialize *vc.
3065          */
3066         init_vcore_to_run(vc);
3067         vc->preempt_tb = TB_NIL;
3068
3069         /*
3070          * Number of threads that we will be controlling: the same as
3071          * the number of threads per subcore, except on POWER9,
3072          * where it's 1 because the threads are (mostly) independent.
3073          */
3074         controlled_threads = threads_per_vcore(vc->kvm);
3075
3076         /*
3077          * Make sure we are running on primary threads, and that secondary
3078          * threads are offline.  Also check if the number of threads in this
3079          * guest are greater than the current system threads per guest.
3080          * On POWER9, we need to be not in independent-threads mode if
3081          * this is a HPT guest on a radix host machine where the
3082          * CPU threads may not be in different MMU modes.
3083          */
3084         hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
3085                 !kvm_is_radix(vc->kvm);
3086         if (((controlled_threads > 1) &&
3087              ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
3088             (hpt_on_radix && vc->kvm->arch.threads_indep)) {
3089                 for_each_runnable_thread(i, vcpu, vc) {
3090                         vcpu->arch.ret = -EBUSY;
3091                         kvmppc_remove_runnable(vc, vcpu);
3092                         wake_up(&vcpu->arch.cpu_run);
3093                 }
3094                 goto out;
3095         }
3096
3097         /*
3098          * See if we could run any other vcores on the physical core
3099          * along with this one.
3100          */
3101         init_core_info(&core_info, vc);
3102         pcpu = smp_processor_id();
3103         target_threads = controlled_threads;
3104         if (target_smt_mode && target_smt_mode < target_threads)
3105                 target_threads = target_smt_mode;
3106         if (vc->num_threads < target_threads)
3107                 collect_piggybacks(&core_info, target_threads);
3108
3109         /*
3110          * On radix, arrange for TLB flushing if necessary.
3111          * This has to be done before disabling interrupts since
3112          * it uses smp_call_function().
3113          */
3114         pcpu = smp_processor_id();
3115         if (kvm_is_radix(vc->kvm)) {
3116                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3117                         for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3118                                 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3119         }
3120
3121         /*
3122          * Hard-disable interrupts, and check resched flag and signals.
3123          * If we need to reschedule or deliver a signal, clean up
3124          * and return without going into the guest(s).
3125          * If the mmu_ready flag has been cleared, don't go into the
3126          * guest because that means a HPT resize operation is in progress.
3127          */
3128         local_irq_disable();
3129         hard_irq_disable();
3130         if (lazy_irq_pending() || need_resched() ||
3131             recheck_signals_and_mmu(&core_info)) {
3132                 local_irq_enable();
3133                 vc->vcore_state = VCORE_INACTIVE;
3134                 /* Unlock all except the primary vcore */
3135                 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3136                         pvc = core_info.vc[sub];
3137                         /* Put back on to the preempted vcores list */
3138                         kvmppc_vcore_preempt(pvc);
3139                         spin_unlock(&pvc->lock);
3140                 }
3141                 for (i = 0; i < controlled_threads; ++i)
3142                         kvmppc_release_hwthread(pcpu + i);
3143                 return;
3144         }
3145
3146         kvmppc_clear_host_core(pcpu);
3147
3148         /* Decide on micro-threading (split-core) mode */
3149         subcore_size = threads_per_subcore;
3150         cmd_bit = stat_bit = 0;
3151         split = core_info.n_subcores;
3152         sip = NULL;
3153         is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3154                 && !cpu_has_feature(CPU_FTR_ARCH_300);
3155
3156         if (split > 1 || hpt_on_radix) {
3157                 sip = &split_info;
3158                 memset(&split_info, 0, sizeof(split_info));
3159                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3160                         split_info.vc[sub] = core_info.vc[sub];
3161
3162                 if (is_power8) {
3163                         if (split == 2 && (dynamic_mt_modes & 2)) {
3164                                 cmd_bit = HID0_POWER8_1TO2LPAR;
3165                                 stat_bit = HID0_POWER8_2LPARMODE;
3166                         } else {
3167                                 split = 4;
3168                                 cmd_bit = HID0_POWER8_1TO4LPAR;
3169                                 stat_bit = HID0_POWER8_4LPARMODE;
3170                         }
3171                         subcore_size = MAX_SMT_THREADS / split;
3172                         split_info.rpr = mfspr(SPRN_RPR);
3173                         split_info.pmmar = mfspr(SPRN_PMMAR);
3174                         split_info.ldbar = mfspr(SPRN_LDBAR);
3175                         split_info.subcore_size = subcore_size;
3176                 } else {
3177                         split_info.subcore_size = 1;
3178                         if (hpt_on_radix) {
3179                                 /* Use the split_info for LPCR/LPIDR changes */
3180                                 split_info.lpcr_req = vc->lpcr;
3181                                 split_info.lpidr_req = vc->kvm->arch.lpid;
3182                                 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
3183                                 split_info.do_set = 1;
3184                         }
3185                 }
3186
3187                 /* order writes to split_info before kvm_split_mode pointer */
3188                 smp_wmb();
3189         }
3190
3191         for (thr = 0; thr < controlled_threads; ++thr) {
3192                 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3193
3194                 paca->kvm_hstate.tid = thr;
3195                 paca->kvm_hstate.napping = 0;
3196                 paca->kvm_hstate.kvm_split_mode = sip;
3197         }
3198
3199         /* Initiate micro-threading (split-core) on POWER8 if required */
3200         if (cmd_bit) {
3201                 unsigned long hid0 = mfspr(SPRN_HID0);
3202
3203                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3204                 mb();
3205                 mtspr(SPRN_HID0, hid0);
3206                 isync();
3207                 for (;;) {
3208                         hid0 = mfspr(SPRN_HID0);
3209                         if (hid0 & stat_bit)
3210                                 break;
3211                         cpu_relax();
3212                 }
3213         }
3214
3215         /*
3216          * On POWER8, set RWMR register.
3217          * Since it only affects PURR and SPURR, it doesn't affect
3218          * the host, so we don't save/restore the host value.
3219          */
3220         if (is_power8) {
3221                 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3222                 int n_online = atomic_read(&vc->online_count);
3223
3224                 /*
3225                  * Use the 8-thread value if we're doing split-core
3226                  * or if the vcore's online count looks bogus.
3227                  */
3228                 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3229                     n_online >= 1 && n_online <= MAX_SMT_THREADS)
3230                         rwmr_val = p8_rwmr_values[n_online];
3231                 mtspr(SPRN_RWMR, rwmr_val);
3232         }
3233
3234         /* Start all the threads */
3235         active = 0;
3236         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3237                 thr = is_power8 ? subcore_thread_map[sub] : sub;
3238                 thr0_done = false;
3239                 active |= 1 << thr;
3240                 pvc = core_info.vc[sub];
3241                 pvc->pcpu = pcpu + thr;
3242                 for_each_runnable_thread(i, vcpu, pvc) {
3243                         kvmppc_start_thread(vcpu, pvc);
3244                         kvmppc_create_dtl_entry(vcpu, pvc);
3245                         trace_kvm_guest_enter(vcpu);
3246                         if (!vcpu->arch.ptid)
3247                                 thr0_done = true;
3248                         active |= 1 << (thr + vcpu->arch.ptid);
3249                 }
3250                 /*
3251                  * We need to start the first thread of each subcore
3252                  * even if it doesn't have a vcpu.
3253                  */
3254                 if (!thr0_done)
3255                         kvmppc_start_thread(NULL, pvc);
3256         }
3257
3258         /*
3259          * Ensure that split_info.do_nap is set after setting
3260          * the vcore pointer in the PACA of the secondaries.
3261          */
3262         smp_mb();
3263
3264         /*
3265          * When doing micro-threading, poke the inactive threads as well.
3266          * This gets them to the nap instruction after kvm_do_nap,
3267          * which reduces the time taken to unsplit later.
3268          * For POWER9 HPT guest on radix host, we need all the secondary
3269          * threads woken up so they can do the LPCR/LPIDR change.
3270          */
3271         if (cmd_bit || hpt_on_radix) {
3272                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
3273                 for (thr = 1; thr < threads_per_subcore; ++thr)
3274                         if (!(active & (1 << thr)))
3275                                 kvmppc_ipi_thread(pcpu + thr);
3276         }
3277
3278         vc->vcore_state = VCORE_RUNNING;
3279         preempt_disable();
3280
3281         trace_kvmppc_run_core(vc, 0);
3282
3283         for (sub = 0; sub < core_info.n_subcores; ++sub)
3284                 spin_unlock(&core_info.vc[sub]->lock);
3285
3286         guest_enter_irqoff();
3287
3288         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3289
3290         this_cpu_disable_ftrace();
3291
3292         /*
3293          * Interrupts will be enabled once we get into the guest,
3294          * so tell lockdep that we're about to enable interrupts.
3295          */
3296         trace_hardirqs_on();
3297
3298         trap = __kvmppc_vcore_entry();
3299
3300         trace_hardirqs_off();
3301
3302         this_cpu_enable_ftrace();
3303
3304         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3305
3306         set_irq_happened(trap);
3307
3308         spin_lock(&vc->lock);
3309         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3310         vc->vcore_state = VCORE_EXITING;
3311
3312         /* wait for secondary threads to finish writing their state to memory */
3313         kvmppc_wait_for_nap(controlled_threads);
3314
3315         /* Return to whole-core mode if we split the core earlier */
3316         if (cmd_bit) {
3317                 unsigned long hid0 = mfspr(SPRN_HID0);
3318                 unsigned long loops = 0;
3319
3320                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3321                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3322                 mb();
3323                 mtspr(SPRN_HID0, hid0);
3324                 isync();
3325                 for (;;) {
3326                         hid0 = mfspr(SPRN_HID0);
3327                         if (!(hid0 & stat_bit))
3328                                 break;
3329                         cpu_relax();
3330                         ++loops;
3331                 }
3332         } else if (hpt_on_radix) {
3333                 /* Wait for all threads to have seen final sync */
3334                 for (thr = 1; thr < controlled_threads; ++thr) {
3335                         struct paca_struct *paca = paca_ptrs[pcpu + thr];
3336
3337                         while (paca->kvm_hstate.kvm_split_mode) {
3338                                 HMT_low();
3339                                 barrier();
3340                         }
3341                         HMT_medium();
3342                 }
3343         }
3344         split_info.do_nap = 0;
3345
3346         kvmppc_set_host_core(pcpu);
3347
3348         local_irq_enable();
3349         guest_exit();
3350
3351         /* Let secondaries go back to the offline loop */
3352         for (i = 0; i < controlled_threads; ++i) {
3353                 kvmppc_release_hwthread(pcpu + i);
3354                 if (sip && sip->napped[i])
3355                         kvmppc_ipi_thread(pcpu + i);
3356                 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3357         }
3358
3359         spin_unlock(&vc->lock);
3360
3361         /* make sure updates to secondary vcpu structs are visible now */
3362         smp_mb();
3363
3364         preempt_enable();
3365
3366         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3367                 pvc = core_info.vc[sub];
3368                 post_guest_process(pvc, pvc == vc);
3369         }
3370
3371         spin_lock(&vc->lock);
3372
3373  out:
3374         vc->vcore_state = VCORE_INACTIVE;
3375         trace_kvmppc_run_core(vc, 1);
3376 }
3377
3378 /*
3379  * Load up hypervisor-mode registers on P9.
3380  */
3381 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3382                                      unsigned long lpcr)
3383 {
3384         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3385         s64 hdec;
3386         u64 tb, purr, spurr;
3387         int trap;
3388         unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3389         unsigned long host_ciabr = mfspr(SPRN_CIABR);
3390         unsigned long host_dawr = mfspr(SPRN_DAWR0);
3391         unsigned long host_dawrx = mfspr(SPRN_DAWRX0);
3392         unsigned long host_psscr = mfspr(SPRN_PSSCR);
3393         unsigned long host_pidr = mfspr(SPRN_PID);
3394
3395         hdec = time_limit - mftb();
3396         if (hdec < 0)
3397                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3398         mtspr(SPRN_HDEC, hdec);
3399
3400         if (vc->tb_offset) {
3401                 u64 new_tb = mftb() + vc->tb_offset;
3402                 mtspr(SPRN_TBU40, new_tb);
3403                 tb = mftb();
3404                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3405                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3406                 vc->tb_offset_applied = vc->tb_offset;
3407         }
3408
3409         if (vc->pcr)
3410                 mtspr(SPRN_PCR, vc->pcr | PCR_MASK);
3411         mtspr(SPRN_DPDES, vc->dpdes);
3412         mtspr(SPRN_VTB, vc->vtb);
3413
3414         local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3415         local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3416         mtspr(SPRN_PURR, vcpu->arch.purr);
3417         mtspr(SPRN_SPURR, vcpu->arch.spurr);
3418
3419         if (dawr_enabled()) {
3420                 mtspr(SPRN_DAWR0, vcpu->arch.dawr);
3421                 mtspr(SPRN_DAWRX0, vcpu->arch.dawrx);
3422         }
3423         mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3424         mtspr(SPRN_IC, vcpu->arch.ic);
3425         mtspr(SPRN_PID, vcpu->arch.pid);
3426
3427         mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3428               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3429
3430         mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3431
3432         mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3433         mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3434         mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3435         mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3436
3437         mtspr(SPRN_AMOR, ~0UL);
3438
3439         mtspr(SPRN_LPCR, lpcr);
3440         isync();
3441
3442         kvmppc_xive_push_vcpu(vcpu);
3443
3444         mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3445         mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3446
3447         trap = __kvmhv_vcpu_entry_p9(vcpu);
3448
3449         /* Advance host PURR/SPURR by the amount used by guest */
3450         purr = mfspr(SPRN_PURR);
3451         spurr = mfspr(SPRN_SPURR);
3452         mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3453               purr - vcpu->arch.purr);
3454         mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3455               spurr - vcpu->arch.spurr);
3456         vcpu->arch.purr = purr;
3457         vcpu->arch.spurr = spurr;
3458
3459         vcpu->arch.ic = mfspr(SPRN_IC);
3460         vcpu->arch.pid = mfspr(SPRN_PID);
3461         vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3462
3463         vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3464         vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3465         vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3466         vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3467
3468         /* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3469         mtspr(SPRN_PSSCR, host_psscr |
3470               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3471         mtspr(SPRN_HFSCR, host_hfscr);
3472         mtspr(SPRN_CIABR, host_ciabr);
3473         mtspr(SPRN_DAWR0, host_dawr);
3474         mtspr(SPRN_DAWRX0, host_dawrx);
3475         mtspr(SPRN_PID, host_pidr);
3476
3477         /*
3478          * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3479          * case we interrupted the guest between a tlbie and a ptesync.
3480          */
3481         asm volatile("eieio; tlbsync; ptesync");
3482
3483         mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid);    /* restore host LPID */
3484         isync();
3485
3486         vc->dpdes = mfspr(SPRN_DPDES);
3487         vc->vtb = mfspr(SPRN_VTB);
3488         mtspr(SPRN_DPDES, 0);
3489         if (vc->pcr)
3490                 mtspr(SPRN_PCR, PCR_MASK);
3491
3492         if (vc->tb_offset_applied) {
3493                 u64 new_tb = mftb() - vc->tb_offset_applied;
3494                 mtspr(SPRN_TBU40, new_tb);
3495                 tb = mftb();
3496                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3497                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3498                 vc->tb_offset_applied = 0;
3499         }
3500
3501         mtspr(SPRN_HDEC, 0x7fffffff);
3502         mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3503
3504         return trap;
3505 }
3506
3507 /*
3508  * Virtual-mode guest entry for POWER9 and later when the host and
3509  * guest are both using the radix MMU.  The LPIDR has already been set.
3510  */
3511 int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3512                          unsigned long lpcr)
3513 {
3514         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3515         unsigned long host_dscr = mfspr(SPRN_DSCR);
3516         unsigned long host_tidr = mfspr(SPRN_TIDR);
3517         unsigned long host_iamr = mfspr(SPRN_IAMR);
3518         unsigned long host_amr = mfspr(SPRN_AMR);
3519         s64 dec;
3520         u64 tb;
3521         int trap, save_pmu;
3522
3523         dec = mfspr(SPRN_DEC);
3524         tb = mftb();
3525         if (dec < 512)
3526                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3527         local_paca->kvm_hstate.dec_expires = dec + tb;
3528         if (local_paca->kvm_hstate.dec_expires < time_limit)
3529                 time_limit = local_paca->kvm_hstate.dec_expires;
3530
3531         vcpu->arch.ceded = 0;
3532
3533         kvmhv_save_host_pmu();          /* saves it to PACA kvm_hstate */
3534
3535         kvmppc_subcore_enter_guest();
3536
3537         vc->entry_exit_map = 1;
3538         vc->in_guest = 1;
3539
3540         if (vcpu->arch.vpa.pinned_addr) {
3541                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3542                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3543                 lp->yield_count = cpu_to_be32(yield_count);
3544                 vcpu->arch.vpa.dirty = 1;
3545         }
3546
3547         if (cpu_has_feature(CPU_FTR_TM) ||
3548             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3549                 kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3550
3551         kvmhv_load_guest_pmu(vcpu);
3552
3553         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3554         load_fp_state(&vcpu->arch.fp);
3555 #ifdef CONFIG_ALTIVEC
3556         load_vr_state(&vcpu->arch.vr);
3557 #endif
3558         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3559
3560         mtspr(SPRN_DSCR, vcpu->arch.dscr);
3561         mtspr(SPRN_IAMR, vcpu->arch.iamr);
3562         mtspr(SPRN_PSPB, vcpu->arch.pspb);
3563         mtspr(SPRN_FSCR, vcpu->arch.fscr);
3564         mtspr(SPRN_TAR, vcpu->arch.tar);
3565         mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3566         mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3567         mtspr(SPRN_BESCR, vcpu->arch.bescr);
3568         mtspr(SPRN_WORT, vcpu->arch.wort);
3569         mtspr(SPRN_TIDR, vcpu->arch.tid);
3570         mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3571         mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3572         mtspr(SPRN_AMR, vcpu->arch.amr);
3573         mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3574
3575         if (!(vcpu->arch.ctrl & 1))
3576                 mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3577
3578         mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3579
3580         if (kvmhv_on_pseries()) {
3581                 /*
3582                  * We need to save and restore the guest visible part of the
3583                  * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3584                  * doesn't do this for us. Note only required if pseries since
3585                  * this is done in kvmhv_load_hv_regs_and_go() below otherwise.
3586                  */
3587                 unsigned long host_psscr;
3588                 /* call our hypervisor to load up HV regs and go */
3589                 struct hv_guest_state hvregs;
3590
3591                 host_psscr = mfspr(SPRN_PSSCR_PR);
3592                 mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3593                 kvmhv_save_hv_regs(vcpu, &hvregs);
3594                 hvregs.lpcr = lpcr;
3595                 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3596                 hvregs.version = HV_GUEST_STATE_VERSION;
3597                 if (vcpu->arch.nested) {
3598                         hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3599                         hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3600                 } else {
3601                         hvregs.lpid = vcpu->kvm->arch.lpid;
3602                         hvregs.vcpu_token = vcpu->vcpu_id;
3603                 }
3604                 hvregs.hdec_expiry = time_limit;
3605                 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3606                                           __pa(&vcpu->arch.regs));
3607                 kvmhv_restore_hv_return_state(vcpu, &hvregs);
3608                 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3609                 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3610                 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3611                 vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
3612                 mtspr(SPRN_PSSCR_PR, host_psscr);
3613
3614                 /* H_CEDE has to be handled now, not later */
3615                 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3616                     kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3617                         kvmppc_nested_cede(vcpu);
3618                         kvmppc_set_gpr(vcpu, 3, 0);
3619                         trap = 0;
3620                 }
3621         } else {
3622                 trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3623         }
3624
3625         vcpu->arch.slb_max = 0;
3626         dec = mfspr(SPRN_DEC);
3627         if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
3628                 dec = (s32) dec;
3629         tb = mftb();
3630         vcpu->arch.dec_expires = dec + tb;
3631         vcpu->cpu = -1;
3632         vcpu->arch.thread_cpu = -1;
3633         vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3634
3635         vcpu->arch.iamr = mfspr(SPRN_IAMR);
3636         vcpu->arch.pspb = mfspr(SPRN_PSPB);
3637         vcpu->arch.fscr = mfspr(SPRN_FSCR);
3638         vcpu->arch.tar = mfspr(SPRN_TAR);
3639         vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3640         vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3641         vcpu->arch.bescr = mfspr(SPRN_BESCR);
3642         vcpu->arch.wort = mfspr(SPRN_WORT);
3643         vcpu->arch.tid = mfspr(SPRN_TIDR);
3644         vcpu->arch.amr = mfspr(SPRN_AMR);
3645         vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3646         vcpu->arch.dscr = mfspr(SPRN_DSCR);
3647
3648         mtspr(SPRN_PSPB, 0);
3649         mtspr(SPRN_WORT, 0);
3650         mtspr(SPRN_UAMOR, 0);
3651         mtspr(SPRN_DSCR, host_dscr);
3652         mtspr(SPRN_TIDR, host_tidr);
3653         mtspr(SPRN_IAMR, host_iamr);
3654         mtspr(SPRN_PSPB, 0);
3655
3656         if (host_amr != vcpu->arch.amr)
3657                 mtspr(SPRN_AMR, host_amr);
3658
3659         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3660         store_fp_state(&vcpu->arch.fp);
3661 #ifdef CONFIG_ALTIVEC
3662         store_vr_state(&vcpu->arch.vr);
3663 #endif
3664         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3665
3666         if (cpu_has_feature(CPU_FTR_TM) ||
3667             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3668                 kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3669
3670         save_pmu = 1;
3671         if (vcpu->arch.vpa.pinned_addr) {
3672                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3673                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3674                 lp->yield_count = cpu_to_be32(yield_count);
3675                 vcpu->arch.vpa.dirty = 1;
3676                 save_pmu = lp->pmcregs_in_use;
3677         }
3678         /* Must save pmu if this guest is capable of running nested guests */
3679         save_pmu |= nesting_enabled(vcpu->kvm);
3680
3681         kvmhv_save_guest_pmu(vcpu, save_pmu);
3682
3683         vc->entry_exit_map = 0x101;
3684         vc->in_guest = 0;
3685
3686         mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3687         mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3688
3689         kvmhv_load_host_pmu();
3690
3691         kvmppc_subcore_exit_guest();
3692
3693         return trap;
3694 }
3695
3696 /*
3697  * Wait for some other vcpu thread to execute us, and
3698  * wake us up when we need to handle something in the host.
3699  */
3700 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3701                                  struct kvm_vcpu *vcpu, int wait_state)
3702 {
3703         DEFINE_WAIT(wait);
3704
3705         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3706         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3707                 spin_unlock(&vc->lock);
3708                 schedule();
3709                 spin_lock(&vc->lock);
3710         }
3711         finish_wait(&vcpu->arch.cpu_run, &wait);
3712 }
3713
3714 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3715 {
3716         if (!halt_poll_ns_grow)
3717                 return;
3718
3719         vc->halt_poll_ns *= halt_poll_ns_grow;
3720         if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3721                 vc->halt_poll_ns = halt_poll_ns_grow_start;
3722 }
3723
3724 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3725 {
3726         if (halt_poll_ns_shrink == 0)
3727                 vc->halt_poll_ns = 0;
3728         else
3729                 vc->halt_poll_ns /= halt_poll_ns_shrink;
3730 }
3731
3732 #ifdef CONFIG_KVM_XICS
3733 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3734 {
3735         if (!xics_on_xive())
3736                 return false;
3737         return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3738                 vcpu->arch.xive_saved_state.cppr;
3739 }
3740 #else
3741 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3742 {
3743         return false;
3744 }
3745 #endif /* CONFIG_KVM_XICS */
3746
3747 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3748 {
3749         if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3750             kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3751                 return true;
3752
3753         return false;
3754 }
3755
3756 /*
3757  * Check to see if any of the runnable vcpus on the vcore have pending
3758  * exceptions or are no longer ceded
3759  */
3760 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3761 {
3762         struct kvm_vcpu *vcpu;
3763         int i;
3764
3765         for_each_runnable_thread(i, vcpu, vc) {
3766                 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3767                         return 1;
3768         }
3769
3770         return 0;
3771 }
3772
3773 /*
3774  * All the vcpus in this vcore are idle, so wait for a decrementer
3775  * or external interrupt to one of the vcpus.  vc->lock is held.
3776  */
3777 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3778 {
3779         ktime_t cur, start_poll, start_wait;
3780         int do_sleep = 1;
3781         u64 block_ns;
3782
3783         /* Poll for pending exceptions and ceded state */
3784         cur = start_poll = ktime_get();
3785         if (vc->halt_poll_ns) {
3786                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3787                 ++vc->runner->stat.halt_attempted_poll;
3788
3789                 vc->vcore_state = VCORE_POLLING;
3790                 spin_unlock(&vc->lock);
3791
3792                 do {
3793                         if (kvmppc_vcore_check_block(vc)) {
3794                                 do_sleep = 0;
3795                                 break;
3796                         }
3797                         cur = ktime_get();
3798                 } while (single_task_running() && ktime_before(cur, stop));
3799
3800                 spin_lock(&vc->lock);
3801                 vc->vcore_state = VCORE_INACTIVE;
3802
3803                 if (!do_sleep) {
3804                         ++vc->runner->stat.halt_successful_poll;
3805                         goto out;
3806                 }
3807         }
3808
3809         prepare_to_rcuwait(&vc->wait);
3810         set_current_state(TASK_INTERRUPTIBLE);
3811         if (kvmppc_vcore_check_block(vc)) {
3812                 finish_rcuwait(&vc->wait);
3813                 do_sleep = 0;
3814                 /* If we polled, count this as a successful poll */
3815                 if (vc->halt_poll_ns)
3816                         ++vc->runner->stat.halt_successful_poll;
3817                 goto out;
3818         }
3819
3820         start_wait = ktime_get();
3821
3822         vc->vcore_state = VCORE_SLEEPING;
3823         trace_kvmppc_vcore_blocked(vc, 0);
3824         spin_unlock(&vc->lock);
3825         schedule();
3826         finish_rcuwait(&vc->wait);
3827         spin_lock(&vc->lock);
3828         vc->vcore_state = VCORE_INACTIVE;
3829         trace_kvmppc_vcore_blocked(vc, 1);
3830         ++vc->runner->stat.halt_successful_wait;
3831
3832         cur = ktime_get();
3833
3834 out:
3835         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3836
3837         /* Attribute wait time */
3838         if (do_sleep) {
3839                 vc->runner->stat.halt_wait_ns +=
3840                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
3841                 /* Attribute failed poll time */
3842                 if (vc->halt_poll_ns)
3843                         vc->runner->stat.halt_poll_fail_ns +=
3844                                 ktime_to_ns(start_wait) -
3845                                 ktime_to_ns(start_poll);
3846         } else {
3847                 /* Attribute successful poll time */
3848                 if (vc->halt_poll_ns)
3849                         vc->runner->stat.halt_poll_success_ns +=
3850                                 ktime_to_ns(cur) -
3851                                 ktime_to_ns(start_poll);
3852         }
3853
3854         /* Adjust poll time */
3855         if (halt_poll_ns) {
3856                 if (block_ns <= vc->halt_poll_ns)
3857                         ;
3858                 /* We slept and blocked for longer than the max halt time */
3859                 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3860                         shrink_halt_poll_ns(vc);
3861                 /* We slept and our poll time is too small */
3862                 else if (vc->halt_poll_ns < halt_poll_ns &&
3863                                 block_ns < halt_poll_ns)
3864                         grow_halt_poll_ns(vc);
3865                 if (vc->halt_poll_ns > halt_poll_ns)
3866                         vc->halt_poll_ns = halt_poll_ns;
3867         } else
3868                 vc->halt_poll_ns = 0;
3869
3870         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3871 }
3872
3873 /*
3874  * This never fails for a radix guest, as none of the operations it does
3875  * for a radix guest can fail or have a way to report failure.
3876  * kvmhv_run_single_vcpu() relies on this fact.
3877  */
3878 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3879 {
3880         int r = 0;
3881         struct kvm *kvm = vcpu->kvm;
3882
3883         mutex_lock(&kvm->arch.mmu_setup_lock);
3884         if (!kvm->arch.mmu_ready) {
3885                 if (!kvm_is_radix(kvm))
3886                         r = kvmppc_hv_setup_htab_rma(vcpu);
3887                 if (!r) {
3888                         if (cpu_has_feature(CPU_FTR_ARCH_300))
3889                                 kvmppc_setup_partition_table(kvm);
3890                         kvm->arch.mmu_ready = 1;
3891                 }
3892         }
3893         mutex_unlock(&kvm->arch.mmu_setup_lock);
3894         return r;
3895 }
3896
3897 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3898 {
3899         int n_ceded, i, r;
3900         struct kvmppc_vcore *vc;
3901         struct kvm_vcpu *v;
3902
3903         trace_kvmppc_run_vcpu_enter(vcpu);
3904
3905         kvm_run->exit_reason = 0;
3906         vcpu->arch.ret = RESUME_GUEST;
3907         vcpu->arch.trap = 0;
3908         kvmppc_update_vpas(vcpu);
3909
3910         /*
3911          * Synchronize with other threads in this virtual core
3912          */
3913         vc = vcpu->arch.vcore;
3914         spin_lock(&vc->lock);
3915         vcpu->arch.ceded = 0;
3916         vcpu->arch.run_task = current;
3917         vcpu->arch.kvm_run = kvm_run;
3918         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3919         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3920         vcpu->arch.busy_preempt = TB_NIL;
3921         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3922         ++vc->n_runnable;
3923
3924         /*
3925          * This happens the first time this is called for a vcpu.
3926          * If the vcore is already running, we may be able to start
3927          * this thread straight away and have it join in.
3928          */
3929         if (!signal_pending(current)) {
3930                 if ((vc->vcore_state == VCORE_PIGGYBACK ||
3931                      vc->vcore_state == VCORE_RUNNING) &&
3932                            !VCORE_IS_EXITING(vc)) {
3933                         kvmppc_create_dtl_entry(vcpu, vc);
3934                         kvmppc_start_thread(vcpu, vc);
3935                         trace_kvm_guest_enter(vcpu);
3936                 } else if (vc->vcore_state == VCORE_SLEEPING) {
3937                         rcuwait_wake_up(&vc->wait);
3938                 }
3939
3940         }
3941
3942         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3943                !signal_pending(current)) {
3944                 /* See if the MMU is ready to go */
3945                 if (!vcpu->kvm->arch.mmu_ready) {
3946                         spin_unlock(&vc->lock);
3947                         r = kvmhv_setup_mmu(vcpu);
3948                         spin_lock(&vc->lock);
3949                         if (r) {
3950                                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3951                                 kvm_run->fail_entry.
3952                                         hardware_entry_failure_reason = 0;
3953                                 vcpu->arch.ret = r;
3954                                 break;
3955                         }
3956                 }
3957
3958                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3959                         kvmppc_vcore_end_preempt(vc);
3960
3961                 if (vc->vcore_state != VCORE_INACTIVE) {
3962                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3963                         continue;
3964                 }
3965                 for_each_runnable_thread(i, v, vc) {
3966                         kvmppc_core_prepare_to_enter(v);
3967                         if (signal_pending(v->arch.run_task)) {
3968                                 kvmppc_remove_runnable(vc, v);
3969                                 v->stat.signal_exits++;
3970                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3971                                 v->arch.ret = -EINTR;
3972                                 wake_up(&v->arch.cpu_run);
3973                         }
3974                 }
3975                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3976                         break;
3977                 n_ceded = 0;
3978                 for_each_runnable_thread(i, v, vc) {
3979                         if (!kvmppc_vcpu_woken(v))
3980                                 n_ceded += v->arch.ceded;
3981                         else
3982                                 v->arch.ceded = 0;
3983                 }
3984                 vc->runner = vcpu;
3985                 if (n_ceded == vc->n_runnable) {
3986                         kvmppc_vcore_blocked(vc);
3987                 } else if (need_resched()) {
3988                         kvmppc_vcore_preempt(vc);
3989                         /* Let something else run */
3990                         cond_resched_lock(&vc->lock);
3991                         if (vc->vcore_state == VCORE_PREEMPT)
3992                                 kvmppc_vcore_end_preempt(vc);
3993                 } else {
3994                         kvmppc_run_core(vc);
3995                 }
3996                 vc->runner = NULL;
3997         }
3998
3999         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4000                (vc->vcore_state == VCORE_RUNNING ||
4001                 vc->vcore_state == VCORE_EXITING ||
4002                 vc->vcore_state == VCORE_PIGGYBACK))
4003                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4004
4005         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4006                 kvmppc_vcore_end_preempt(vc);
4007
4008         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4009                 kvmppc_remove_runnable(vc, vcpu);
4010                 vcpu->stat.signal_exits++;
4011                 kvm_run->exit_reason = KVM_EXIT_INTR;
4012                 vcpu->arch.ret = -EINTR;
4013         }
4014
4015         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4016                 /* Wake up some vcpu to run the core */
4017                 i = -1;
4018                 v = next_runnable_thread(vc, &i);
4019                 wake_up(&v->arch.cpu_run);
4020         }
4021
4022         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4023         spin_unlock(&vc->lock);
4024         return vcpu->arch.ret;
4025 }
4026
4027 int kvmhv_run_single_vcpu(struct kvm_run *kvm_run,
4028                           struct kvm_vcpu *vcpu, u64 time_limit,
4029                           unsigned long lpcr)
4030 {
4031         int trap, r, pcpu;
4032         int srcu_idx, lpid;
4033         struct kvmppc_vcore *vc;
4034         struct kvm *kvm = vcpu->kvm;
4035         struct kvm_nested_guest *nested = vcpu->arch.nested;
4036
4037         trace_kvmppc_run_vcpu_enter(vcpu);
4038
4039         kvm_run->exit_reason = 0;
4040         vcpu->arch.ret = RESUME_GUEST;
4041         vcpu->arch.trap = 0;
4042
4043         vc = vcpu->arch.vcore;
4044         vcpu->arch.ceded = 0;
4045         vcpu->arch.run_task = current;
4046         vcpu->arch.kvm_run = kvm_run;
4047         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4048         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4049         vcpu->arch.busy_preempt = TB_NIL;
4050         vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4051         vc->runnable_threads[0] = vcpu;
4052         vc->n_runnable = 1;
4053         vc->runner = vcpu;
4054
4055         /* See if the MMU is ready to go */
4056         if (!kvm->arch.mmu_ready)
4057                 kvmhv_setup_mmu(vcpu);
4058
4059         if (need_resched())
4060                 cond_resched();
4061
4062         kvmppc_update_vpas(vcpu);
4063
4064         init_vcore_to_run(vc);
4065         vc->preempt_tb = TB_NIL;
4066
4067         preempt_disable();
4068         pcpu = smp_processor_id();
4069         vc->pcpu = pcpu;
4070         kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4071
4072         local_irq_disable();
4073         hard_irq_disable();
4074         if (signal_pending(current))
4075                 goto sigpend;
4076         if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4077                 goto out;
4078
4079         if (!nested) {
4080                 kvmppc_core_prepare_to_enter(vcpu);
4081                 if (vcpu->arch.doorbell_request) {
4082                         vc->dpdes = 1;
4083                         smp_wmb();
4084                         vcpu->arch.doorbell_request = 0;
4085                 }
4086                 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4087                              &vcpu->arch.pending_exceptions))
4088                         lpcr |= LPCR_MER;
4089         } else if (vcpu->arch.pending_exceptions ||
4090                    vcpu->arch.doorbell_request ||
4091                    xive_interrupt_pending(vcpu)) {
4092                 vcpu->arch.ret = RESUME_HOST;
4093                 goto out;
4094         }
4095
4096         kvmppc_clear_host_core(pcpu);
4097
4098         local_paca->kvm_hstate.tid = 0;
4099         local_paca->kvm_hstate.napping = 0;
4100         local_paca->kvm_hstate.kvm_split_mode = NULL;
4101         kvmppc_start_thread(vcpu, vc);
4102         kvmppc_create_dtl_entry(vcpu, vc);
4103         trace_kvm_guest_enter(vcpu);
4104
4105         vc->vcore_state = VCORE_RUNNING;
4106         trace_kvmppc_run_core(vc, 0);
4107
4108         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4109                 lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
4110                 mtspr(SPRN_LPID, lpid);
4111                 isync();
4112                 kvmppc_check_need_tlb_flush(kvm, pcpu, nested);
4113         }
4114
4115         guest_enter_irqoff();
4116
4117         srcu_idx = srcu_read_lock(&kvm->srcu);
4118
4119         this_cpu_disable_ftrace();
4120
4121         /* Tell lockdep that we're about to enable interrupts */
4122         trace_hardirqs_on();
4123
4124         trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4125         vcpu->arch.trap = trap;
4126
4127         trace_hardirqs_off();
4128
4129         this_cpu_enable_ftrace();
4130
4131         srcu_read_unlock(&kvm->srcu, srcu_idx);
4132
4133         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4134                 mtspr(SPRN_LPID, kvm->arch.host_lpid);
4135                 isync();
4136         }
4137
4138         set_irq_happened(trap);
4139
4140         kvmppc_set_host_core(pcpu);
4141
4142         local_irq_enable();
4143         guest_exit();
4144
4145         cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4146
4147         preempt_enable();
4148
4149         /*
4150          * cancel pending decrementer exception if DEC is now positive, or if
4151          * entering a nested guest in which case the decrementer is now owned
4152          * by L2 and the L1 decrementer is provided in hdec_expires
4153          */
4154         if (kvmppc_core_pending_dec(vcpu) &&
4155                         ((get_tb() < vcpu->arch.dec_expires) ||
4156                          (trap == BOOK3S_INTERRUPT_SYSCALL &&
4157                           kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4158                 kvmppc_core_dequeue_dec(vcpu);
4159
4160         trace_kvm_guest_exit(vcpu);
4161         r = RESUME_GUEST;
4162         if (trap) {
4163                 if (!nested)
4164                         r = kvmppc_handle_exit_hv(kvm_run, vcpu, current);
4165                 else
4166                         r = kvmppc_handle_nested_exit(kvm_run, vcpu);
4167         }
4168         vcpu->arch.ret = r;
4169
4170         if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4171             !kvmppc_vcpu_woken(vcpu)) {
4172                 kvmppc_set_timer(vcpu);
4173                 while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4174                         if (signal_pending(current)) {
4175                                 vcpu->stat.signal_exits++;
4176                                 kvm_run->exit_reason = KVM_EXIT_INTR;
4177                                 vcpu->arch.ret = -EINTR;
4178                                 break;
4179                         }
4180                         spin_lock(&vc->lock);
4181                         kvmppc_vcore_blocked(vc);
4182                         spin_unlock(&vc->lock);
4183                 }
4184         }
4185         vcpu->arch.ceded = 0;
4186
4187         vc->vcore_state = VCORE_INACTIVE;
4188         trace_kvmppc_run_core(vc, 1);
4189
4190  done:
4191         kvmppc_remove_runnable(vc, vcpu);
4192         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4193
4194         return vcpu->arch.ret;
4195
4196  sigpend:
4197         vcpu->stat.signal_exits++;
4198         kvm_run->exit_reason = KVM_EXIT_INTR;
4199         vcpu->arch.ret = -EINTR;
4200  out:
4201         local_irq_enable();
4202         preempt_enable();
4203         goto done;
4204 }
4205
4206 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
4207 {
4208         int r;
4209         int srcu_idx;
4210         unsigned long ebb_regs[3] = {}; /* shut up GCC */
4211         unsigned long user_tar = 0;
4212         unsigned int user_vrsave;
4213         struct kvm *kvm;
4214
4215         if (!vcpu->arch.sane) {
4216                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4217                 return -EINVAL;
4218         }
4219
4220         /*
4221          * Don't allow entry with a suspended transaction, because
4222          * the guest entry/exit code will lose it.
4223          * If the guest has TM enabled, save away their TM-related SPRs
4224          * (they will get restored by the TM unavailable interrupt).
4225          */
4226 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4227         if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4228             (current->thread.regs->msr & MSR_TM)) {
4229                 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4230                         run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4231                         run->fail_entry.hardware_entry_failure_reason = 0;
4232                         return -EINVAL;
4233                 }
4234                 /* Enable TM so we can read the TM SPRs */
4235                 mtmsr(mfmsr() | MSR_TM);
4236                 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4237                 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4238                 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4239                 current->thread.regs->msr &= ~MSR_TM;
4240         }
4241 #endif
4242
4243         /*
4244          * Force online to 1 for the sake of old userspace which doesn't
4245          * set it.
4246          */
4247         if (!vcpu->arch.online) {
4248                 atomic_inc(&vcpu->arch.vcore->online_count);
4249                 vcpu->arch.online = 1;
4250         }
4251
4252         kvmppc_core_prepare_to_enter(vcpu);
4253
4254         /* No need to go into the guest when all we'll do is come back out */
4255         if (signal_pending(current)) {
4256                 run->exit_reason = KVM_EXIT_INTR;
4257                 return -EINTR;
4258         }
4259
4260         kvm = vcpu->kvm;
4261         atomic_inc(&kvm->arch.vcpus_running);
4262         /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4263         smp_mb();
4264
4265         flush_all_to_thread(current);
4266
4267         /* Save userspace EBB and other register values */
4268         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4269                 ebb_regs[0] = mfspr(SPRN_EBBHR);
4270                 ebb_regs[1] = mfspr(SPRN_EBBRR);
4271                 ebb_regs[2] = mfspr(SPRN_BESCR);
4272                 user_tar = mfspr(SPRN_TAR);
4273         }
4274         user_vrsave = mfspr(SPRN_VRSAVE);
4275
4276         vcpu->arch.waitp = &vcpu->arch.vcore->wait;
4277         vcpu->arch.pgdir = kvm->mm->pgd;
4278         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4279
4280         do {
4281                 /*
4282                  * The early POWER9 chips that can't mix radix and HPT threads
4283                  * on the same core also need the workaround for the problem
4284                  * where the TLB would prefetch entries in the guest exit path
4285                  * for radix guests using the guest PIDR value and LPID 0.
4286                  * The workaround is in the old path (kvmppc_run_vcpu())
4287                  * but not the new path (kvmhv_run_single_vcpu()).
4288                  */
4289                 if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4290                     !no_mixing_hpt_and_radix)
4291                         r = kvmhv_run_single_vcpu(run, vcpu, ~(u64)0,
4292                                                   vcpu->arch.vcore->lpcr);
4293                 else
4294                         r = kvmppc_run_vcpu(run, vcpu);
4295
4296                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4297                     !(vcpu->arch.shregs.msr & MSR_PR)) {
4298                         trace_kvm_hcall_enter(vcpu);
4299                         r = kvmppc_pseries_do_hcall(vcpu);
4300                         trace_kvm_hcall_exit(vcpu, r);
4301                         kvmppc_core_prepare_to_enter(vcpu);
4302                 } else if (r == RESUME_PAGE_FAULT) {
4303                         srcu_idx = srcu_read_lock(&kvm->srcu);
4304                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
4305                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4306                         srcu_read_unlock(&kvm->srcu, srcu_idx);
4307                 } else if (r == RESUME_PASSTHROUGH) {
4308                         if (WARN_ON(xics_on_xive()))
4309                                 r = H_SUCCESS;
4310                         else
4311                                 r = kvmppc_xics_rm_complete(vcpu, 0);
4312                 }
4313         } while (is_kvmppc_resume_guest(r));
4314
4315         /* Restore userspace EBB and other register values */
4316         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4317                 mtspr(SPRN_EBBHR, ebb_regs[0]);
4318                 mtspr(SPRN_EBBRR, ebb_regs[1]);
4319                 mtspr(SPRN_BESCR, ebb_regs[2]);
4320                 mtspr(SPRN_TAR, user_tar);
4321                 mtspr(SPRN_FSCR, current->thread.fscr);
4322         }
4323         mtspr(SPRN_VRSAVE, user_vrsave);
4324
4325         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4326         atomic_dec(&kvm->arch.vcpus_running);
4327         return r;
4328 }
4329
4330 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4331                                      int shift, int sllp)
4332 {
4333         (*sps)->page_shift = shift;
4334         (*sps)->slb_enc = sllp;
4335         (*sps)->enc[0].page_shift = shift;
4336         (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4337         /*
4338          * Add 16MB MPSS support (may get filtered out by userspace)
4339          */
4340         if (shift != 24) {
4341                 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4342                 if (penc != -1) {
4343                         (*sps)->enc[1].page_shift = 24;
4344                         (*sps)->enc[1].pte_enc = penc;
4345                 }
4346         }
4347         (*sps)++;
4348 }
4349
4350 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4351                                          struct kvm_ppc_smmu_info *info)
4352 {
4353         struct kvm_ppc_one_seg_page_size *sps;
4354
4355         /*
4356          * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4357          * POWER7 doesn't support keys for instruction accesses,
4358          * POWER8 and POWER9 do.
4359          */
4360         info->data_keys = 32;
4361         info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4362
4363         /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4364         info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4365         info->slb_size = 32;
4366
4367         /* We only support these sizes for now, and no muti-size segments */
4368         sps = &info->sps[0];
4369         kvmppc_add_seg_page_size(&sps, 12, 0);
4370         kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4371         kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4372
4373         /* If running as a nested hypervisor, we don't support HPT guests */
4374         if (kvmhv_on_pseries())
4375                 info->flags |= KVM_PPC_NO_HASH;
4376
4377         return 0;
4378 }
4379
4380 /*
4381  * Get (and clear) the dirty memory log for a memory slot.
4382  */
4383 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4384                                          struct kvm_dirty_log *log)
4385 {
4386         struct kvm_memslots *slots;
4387         struct kvm_memory_slot *memslot;
4388         int i, r;
4389         unsigned long n;
4390         unsigned long *buf, *p;
4391         struct kvm_vcpu *vcpu;
4392
4393         mutex_lock(&kvm->slots_lock);
4394
4395         r = -EINVAL;
4396         if (log->slot >= KVM_USER_MEM_SLOTS)
4397                 goto out;
4398
4399         slots = kvm_memslots(kvm);
4400         memslot = id_to_memslot(slots, log->slot);
4401         r = -ENOENT;
4402         if (!memslot || !memslot->dirty_bitmap)
4403                 goto out;
4404
4405         /*
4406          * Use second half of bitmap area because both HPT and radix
4407          * accumulate bits in the first half.
4408          */
4409         n = kvm_dirty_bitmap_bytes(memslot);
4410         buf = memslot->dirty_bitmap + n / sizeof(long);
4411         memset(buf, 0, n);
4412
4413         if (kvm_is_radix(kvm))
4414                 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4415         else
4416                 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4417         if (r)
4418                 goto out;
4419
4420         /*
4421          * We accumulate dirty bits in the first half of the
4422          * memslot's dirty_bitmap area, for when pages are paged
4423          * out or modified by the host directly.  Pick up these
4424          * bits and add them to the map.
4425          */
4426         p = memslot->dirty_bitmap;
4427         for (i = 0; i < n / sizeof(long); ++i)
4428                 buf[i] |= xchg(&p[i], 0);
4429
4430         /* Harvest dirty bits from VPA and DTL updates */
4431         /* Note: we never modify the SLB shadow buffer areas */
4432         kvm_for_each_vcpu(i, vcpu, kvm) {
4433                 spin_lock(&vcpu->arch.vpa_update_lock);
4434                 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4435                 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4436                 spin_unlock(&vcpu->arch.vpa_update_lock);
4437         }
4438
4439         r = -EFAULT;
4440         if (copy_to_user(log->dirty_bitmap, buf, n))
4441                 goto out;
4442
4443         r = 0;
4444 out:
4445         mutex_unlock(&kvm->slots_lock);
4446         return r;
4447 }
4448
4449 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot)
4450 {
4451         vfree(slot->arch.rmap);
4452         slot->arch.rmap = NULL;
4453 }
4454
4455 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4456                                         struct kvm_memory_slot *slot,
4457                                         const struct kvm_userspace_memory_region *mem,
4458                                         enum kvm_mr_change change)
4459 {
4460         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4461
4462         if (change == KVM_MR_CREATE) {
4463                 slot->arch.rmap = vzalloc(array_size(npages,
4464                                           sizeof(*slot->arch.rmap)));
4465                 if (!slot->arch.rmap)
4466                         return -ENOMEM;
4467         }
4468
4469         return 0;
4470 }
4471
4472 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4473                                 const struct kvm_userspace_memory_region *mem,
4474                                 const struct kvm_memory_slot *old,
4475                                 const struct kvm_memory_slot *new,
4476                                 enum kvm_mr_change change)
4477 {
4478         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4479
4480         /*
4481          * If we are making a new memslot, it might make
4482          * some address that was previously cached as emulated
4483          * MMIO be no longer emulated MMIO, so invalidate
4484          * all the caches of emulated MMIO translations.
4485          */
4486         if (npages)
4487                 atomic64_inc(&kvm->arch.mmio_update);
4488
4489         /*
4490          * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4491          * have already called kvm_arch_flush_shadow_memslot() to
4492          * flush shadow mappings.  For KVM_MR_CREATE we have no
4493          * previous mappings.  So the only case to handle is
4494          * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4495          * has been changed.
4496          * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4497          * to get rid of any THP PTEs in the partition-scoped page tables
4498          * so we can track dirtiness at the page level; we flush when
4499          * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4500          * using THP PTEs.
4501          */
4502         if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4503             ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4504                 kvmppc_radix_flush_memslot(kvm, old);
4505         /*
4506          * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
4507          */
4508         if (!kvm->arch.secure_guest)
4509                 return;
4510
4511         switch (change) {
4512         case KVM_MR_CREATE:
4513                 if (kvmppc_uvmem_slot_init(kvm, new))
4514                         return;
4515                 uv_register_mem_slot(kvm->arch.lpid,
4516                                      new->base_gfn << PAGE_SHIFT,
4517                                      new->npages * PAGE_SIZE,
4518                                      0, new->id);
4519                 break;
4520         case KVM_MR_DELETE:
4521                 uv_unregister_mem_slot(kvm->arch.lpid, old->id);
4522                 kvmppc_uvmem_slot_free(kvm, old);
4523                 break;
4524         default:
4525                 /* TODO: Handle KVM_MR_MOVE */
4526                 break;
4527         }
4528 }
4529
4530 /*
4531  * Update LPCR values in kvm->arch and in vcores.
4532  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4533  * of kvm->arch.lpcr update).
4534  */
4535 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4536 {
4537         long int i;
4538         u32 cores_done = 0;
4539
4540         if ((kvm->arch.lpcr & mask) == lpcr)
4541                 return;
4542
4543         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4544
4545         for (i = 0; i < KVM_MAX_VCORES; ++i) {
4546                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4547                 if (!vc)
4548                         continue;
4549                 spin_lock(&vc->lock);
4550                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4551                 spin_unlock(&vc->lock);
4552                 if (++cores_done >= kvm->arch.online_vcores)
4553                         break;
4554         }
4555 }
4556
4557 void kvmppc_setup_partition_table(struct kvm *kvm)
4558 {
4559         unsigned long dw0, dw1;
4560
4561         if (!kvm_is_radix(kvm)) {
4562                 /* PS field - page size for VRMA */
4563                 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4564                         ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4565                 /* HTABSIZE and HTABORG fields */
4566                 dw0 |= kvm->arch.sdr1;
4567
4568                 /* Second dword as set by userspace */
4569                 dw1 = kvm->arch.process_table;
4570         } else {
4571                 dw0 = PATB_HR | radix__get_tree_size() |
4572                         __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4573                 dw1 = PATB_GR | kvm->arch.process_table;
4574         }
4575         kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4576 }
4577
4578 /*
4579  * Set up HPT (hashed page table) and RMA (real-mode area).
4580  * Must be called with kvm->arch.mmu_setup_lock held.
4581  */
4582 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4583 {
4584         int err = 0;
4585         struct kvm *kvm = vcpu->kvm;
4586         unsigned long hva;
4587         struct kvm_memory_slot *memslot;
4588         struct vm_area_struct *vma;
4589         unsigned long lpcr = 0, senc;
4590         unsigned long psize, porder;
4591         int srcu_idx;
4592
4593         /* Allocate hashed page table (if not done already) and reset it */
4594         if (!kvm->arch.hpt.virt) {
4595                 int order = KVM_DEFAULT_HPT_ORDER;
4596                 struct kvm_hpt_info info;
4597
4598                 err = kvmppc_allocate_hpt(&info, order);
4599                 /* If we get here, it means userspace didn't specify a
4600                  * size explicitly.  So, try successively smaller
4601                  * sizes if the default failed. */
4602                 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4603                         err  = kvmppc_allocate_hpt(&info, order);
4604
4605                 if (err < 0) {
4606                         pr_err("KVM: Couldn't alloc HPT\n");
4607                         goto out;
4608                 }
4609
4610                 kvmppc_set_hpt(kvm, &info);
4611         }
4612
4613         /* Look up the memslot for guest physical address 0 */
4614         srcu_idx = srcu_read_lock(&kvm->srcu);
4615         memslot = gfn_to_memslot(kvm, 0);
4616
4617         /* We must have some memory at 0 by now */
4618         err = -EINVAL;
4619         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4620                 goto out_srcu;
4621
4622         /* Look up the VMA for the start of this memory slot */
4623         hva = memslot->userspace_addr;
4624         down_read(&kvm->mm->mmap_sem);
4625         vma = find_vma(kvm->mm, hva);
4626         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4627                 goto up_out;
4628
4629         psize = vma_kernel_pagesize(vma);
4630
4631         up_read(&kvm->mm->mmap_sem);
4632
4633         /* We can handle 4k, 64k or 16M pages in the VRMA */
4634         if (psize >= 0x1000000)
4635                 psize = 0x1000000;
4636         else if (psize >= 0x10000)
4637                 psize = 0x10000;
4638         else
4639                 psize = 0x1000;
4640         porder = __ilog2(psize);
4641
4642         senc = slb_pgsize_encoding(psize);
4643         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4644                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4645         /* Create HPTEs in the hash page table for the VRMA */
4646         kvmppc_map_vrma(vcpu, memslot, porder);
4647
4648         /* Update VRMASD field in the LPCR */
4649         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4650                 /* the -4 is to account for senc values starting at 0x10 */
4651                 lpcr = senc << (LPCR_VRMASD_SH - 4);
4652                 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4653         }
4654
4655         /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4656         smp_wmb();
4657         err = 0;
4658  out_srcu:
4659         srcu_read_unlock(&kvm->srcu, srcu_idx);
4660  out:
4661         return err;
4662
4663  up_out:
4664         up_read(&kvm->mm->mmap_sem);
4665         goto out_srcu;
4666 }
4667
4668 /*
4669  * Must be called with kvm->arch.mmu_setup_lock held and
4670  * mmu_ready = 0 and no vcpus running.
4671  */
4672 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4673 {
4674         if (nesting_enabled(kvm))
4675                 kvmhv_release_all_nested(kvm);
4676         kvmppc_rmap_reset(kvm);
4677         kvm->arch.process_table = 0;
4678         /* Mutual exclusion with kvm_unmap_hva_range etc. */
4679         spin_lock(&kvm->mmu_lock);
4680         kvm->arch.radix = 0;
4681         spin_unlock(&kvm->mmu_lock);
4682         kvmppc_free_radix(kvm);
4683         kvmppc_update_lpcr(kvm, LPCR_VPM1,
4684                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4685         return 0;
4686 }
4687
4688 /*
4689  * Must be called with kvm->arch.mmu_setup_lock held and
4690  * mmu_ready = 0 and no vcpus running.
4691  */
4692 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4693 {
4694         int err;
4695
4696         err = kvmppc_init_vm_radix(kvm);
4697         if (err)
4698                 return err;
4699         kvmppc_rmap_reset(kvm);
4700         /* Mutual exclusion with kvm_unmap_hva_range etc. */
4701         spin_lock(&kvm->mmu_lock);
4702         kvm->arch.radix = 1;
4703         spin_unlock(&kvm->mmu_lock);
4704         kvmppc_free_hpt(&kvm->arch.hpt);
4705         kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4706                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4707         return 0;
4708 }
4709
4710 #ifdef CONFIG_KVM_XICS
4711 /*
4712  * Allocate a per-core structure for managing state about which cores are
4713  * running in the host versus the guest and for exchanging data between
4714  * real mode KVM and CPU running in the host.
4715  * This is only done for the first VM.
4716  * The allocated structure stays even if all VMs have stopped.
4717  * It is only freed when the kvm-hv module is unloaded.
4718  * It's OK for this routine to fail, we just don't support host
4719  * core operations like redirecting H_IPI wakeups.
4720  */
4721 void kvmppc_alloc_host_rm_ops(void)
4722 {
4723         struct kvmppc_host_rm_ops *ops;
4724         unsigned long l_ops;
4725         int cpu, core;
4726         int size;
4727
4728         /* Not the first time here ? */
4729         if (kvmppc_host_rm_ops_hv != NULL)
4730                 return;
4731
4732         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4733         if (!ops)
4734                 return;
4735
4736         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4737         ops->rm_core = kzalloc(size, GFP_KERNEL);
4738
4739         if (!ops->rm_core) {
4740                 kfree(ops);
4741                 return;
4742         }
4743
4744         cpus_read_lock();
4745
4746         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4747                 if (!cpu_online(cpu))
4748                         continue;
4749
4750                 core = cpu >> threads_shift;
4751                 ops->rm_core[core].rm_state.in_host = 1;
4752         }
4753
4754         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4755
4756         /*
4757          * Make the contents of the kvmppc_host_rm_ops structure visible
4758          * to other CPUs before we assign it to the global variable.
4759          * Do an atomic assignment (no locks used here), but if someone
4760          * beats us to it, just free our copy and return.
4761          */
4762         smp_wmb();
4763         l_ops = (unsigned long) ops;
4764
4765         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4766                 cpus_read_unlock();
4767                 kfree(ops->rm_core);
4768                 kfree(ops);
4769                 return;
4770         }
4771
4772         cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4773                                              "ppc/kvm_book3s:prepare",
4774                                              kvmppc_set_host_core,
4775                                              kvmppc_clear_host_core);
4776         cpus_read_unlock();
4777 }
4778
4779 void kvmppc_free_host_rm_ops(void)
4780 {
4781         if (kvmppc_host_rm_ops_hv) {
4782                 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4783                 kfree(kvmppc_host_rm_ops_hv->rm_core);
4784                 kfree(kvmppc_host_rm_ops_hv);
4785                 kvmppc_host_rm_ops_hv = NULL;
4786         }
4787 }
4788 #endif
4789
4790 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4791 {
4792         unsigned long lpcr, lpid;
4793         char buf[32];
4794         int ret;
4795
4796         mutex_init(&kvm->arch.uvmem_lock);
4797         INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
4798         mutex_init(&kvm->arch.mmu_setup_lock);
4799
4800         /* Allocate the guest's logical partition ID */
4801
4802         lpid = kvmppc_alloc_lpid();
4803         if ((long)lpid < 0)
4804                 return -ENOMEM;
4805         kvm->arch.lpid = lpid;
4806
4807         kvmppc_alloc_host_rm_ops();
4808
4809         kvmhv_vm_nested_init(kvm);
4810
4811         /*
4812          * Since we don't flush the TLB when tearing down a VM,
4813          * and this lpid might have previously been used,
4814          * make sure we flush on each core before running the new VM.
4815          * On POWER9, the tlbie in mmu_partition_table_set_entry()
4816          * does this flush for us.
4817          */
4818         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4819                 cpumask_setall(&kvm->arch.need_tlb_flush);
4820
4821         /* Start out with the default set of hcalls enabled */
4822         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4823                sizeof(kvm->arch.enabled_hcalls));
4824
4825         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4826                 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4827
4828         /* Init LPCR for virtual RMA mode */
4829         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4830                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
4831                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4832                 lpcr &= LPCR_PECE | LPCR_LPES;
4833         } else {
4834                 lpcr = 0;
4835         }
4836         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4837                 LPCR_VPM0 | LPCR_VPM1;
4838         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4839                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4840         /* On POWER8 turn on online bit to enable PURR/SPURR */
4841         if (cpu_has_feature(CPU_FTR_ARCH_207S))
4842                 lpcr |= LPCR_ONL;
4843         /*
4844          * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4845          * Set HVICE bit to enable hypervisor virtualization interrupts.
4846          * Set HEIC to prevent OS interrupts to go to hypervisor (should
4847          * be unnecessary but better safe than sorry in case we re-enable
4848          * EE in HV mode with this LPCR still set)
4849          */
4850         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4851                 lpcr &= ~LPCR_VPM0;
4852                 lpcr |= LPCR_HVICE | LPCR_HEIC;
4853
4854                 /*
4855                  * If xive is enabled, we route 0x500 interrupts directly
4856                  * to the guest.
4857                  */
4858                 if (xics_on_xive())
4859                         lpcr |= LPCR_LPES;
4860         }
4861
4862         /*
4863          * If the host uses radix, the guest starts out as radix.
4864          */
4865         if (radix_enabled()) {
4866                 kvm->arch.radix = 1;
4867                 kvm->arch.mmu_ready = 1;
4868                 lpcr &= ~LPCR_VPM1;
4869                 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4870                 ret = kvmppc_init_vm_radix(kvm);
4871                 if (ret) {
4872                         kvmppc_free_lpid(kvm->arch.lpid);
4873                         return ret;
4874                 }
4875                 kvmppc_setup_partition_table(kvm);
4876         }
4877
4878         kvm->arch.lpcr = lpcr;
4879
4880         /* Initialization for future HPT resizes */
4881         kvm->arch.resize_hpt = NULL;
4882
4883         /*
4884          * Work out how many sets the TLB has, for the use of
4885          * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4886          */
4887         if (radix_enabled())
4888                 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;     /* 128 */
4889         else if (cpu_has_feature(CPU_FTR_ARCH_300))
4890                 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;      /* 256 */
4891         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4892                 kvm->arch.tlb_sets = POWER8_TLB_SETS;           /* 512 */
4893         else
4894                 kvm->arch.tlb_sets = POWER7_TLB_SETS;           /* 128 */
4895
4896         /*
4897          * Track that we now have a HV mode VM active. This blocks secondary
4898          * CPU threads from coming online.
4899          * On POWER9, we only need to do this if the "indep_threads_mode"
4900          * module parameter has been set to N.
4901          */
4902         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4903                 if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
4904                         pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
4905                         kvm->arch.threads_indep = true;
4906                 } else {
4907                         kvm->arch.threads_indep = indep_threads_mode;
4908                 }
4909         }
4910         if (!kvm->arch.threads_indep)
4911                 kvm_hv_vm_activated();
4912
4913         /*
4914          * Initialize smt_mode depending on processor.
4915          * POWER8 and earlier have to use "strict" threading, where
4916          * all vCPUs in a vcore have to run on the same (sub)core,
4917          * whereas on POWER9 the threads can each run a different
4918          * guest.
4919          */
4920         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4921                 kvm->arch.smt_mode = threads_per_subcore;
4922         else
4923                 kvm->arch.smt_mode = 1;
4924         kvm->arch.emul_smt_mode = 1;
4925
4926         /*
4927          * Create a debugfs directory for the VM
4928          */
4929         snprintf(buf, sizeof(buf), "vm%d", current->pid);
4930         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4931         kvmppc_mmu_debugfs_init(kvm);
4932         if (radix_enabled())
4933                 kvmhv_radix_debugfs_init(kvm);
4934
4935         return 0;
4936 }
4937
4938 static void kvmppc_free_vcores(struct kvm *kvm)
4939 {
4940         long int i;
4941
4942         for (i = 0; i < KVM_MAX_VCORES; ++i)
4943                 kfree(kvm->arch.vcores[i]);
4944         kvm->arch.online_vcores = 0;
4945 }
4946
4947 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4948 {
4949         debugfs_remove_recursive(kvm->arch.debugfs_dir);
4950
4951         if (!kvm->arch.threads_indep)
4952                 kvm_hv_vm_deactivated();
4953
4954         kvmppc_free_vcores(kvm);
4955
4956
4957         if (kvm_is_radix(kvm))
4958                 kvmppc_free_radix(kvm);
4959         else
4960                 kvmppc_free_hpt(&kvm->arch.hpt);
4961
4962         /* Perform global invalidation and return lpid to the pool */
4963         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4964                 if (nesting_enabled(kvm))
4965                         kvmhv_release_all_nested(kvm);
4966                 kvm->arch.process_table = 0;
4967                 if (kvm->arch.secure_guest)
4968                         uv_svm_terminate(kvm->arch.lpid);
4969                 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
4970         }
4971
4972         kvmppc_free_lpid(kvm->arch.lpid);
4973
4974         kvmppc_free_pimap(kvm);
4975 }
4976
4977 /* We don't need to emulate any privileged instructions or dcbz */
4978 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
4979                                      unsigned int inst, int *advance)
4980 {
4981         return EMULATE_FAIL;
4982 }
4983
4984 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
4985                                         ulong spr_val)
4986 {
4987         return EMULATE_FAIL;
4988 }
4989
4990 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
4991                                         ulong *spr_val)
4992 {
4993         return EMULATE_FAIL;
4994 }
4995
4996 static int kvmppc_core_check_processor_compat_hv(void)
4997 {
4998         if (cpu_has_feature(CPU_FTR_HVMODE) &&
4999             cpu_has_feature(CPU_FTR_ARCH_206))
5000                 return 0;
5001
5002         /* POWER9 in radix mode is capable of being a nested hypervisor. */
5003         if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5004                 return 0;
5005
5006         return -EIO;
5007 }
5008
5009 #ifdef CONFIG_KVM_XICS
5010
5011 void kvmppc_free_pimap(struct kvm *kvm)
5012 {
5013         kfree(kvm->arch.pimap);
5014 }
5015
5016 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5017 {
5018         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5019 }
5020
5021 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5022 {
5023         struct irq_desc *desc;
5024         struct kvmppc_irq_map *irq_map;
5025         struct kvmppc_passthru_irqmap *pimap;
5026         struct irq_chip *chip;
5027         int i, rc = 0;
5028
5029         if (!kvm_irq_bypass)
5030                 return 1;
5031
5032         desc = irq_to_desc(host_irq);
5033         if (!desc)
5034                 return -EIO;
5035
5036         mutex_lock(&kvm->lock);
5037
5038         pimap = kvm->arch.pimap;
5039         if (pimap == NULL) {
5040                 /* First call, allocate structure to hold IRQ map */
5041                 pimap = kvmppc_alloc_pimap();
5042                 if (pimap == NULL) {
5043                         mutex_unlock(&kvm->lock);
5044                         return -ENOMEM;
5045                 }
5046                 kvm->arch.pimap = pimap;
5047         }
5048
5049         /*
5050          * For now, we only support interrupts for which the EOI operation
5051          * is an OPAL call followed by a write to XIRR, since that's
5052          * what our real-mode EOI code does, or a XIVE interrupt
5053          */
5054         chip = irq_data_get_irq_chip(&desc->irq_data);
5055         if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5056                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5057                         host_irq, guest_gsi);
5058                 mutex_unlock(&kvm->lock);
5059                 return -ENOENT;
5060         }
5061
5062         /*
5063          * See if we already have an entry for this guest IRQ number.
5064          * If it's mapped to a hardware IRQ number, that's an error,
5065          * otherwise re-use this entry.
5066          */
5067         for (i = 0; i < pimap->n_mapped; i++) {
5068                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
5069                         if (pimap->mapped[i].r_hwirq) {
5070                                 mutex_unlock(&kvm->lock);
5071                                 return -EINVAL;
5072                         }
5073                         break;
5074                 }
5075         }
5076
5077         if (i == KVMPPC_PIRQ_MAPPED) {
5078                 mutex_unlock(&kvm->lock);
5079                 return -EAGAIN;         /* table is full */
5080         }
5081
5082         irq_map = &pimap->mapped[i];
5083
5084         irq_map->v_hwirq = guest_gsi;
5085         irq_map->desc = desc;
5086
5087         /*
5088          * Order the above two stores before the next to serialize with
5089          * the KVM real mode handler.
5090          */
5091         smp_wmb();
5092         irq_map->r_hwirq = desc->irq_data.hwirq;
5093
5094         if (i == pimap->n_mapped)
5095                 pimap->n_mapped++;
5096
5097         if (xics_on_xive())
5098                 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5099         else
5100                 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5101         if (rc)
5102                 irq_map->r_hwirq = 0;
5103
5104         mutex_unlock(&kvm->lock);
5105
5106         return 0;
5107 }
5108
5109 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5110 {
5111         struct irq_desc *desc;
5112         struct kvmppc_passthru_irqmap *pimap;
5113         int i, rc = 0;
5114
5115         if (!kvm_irq_bypass)
5116                 return 0;
5117
5118         desc = irq_to_desc(host_irq);
5119         if (!desc)
5120                 return -EIO;
5121
5122         mutex_lock(&kvm->lock);
5123         if (!kvm->arch.pimap)
5124                 goto unlock;
5125
5126         pimap = kvm->arch.pimap;
5127
5128         for (i = 0; i < pimap->n_mapped; i++) {
5129                 if (guest_gsi == pimap->mapped[i].v_hwirq)
5130                         break;
5131         }
5132
5133         if (i == pimap->n_mapped) {
5134                 mutex_unlock(&kvm->lock);
5135                 return -ENODEV;
5136         }
5137
5138         if (xics_on_xive())
5139                 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5140         else
5141                 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5142
5143         /* invalidate the entry (what do do on error from the above ?) */
5144         pimap->mapped[i].r_hwirq = 0;
5145
5146         /*
5147          * We don't free this structure even when the count goes to
5148          * zero. The structure is freed when we destroy the VM.
5149          */
5150  unlock:
5151         mutex_unlock(&kvm->lock);
5152         return rc;
5153 }
5154
5155 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5156                                              struct irq_bypass_producer *prod)
5157 {
5158         int ret = 0;
5159         struct kvm_kernel_irqfd *irqfd =
5160                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5161
5162         irqfd->producer = prod;
5163
5164         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5165         if (ret)
5166                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5167                         prod->irq, irqfd->gsi, ret);
5168
5169         return ret;
5170 }
5171
5172 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5173                                               struct irq_bypass_producer *prod)
5174 {
5175         int ret;
5176         struct kvm_kernel_irqfd *irqfd =
5177                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5178
5179         irqfd->producer = NULL;
5180
5181         /*
5182          * When producer of consumer is unregistered, we change back to
5183          * default external interrupt handling mode - KVM real mode
5184          * will switch back to host.
5185          */
5186         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5187         if (ret)
5188                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5189                         prod->irq, irqfd->gsi, ret);
5190 }
5191 #endif
5192
5193 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5194                                  unsigned int ioctl, unsigned long arg)
5195 {
5196         struct kvm *kvm __maybe_unused = filp->private_data;
5197         void __user *argp = (void __user *)arg;
5198         long r;
5199
5200         switch (ioctl) {
5201
5202         case KVM_PPC_ALLOCATE_HTAB: {
5203                 u32 htab_order;
5204
5205                 r = -EFAULT;
5206                 if (get_user(htab_order, (u32 __user *)argp))
5207                         break;
5208                 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5209                 if (r)
5210                         break;
5211                 r = 0;
5212                 break;
5213         }
5214
5215         case KVM_PPC_GET_HTAB_FD: {
5216                 struct kvm_get_htab_fd ghf;
5217
5218                 r = -EFAULT;
5219                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5220                         break;
5221                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5222                 break;
5223         }
5224
5225         case KVM_PPC_RESIZE_HPT_PREPARE: {
5226                 struct kvm_ppc_resize_hpt rhpt;
5227
5228                 r = -EFAULT;
5229                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5230                         break;
5231
5232                 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5233                 break;
5234         }
5235
5236         case KVM_PPC_RESIZE_HPT_COMMIT: {
5237                 struct kvm_ppc_resize_hpt rhpt;
5238
5239                 r = -EFAULT;
5240                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5241                         break;
5242
5243                 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5244                 break;
5245         }
5246
5247         default:
5248                 r = -ENOTTY;
5249         }
5250
5251         return r;
5252 }
5253
5254 /*
5255  * List of hcall numbers to enable by default.
5256  * For compatibility with old userspace, we enable by default
5257  * all hcalls that were implemented before the hcall-enabling
5258  * facility was added.  Note this list should not include H_RTAS.
5259  */
5260 static unsigned int default_hcall_list[] = {
5261         H_REMOVE,
5262         H_ENTER,
5263         H_READ,
5264         H_PROTECT,
5265         H_BULK_REMOVE,
5266         H_GET_TCE,
5267         H_PUT_TCE,
5268         H_SET_DABR,
5269         H_SET_XDABR,
5270         H_CEDE,
5271         H_PROD,
5272         H_CONFER,
5273         H_REGISTER_VPA,
5274 #ifdef CONFIG_KVM_XICS
5275         H_EOI,
5276         H_CPPR,
5277         H_IPI,
5278         H_IPOLL,
5279         H_XIRR,
5280         H_XIRR_X,
5281 #endif
5282         0
5283 };
5284
5285 static void init_default_hcalls(void)
5286 {
5287         int i;
5288         unsigned int hcall;
5289
5290         for (i = 0; default_hcall_list[i]; ++i) {
5291                 hcall = default_hcall_list[i];
5292                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5293                 __set_bit(hcall / 4, default_enabled_hcalls);
5294         }
5295 }
5296
5297 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5298 {
5299         unsigned long lpcr;
5300         int radix;
5301         int err;
5302
5303         /* If not on a POWER9, reject it */
5304         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5305                 return -ENODEV;
5306
5307         /* If any unknown flags set, reject it */
5308         if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5309                 return -EINVAL;
5310
5311         /* GR (guest radix) bit in process_table field must match */
5312         radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5313         if (!!(cfg->process_table & PATB_GR) != radix)
5314                 return -EINVAL;
5315
5316         /* Process table size field must be reasonable, i.e. <= 24 */
5317         if ((cfg->process_table & PRTS_MASK) > 24)
5318                 return -EINVAL;
5319
5320         /* We can change a guest to/from radix now, if the host is radix */
5321         if (radix && !radix_enabled())
5322                 return -EINVAL;
5323
5324         /* If we're a nested hypervisor, we currently only support radix */
5325         if (kvmhv_on_pseries() && !radix)
5326                 return -EINVAL;
5327
5328         mutex_lock(&kvm->arch.mmu_setup_lock);
5329         if (radix != kvm_is_radix(kvm)) {
5330                 if (kvm->arch.mmu_ready) {
5331                         kvm->arch.mmu_ready = 0;
5332                         /* order mmu_ready vs. vcpus_running */
5333                         smp_mb();
5334                         if (atomic_read(&kvm->arch.vcpus_running)) {
5335                                 kvm->arch.mmu_ready = 1;
5336                                 err = -EBUSY;
5337                                 goto out_unlock;
5338                         }
5339                 }
5340                 if (radix)
5341                         err = kvmppc_switch_mmu_to_radix(kvm);
5342                 else
5343                         err = kvmppc_switch_mmu_to_hpt(kvm);
5344                 if (err)
5345                         goto out_unlock;
5346         }
5347
5348         kvm->arch.process_table = cfg->process_table;
5349         kvmppc_setup_partition_table(kvm);
5350
5351         lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5352         kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5353         err = 0;
5354
5355  out_unlock:
5356         mutex_unlock(&kvm->arch.mmu_setup_lock);
5357         return err;
5358 }
5359
5360 static int kvmhv_enable_nested(struct kvm *kvm)
5361 {
5362         if (!nested)
5363                 return -EPERM;
5364         if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5365                 return -ENODEV;
5366
5367         /* kvm == NULL means the caller is testing if the capability exists */
5368         if (kvm)
5369                 kvm->arch.nested_enable = true;
5370         return 0;
5371 }
5372
5373 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5374                                  int size)
5375 {
5376         int rc = -EINVAL;
5377
5378         if (kvmhv_vcpu_is_radix(vcpu)) {
5379                 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5380
5381                 if (rc > 0)
5382                         rc = -EINVAL;
5383         }
5384
5385         /* For now quadrants are the only way to access nested guest memory */
5386         if (rc && vcpu->arch.nested)
5387                 rc = -EAGAIN;
5388
5389         return rc;
5390 }
5391
5392 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5393                                 int size)
5394 {
5395         int rc = -EINVAL;
5396
5397         if (kvmhv_vcpu_is_radix(vcpu)) {
5398                 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5399
5400                 if (rc > 0)
5401                         rc = -EINVAL;
5402         }
5403
5404         /* For now quadrants are the only way to access nested guest memory */
5405         if (rc && vcpu->arch.nested)
5406                 rc = -EAGAIN;
5407
5408         return rc;
5409 }
5410
5411 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
5412 {
5413         unpin_vpa(kvm, vpa);
5414         vpa->gpa = 0;
5415         vpa->pinned_addr = NULL;
5416         vpa->dirty = false;
5417         vpa->update_pending = 0;
5418 }
5419
5420 /*
5421  * Enable a guest to become a secure VM, or test whether
5422  * that could be enabled.
5423  * Called when the KVM_CAP_PPC_SECURE_GUEST capability is
5424  * tested (kvm == NULL) or enabled (kvm != NULL).
5425  */
5426 static int kvmhv_enable_svm(struct kvm *kvm)
5427 {
5428         if (!kvmppc_uvmem_available())
5429                 return -EINVAL;
5430         if (kvm)
5431                 kvm->arch.svm_enabled = 1;
5432         return 0;
5433 }
5434
5435 /*
5436  *  IOCTL handler to turn off secure mode of guest
5437  *
5438  * - Release all device pages
5439  * - Issue ucall to terminate the guest on the UV side
5440  * - Unpin the VPA pages.
5441  * - Reinit the partition scoped page tables
5442  */
5443 static int kvmhv_svm_off(struct kvm *kvm)
5444 {
5445         struct kvm_vcpu *vcpu;
5446         int mmu_was_ready;
5447         int srcu_idx;
5448         int ret = 0;
5449         int i;
5450
5451         if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
5452                 return ret;
5453
5454         mutex_lock(&kvm->arch.mmu_setup_lock);
5455         mmu_was_ready = kvm->arch.mmu_ready;
5456         if (kvm->arch.mmu_ready) {
5457                 kvm->arch.mmu_ready = 0;
5458                 /* order mmu_ready vs. vcpus_running */
5459                 smp_mb();
5460                 if (atomic_read(&kvm->arch.vcpus_running)) {
5461                         kvm->arch.mmu_ready = 1;
5462                         ret = -EBUSY;
5463                         goto out;
5464                 }
5465         }
5466
5467         srcu_idx = srcu_read_lock(&kvm->srcu);
5468         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
5469                 struct kvm_memory_slot *memslot;
5470                 struct kvm_memslots *slots = __kvm_memslots(kvm, i);
5471
5472                 if (!slots)
5473                         continue;
5474
5475                 kvm_for_each_memslot(memslot, slots) {
5476                         kvmppc_uvmem_drop_pages(memslot, kvm, true);
5477                         uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
5478                 }
5479         }
5480         srcu_read_unlock(&kvm->srcu, srcu_idx);
5481
5482         ret = uv_svm_terminate(kvm->arch.lpid);
5483         if (ret != U_SUCCESS) {
5484                 ret = -EINVAL;
5485                 goto out;
5486         }
5487
5488         /*
5489          * When secure guest is reset, all the guest pages are sent
5490          * to UV via UV_PAGE_IN before the non-boot vcpus get a
5491          * chance to run and unpin their VPA pages. Unpinning of all
5492          * VPA pages is done here explicitly so that VPA pages
5493          * can be migrated to the secure side.
5494          *
5495          * This is required to for the secure SMP guest to reboot
5496          * correctly.
5497          */
5498         kvm_for_each_vcpu(i, vcpu, kvm) {
5499                 spin_lock(&vcpu->arch.vpa_update_lock);
5500                 unpin_vpa_reset(kvm, &vcpu->arch.dtl);
5501                 unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
5502                 unpin_vpa_reset(kvm, &vcpu->arch.vpa);
5503                 spin_unlock(&vcpu->arch.vpa_update_lock);
5504         }
5505
5506         kvmppc_setup_partition_table(kvm);
5507         kvm->arch.secure_guest = 0;
5508         kvm->arch.mmu_ready = mmu_was_ready;
5509 out:
5510         mutex_unlock(&kvm->arch.mmu_setup_lock);
5511         return ret;
5512 }
5513
5514 static struct kvmppc_ops kvm_ops_hv = {
5515         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5516         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5517         .get_one_reg = kvmppc_get_one_reg_hv,
5518         .set_one_reg = kvmppc_set_one_reg_hv,
5519         .vcpu_load   = kvmppc_core_vcpu_load_hv,
5520         .vcpu_put    = kvmppc_core_vcpu_put_hv,
5521         .inject_interrupt = kvmppc_inject_interrupt_hv,
5522         .set_msr     = kvmppc_set_msr_hv,
5523         .vcpu_run    = kvmppc_vcpu_run_hv,
5524         .vcpu_create = kvmppc_core_vcpu_create_hv,
5525         .vcpu_free   = kvmppc_core_vcpu_free_hv,
5526         .check_requests = kvmppc_core_check_requests_hv,
5527         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5528         .flush_memslot  = kvmppc_core_flush_memslot_hv,
5529         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5530         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5531         .unmap_hva_range = kvm_unmap_hva_range_hv,
5532         .age_hva  = kvm_age_hva_hv,
5533         .test_age_hva = kvm_test_age_hva_hv,
5534         .set_spte_hva = kvm_set_spte_hva_hv,
5535         .free_memslot = kvmppc_core_free_memslot_hv,
5536         .init_vm =  kvmppc_core_init_vm_hv,
5537         .destroy_vm = kvmppc_core_destroy_vm_hv,
5538         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5539         .emulate_op = kvmppc_core_emulate_op_hv,
5540         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5541         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5542         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5543         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
5544         .hcall_implemented = kvmppc_hcall_impl_hv,
5545 #ifdef CONFIG_KVM_XICS
5546         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5547         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5548 #endif
5549         .configure_mmu = kvmhv_configure_mmu,
5550         .get_rmmu_info = kvmhv_get_rmmu_info,
5551         .set_smt_mode = kvmhv_set_smt_mode,
5552         .enable_nested = kvmhv_enable_nested,
5553         .load_from_eaddr = kvmhv_load_from_eaddr,
5554         .store_to_eaddr = kvmhv_store_to_eaddr,
5555         .enable_svm = kvmhv_enable_svm,
5556         .svm_off = kvmhv_svm_off,
5557 };
5558
5559 static int kvm_init_subcore_bitmap(void)
5560 {
5561         int i, j;
5562         int nr_cores = cpu_nr_cores();
5563         struct sibling_subcore_state *sibling_subcore_state;
5564
5565         for (i = 0; i < nr_cores; i++) {
5566                 int first_cpu = i * threads_per_core;
5567                 int node = cpu_to_node(first_cpu);
5568
5569                 /* Ignore if it is already allocated. */
5570                 if (paca_ptrs[first_cpu]->sibling_subcore_state)
5571                         continue;
5572
5573                 sibling_subcore_state =
5574                         kzalloc_node(sizeof(struct sibling_subcore_state),
5575                                                         GFP_KERNEL, node);
5576                 if (!sibling_subcore_state)
5577                         return -ENOMEM;
5578
5579
5580                 for (j = 0; j < threads_per_core; j++) {
5581                         int cpu = first_cpu + j;
5582
5583                         paca_ptrs[cpu]->sibling_subcore_state =
5584                                                 sibling_subcore_state;
5585                 }
5586         }
5587         return 0;
5588 }
5589
5590 static int kvmppc_radix_possible(void)
5591 {
5592         return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5593 }
5594
5595 static int kvmppc_book3s_init_hv(void)
5596 {
5597         int r;
5598
5599         if (!tlbie_capable) {
5600                 pr_err("KVM-HV: Host does not support TLBIE\n");
5601                 return -ENODEV;
5602         }
5603
5604         /*
5605          * FIXME!! Do we need to check on all cpus ?
5606          */
5607         r = kvmppc_core_check_processor_compat_hv();
5608         if (r < 0)
5609                 return -ENODEV;
5610
5611         r = kvmhv_nested_init();
5612         if (r)
5613                 return r;
5614
5615         r = kvm_init_subcore_bitmap();
5616         if (r)
5617                 return r;
5618
5619         /*
5620          * We need a way of accessing the XICS interrupt controller,
5621          * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5622          * indirectly, via OPAL.
5623          */
5624 #ifdef CONFIG_SMP
5625         if (!xics_on_xive() && !kvmhv_on_pseries() &&
5626             !local_paca->kvm_hstate.xics_phys) {
5627                 struct device_node *np;
5628
5629                 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5630                 if (!np) {
5631                         pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5632                         return -ENODEV;
5633                 }
5634                 /* presence of intc confirmed - node can be dropped again */
5635                 of_node_put(np);
5636         }
5637 #endif
5638
5639         kvm_ops_hv.owner = THIS_MODULE;
5640         kvmppc_hv_ops = &kvm_ops_hv;
5641
5642         init_default_hcalls();
5643
5644         init_vcore_lists();
5645
5646         r = kvmppc_mmu_hv_init();
5647         if (r)
5648                 return r;
5649
5650         if (kvmppc_radix_possible())
5651                 r = kvmppc_radix_init();
5652
5653         /*
5654          * POWER9 chips before version 2.02 can't have some threads in
5655          * HPT mode and some in radix mode on the same core.
5656          */
5657         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5658                 unsigned int pvr = mfspr(SPRN_PVR);
5659                 if ((pvr >> 16) == PVR_POWER9 &&
5660                     (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5661                      ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5662                         no_mixing_hpt_and_radix = true;
5663         }
5664
5665         r = kvmppc_uvmem_init();
5666         if (r < 0)
5667                 pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
5668
5669         return r;
5670 }
5671
5672 static void kvmppc_book3s_exit_hv(void)
5673 {
5674         kvmppc_uvmem_free();
5675         kvmppc_free_host_rm_ops();
5676         if (kvmppc_radix_possible())
5677                 kvmppc_radix_exit();
5678         kvmppc_hv_ops = NULL;
5679         kvmhv_nested_exit();
5680 }
5681
5682 module_init(kvmppc_book3s_init_hv);
5683 module_exit(kvmppc_book3s_exit_hv);
5684 MODULE_LICENSE("GPL");
5685 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5686 MODULE_ALIAS("devname:kvm");