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_host.h>
  76 #include <asm/kvm_book3s_uvmem.h>
  77 #include <asm/ultravisor.h>
  78
  79 #include "book3s.h"
  80
  81 #define CREATE_TRACE_POINTS
  82 #include "trace_hv.h"
  83
  84 /* #define EXIT_DEBUG */
  85 /* #define EXIT_DEBUG_SIMPLE */
  86 /* #define EXIT_DEBUG_INT */
  87
  88 /* Used to indicate that a guest page fault needs to be handled */
  89 #define RESUME_PAGE_FAULT       (RESUME_GUEST | RESUME_FLAG_ARCH1)
  90 /* Used to indicate that a guest passthrough interrupt needs to be handled */
  91 #define RESUME_PASSTHROUGH      (RESUME_GUEST | RESUME_FLAG_ARCH2)
  92
  93 /* Used as a "null" value for timebase values */
  94 #define TB_NIL  (~(u64)0)
  95
  96 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1);
  97
  98 static int dynamic_mt_modes = 6;
  99 module_param(dynamic_mt_modes, int, 0644);
 100 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)");
 101 static int target_smt_mode;
 102 module_param(target_smt_mode, int, 0644);
 103 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)");
 104
 105 static bool indep_threads_mode = true;
 106 module_param(indep_threads_mode, bool, S_IRUGO | S_IWUSR);
 107 MODULE_PARM_DESC(indep_threads_mode, "Independent-threads mode (only on POWER9)");
 108
 109 static bool one_vm_per_core;
 110 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR);
 111 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires indep_threads_mode=N)");
 112
 113 #ifdef CONFIG_KVM_XICS
 114 static struct kernel_param_ops module_param_ops = {
 115         .set = param_set_int,
 116         .get = param_get_int,
 117 };
 118
 119 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644);
 120 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization");
 121
 122 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644);
 123 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core");
 124 #endif
 125
 126 /* If set, guests are allowed to create and control nested guests */
 127 static bool nested = true;
 128 module_param(nested, bool, S_IRUGO | S_IWUSR);
 129 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)");
 130
 131 static inline bool nesting_enabled(struct kvm *kvm)
 132 {
 133         return kvm->arch.nested_enable && kvm_is_radix(kvm);
 134 }
 135
 136 /* If set, the threads on each CPU core have to be in the same MMU mode */
 137 static bool no_mixing_hpt_and_radix;
 138
 139 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu);
 140
 141 /*
 142  * RWMR values for POWER8.  These control the rate at which PURR
 143  * and SPURR count and should be set according to the number of
 144  * online threads in the vcore being run.
 145  */
 146 #define RWMR_RPA_P8_1THREAD     0x164520C62609AECAUL
 147 #define RWMR_RPA_P8_2THREAD     0x7FFF2908450D8DA9UL
 148 #define RWMR_RPA_P8_3THREAD     0x164520C62609AECAUL
 149 #define RWMR_RPA_P8_4THREAD     0x199A421245058DA9UL
 150 #define RWMR_RPA_P8_5THREAD     0x164520C62609AECAUL
 151 #define RWMR_RPA_P8_6THREAD     0x164520C62609AECAUL
 152 #define RWMR_RPA_P8_7THREAD     0x164520C62609AECAUL
 153 #define RWMR_RPA_P8_8THREAD     0x164520C62609AECAUL
 154
 155 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = {
 156         RWMR_RPA_P8_1THREAD,
 157         RWMR_RPA_P8_1THREAD,
 158         RWMR_RPA_P8_2THREAD,
 159         RWMR_RPA_P8_3THREAD,
 160         RWMR_RPA_P8_4THREAD,
 161         RWMR_RPA_P8_5THREAD,
 162         RWMR_RPA_P8_6THREAD,
 163         RWMR_RPA_P8_7THREAD,
 164         RWMR_RPA_P8_8THREAD,
 165 };
 166
 167 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc,
 168                 int *ip)
 169 {
 170         int i = *ip;
 171         struct kvm_vcpu *vcpu;
 172
 173         while (++i < MAX_SMT_THREADS) {
 174                 vcpu = READ_ONCE(vc->runnable_threads[i]);
 175                 if (vcpu) {
 176                         *ip = i;
 177                         return vcpu;
 178                 }
 179         }
 180         return NULL;
 181 }
 182
 183 /* Used to traverse the list of runnable threads for a given vcore */
 184 #define for_each_runnable_thread(i, vcpu, vc) \
 185         for (i = -1; (vcpu = next_runnable_thread(vc, &i)); )
 186
 187 static bool kvmppc_ipi_thread(int cpu)
 188 {
 189         unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
 190
 191         /* If we're a nested hypervisor, fall back to ordinary IPIs for now */
 192         if (kvmhv_on_pseries())
 193                 return false;
 194
 195         /* On POWER9 we can use msgsnd to IPI any cpu */
 196         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
 197                 msg |= get_hard_smp_processor_id(cpu);
 198                 smp_mb();
 199                 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 200                 return true;
 201         }
 202
 203         /* On POWER8 for IPIs to threads in the same core, use msgsnd */
 204         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
 205                 preempt_disable();
 206                 if (cpu_first_thread_sibling(cpu) ==
 207                     cpu_first_thread_sibling(smp_processor_id())) {
 208                         msg |= cpu_thread_in_core(cpu);
 209                         smp_mb();
 210                         __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg));
 211                         preempt_enable();
 212                         return true;
 213                 }
 214                 preempt_enable();
 215         }
 216
 217 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP)
 218         if (cpu >= 0 && cpu < nr_cpu_ids) {
 219                 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) {
 220                         xics_wake_cpu(cpu);
 221                         return true;
 222                 }
 223                 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY);
 224                 return true;
 225         }
 226 #endif
 227
 228         return false;
 229 }
 230
 231 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu)
 232 {
 233         int cpu;
 234         struct swait_queue_head *wqp;
 235
 236         wqp = kvm_arch_vcpu_wq(vcpu);
 237         if (swq_has_sleeper(wqp)) {
 238                 swake_up_one(wqp);
 239                 ++vcpu->stat.halt_wakeup;
 240         }
 241
 242         cpu = READ_ONCE(vcpu->arch.thread_cpu);
 243         if (cpu >= 0 && kvmppc_ipi_thread(cpu))
 244                 return;
 245
 246         /* CPU points to the first thread of the core */
 247         cpu = vcpu->cpu;
 248         if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu))
 249                 smp_send_reschedule(cpu);
 250 }
 251
 252 /*
 253  * We use the vcpu_load/put functions to measure stolen time.
 254  * Stolen time is counted as time when either the vcpu is able to
 255  * run as part of a virtual core, but the task running the vcore
 256  * is preempted or sleeping, or when the vcpu needs something done
 257  * in the kernel by the task running the vcpu, but that task is
 258  * preempted or sleeping.  Those two things have to be counted
 259  * separately, since one of the vcpu tasks will take on the job
 260  * of running the core, and the other vcpu tasks in the vcore will
 261  * sleep waiting for it to do that, but that sleep shouldn't count
 262  * as stolen time.
 263  *
 264  * Hence we accumulate stolen time when the vcpu can run as part of
 265  * a vcore using vc->stolen_tb, and the stolen time when the vcpu
 266  * needs its task to do other things in the kernel (for example,
 267  * service a page fault) in busy_stolen.  We don't accumulate
 268  * stolen time for a vcore when it is inactive, or for a vcpu
 269  * when it is in state RUNNING or NOTREADY.  NOTREADY is a bit of
 270  * a misnomer; it means that the vcpu task is not executing in
 271  * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in
 272  * the kernel.  We don't have any way of dividing up that time
 273  * between time that the vcpu is genuinely stopped, time that
 274  * the task is actively working on behalf of the vcpu, and time
 275  * that the task is preempted, so we don't count any of it as
 276  * stolen.
 277  *
 278  * Updates to busy_stolen are protected by arch.tbacct_lock;
 279  * updates to vc->stolen_tb are protected by the vcore->stoltb_lock
 280  * lock.  The stolen times are measured in units of timebase ticks.
 281  * (Note that the != TB_NIL checks below are purely defensive;
 282  * they should never fail.)
 283  */
 284
 285 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc)
 286 {
 287         unsigned long flags;
 288
 289         spin_lock_irqsave(&vc->stoltb_lock, flags);
 290         vc->preempt_tb = mftb();
 291         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 292 }
 293
 294 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc)
 295 {
 296         unsigned long flags;
 297
 298         spin_lock_irqsave(&vc->stoltb_lock, flags);
 299         if (vc->preempt_tb != TB_NIL) {
 300                 vc->stolen_tb += mftb() - vc->preempt_tb;
 301                 vc->preempt_tb = TB_NIL;
 302         }
 303         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 304 }
 305
 306 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu)
 307 {
 308         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 309         unsigned long flags;
 310
 311         /*
 312          * We can test vc->runner without taking the vcore lock,
 313          * because only this task ever sets vc->runner to this
 314          * vcpu, and once it is set to this vcpu, only this task
 315          * ever sets it to NULL.
 316          */
 317         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 318                 kvmppc_core_end_stolen(vc);
 319
 320         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 321         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST &&
 322             vcpu->arch.busy_preempt != TB_NIL) {
 323                 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt;
 324                 vcpu->arch.busy_preempt = TB_NIL;
 325         }
 326         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 327 }
 328
 329 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu)
 330 {
 331         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 332         unsigned long flags;
 333
 334         if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING)
 335                 kvmppc_core_start_stolen(vc);
 336
 337         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 338         if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST)
 339                 vcpu->arch.busy_preempt = mftb();
 340         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 341 }
 342
 343 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr)
 344 {
 345         vcpu->arch.pvr = pvr;
 346 }
 347
 348 /* Dummy value used in computing PCR value below */
 349 #define PCR_ARCH_300    (PCR_ARCH_207 << 1)
 350
 351 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat)
 352 {
 353         unsigned long host_pcr_bit = 0, guest_pcr_bit = 0;
 354         struct kvmppc_vcore *vc = vcpu->arch.vcore;
 355
 356         /* We can (emulate) our own architecture version and anything older */
 357         if (cpu_has_feature(CPU_FTR_ARCH_300))
 358                 host_pcr_bit = PCR_ARCH_300;
 359         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
 360                 host_pcr_bit = PCR_ARCH_207;
 361         else if (cpu_has_feature(CPU_FTR_ARCH_206))
 362                 host_pcr_bit = PCR_ARCH_206;
 363         else
 364                 host_pcr_bit = PCR_ARCH_205;
 365
 366         /* Determine lowest PCR bit needed to run guest in given PVR level */
 367         guest_pcr_bit = host_pcr_bit;
 368         if (arch_compat) {
 369                 switch (arch_compat) {
 370                 case PVR_ARCH_205:
 371                         guest_pcr_bit = PCR_ARCH_205;
 372                         break;
 373                 case PVR_ARCH_206:
 374                 case PVR_ARCH_206p:
 375                         guest_pcr_bit = PCR_ARCH_206;
 376                         break;
 377                 case PVR_ARCH_207:
 378                         guest_pcr_bit = PCR_ARCH_207;
 379                         break;
 380                 case PVR_ARCH_300:
 381                         guest_pcr_bit = PCR_ARCH_300;
 382                         break;
 383                 default:
 384                         return -EINVAL;
 385                 }
 386         }
 387
 388         /* Check requested PCR bits don't exceed our capabilities */
 389         if (guest_pcr_bit > host_pcr_bit)
 390                 return -EINVAL;
 391
 392         spin_lock(&vc->lock);
 393         vc->arch_compat = arch_compat;
 394         /*
 395          * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit
 396          * Also set all reserved PCR bits
 397          */
 398         vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK;
 399         spin_unlock(&vc->lock);
 400
 401         return 0;
 402 }
 403
 404 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu)
 405 {
 406         int r;
 407
 408         pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id);
 409         pr_err("pc  = %.16lx  msr = %.16llx  trap = %x\n",
 410                vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap);
 411         for (r = 0; r < 16; ++r)
 412                 pr_err("r%2d = %.16lx  r%d = %.16lx\n",
 413                        r, kvmppc_get_gpr(vcpu, r),
 414                        r+16, kvmppc_get_gpr(vcpu, r+16));
 415         pr_err("ctr = %.16lx  lr  = %.16lx\n",
 416                vcpu->arch.regs.ctr, vcpu->arch.regs.link);
 417         pr_err("srr0 = %.16llx srr1 = %.16llx\n",
 418                vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1);
 419         pr_err("sprg0 = %.16llx sprg1 = %.16llx\n",
 420                vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1);
 421         pr_err("sprg2 = %.16llx sprg3 = %.16llx\n",
 422                vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3);
 423         pr_err("cr = %.8lx  xer = %.16lx  dsisr = %.8x\n",
 424                vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr);
 425         pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar);
 426         pr_err("fault dar = %.16lx dsisr = %.8x\n",
 427                vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
 428         pr_err("SLB (%d entries):\n", vcpu->arch.slb_max);
 429         for (r = 0; r < vcpu->arch.slb_max; ++r)
 430                 pr_err("  ESID = %.16llx VSID = %.16llx\n",
 431                        vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv);
 432         pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n",
 433                vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1,
 434                vcpu->arch.last_inst);
 435 }
 436
 437 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id)
 438 {
 439         return kvm_get_vcpu_by_id(kvm, id);
 440 }
 441
 442 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa)
 443 {
 444         vpa->__old_status |= LPPACA_OLD_SHARED_PROC;
 445         vpa->yield_count = cpu_to_be32(1);
 446 }
 447
 448 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v,
 449                    unsigned long addr, unsigned long len)
 450 {
 451         /* check address is cacheline aligned */
 452         if (addr & (L1_CACHE_BYTES - 1))
 453                 return -EINVAL;
 454         spin_lock(&vcpu->arch.vpa_update_lock);
 455         if (v->next_gpa != addr || v->len != len) {
 456                 v->next_gpa = addr;
 457                 v->len = addr ? len : 0;
 458                 v->update_pending = 1;
 459         }
 460         spin_unlock(&vcpu->arch.vpa_update_lock);
 461         return 0;
 462 }
 463
 464 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */
 465 struct reg_vpa {
 466         u32 dummy;
 467         union {
 468                 __be16 hword;
 469                 __be32 word;
 470         } length;
 471 };
 472
 473 static int vpa_is_registered(struct kvmppc_vpa *vpap)
 474 {
 475         if (vpap->update_pending)
 476                 return vpap->next_gpa != 0;
 477         return vpap->pinned_addr != NULL;
 478 }
 479
 480 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu,
 481                                        unsigned long flags,
 482                                        unsigned long vcpuid, unsigned long vpa)
 483 {
 484         struct kvm *kvm = vcpu->kvm;
 485         unsigned long len, nb;
 486         void *va;
 487         struct kvm_vcpu *tvcpu;
 488         int err;
 489         int subfunc;
 490         struct kvmppc_vpa *vpap;
 491
 492         tvcpu = kvmppc_find_vcpu(kvm, vcpuid);
 493         if (!tvcpu)
 494                 return H_PARAMETER;
 495
 496         subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK;
 497         if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL ||
 498             subfunc == H_VPA_REG_SLB) {
 499                 /* Registering new area - address must be cache-line aligned */
 500                 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa)
 501                         return H_PARAMETER;
 502
 503                 /* convert logical addr to kernel addr and read length */
 504                 va = kvmppc_pin_guest_page(kvm, vpa, &nb);
 505                 if (va == NULL)
 506                         return H_PARAMETER;
 507                 if (subfunc == H_VPA_REG_VPA)
 508                         len = be16_to_cpu(((struct reg_vpa *)va)->length.hword);
 509                 else
 510                         len = be32_to_cpu(((struct reg_vpa *)va)->length.word);
 511                 kvmppc_unpin_guest_page(kvm, va, vpa, false);
 512
 513                 /* Check length */
 514                 if (len > nb || len < sizeof(struct reg_vpa))
 515                         return H_PARAMETER;
 516         } else {
 517                 vpa = 0;
 518                 len = 0;
 519         }
 520
 521         err = H_PARAMETER;
 522         vpap = NULL;
 523         spin_lock(&tvcpu->arch.vpa_update_lock);
 524
 525         switch (subfunc) {
 526         case H_VPA_REG_VPA:             /* register VPA */
 527                 /*
 528                  * The size of our lppaca is 1kB because of the way we align
 529                  * it for the guest to avoid crossing a 4kB boundary. We only
 530                  * use 640 bytes of the structure though, so we should accept
 531                  * clients that set a size of 640.
 532                  */
 533                 BUILD_BUG_ON(sizeof(struct lppaca) != 640);
 534                 if (len < sizeof(struct lppaca))
 535                         break;
 536                 vpap = &tvcpu->arch.vpa;
 537                 err = 0;
 538                 break;
 539
 540         case H_VPA_REG_DTL:             /* register DTL */
 541                 if (len < sizeof(struct dtl_entry))
 542                         break;
 543                 len -= len % sizeof(struct dtl_entry);
 544
 545                 /* Check that they have previously registered a VPA */
 546                 err = H_RESOURCE;
 547                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 548                         break;
 549
 550                 vpap = &tvcpu->arch.dtl;
 551                 err = 0;
 552                 break;
 553
 554         case H_VPA_REG_SLB:             /* register SLB shadow buffer */
 555                 /* Check that they have previously registered a VPA */
 556                 err = H_RESOURCE;
 557                 if (!vpa_is_registered(&tvcpu->arch.vpa))
 558                         break;
 559
 560                 vpap = &tvcpu->arch.slb_shadow;
 561                 err = 0;
 562                 break;
 563
 564         case H_VPA_DEREG_VPA:           /* deregister VPA */
 565                 /* Check they don't still have a DTL or SLB buf registered */
 566                 err = H_RESOURCE;
 567                 if (vpa_is_registered(&tvcpu->arch.dtl) ||
 568                     vpa_is_registered(&tvcpu->arch.slb_shadow))
 569                         break;
 570
 571                 vpap = &tvcpu->arch.vpa;
 572                 err = 0;
 573                 break;
 574
 575         case H_VPA_DEREG_DTL:           /* deregister DTL */
 576                 vpap = &tvcpu->arch.dtl;
 577                 err = 0;
 578                 break;
 579
 580         case H_VPA_DEREG_SLB:           /* deregister SLB shadow buffer */
 581                 vpap = &tvcpu->arch.slb_shadow;
 582                 err = 0;
 583                 break;
 584         }
 585
 586         if (vpap) {
 587                 vpap->next_gpa = vpa;
 588                 vpap->len = len;
 589                 vpap->update_pending = 1;
 590         }
 591
 592         spin_unlock(&tvcpu->arch.vpa_update_lock);
 593
 594         return err;
 595 }
 596
 597 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap)
 598 {
 599         struct kvm *kvm = vcpu->kvm;
 600         void *va;
 601         unsigned long nb;
 602         unsigned long gpa;
 603
 604         /*
 605          * We need to pin the page pointed to by vpap->next_gpa,
 606          * but we can't call kvmppc_pin_guest_page under the lock
 607          * as it does get_user_pages() and down_read().  So we
 608          * have to drop the lock, pin the page, then get the lock
 609          * again and check that a new area didn't get registered
 610          * in the meantime.
 611          */
 612         for (;;) {
 613                 gpa = vpap->next_gpa;
 614                 spin_unlock(&vcpu->arch.vpa_update_lock);
 615                 va = NULL;
 616                 nb = 0;
 617                 if (gpa)
 618                         va = kvmppc_pin_guest_page(kvm, gpa, &nb);
 619                 spin_lock(&vcpu->arch.vpa_update_lock);
 620                 if (gpa == vpap->next_gpa)
 621                         break;
 622                 /* sigh... unpin that one and try again */
 623                 if (va)
 624                         kvmppc_unpin_guest_page(kvm, va, gpa, false);
 625         }
 626
 627         vpap->update_pending = 0;
 628         if (va && nb < vpap->len) {
 629                 /*
 630                  * If it's now too short, it must be that userspace
 631                  * has changed the mappings underlying guest memory,
 632                  * so unregister the region.
 633                  */
 634                 kvmppc_unpin_guest_page(kvm, va, gpa, false);
 635                 va = NULL;
 636         }
 637         if (vpap->pinned_addr)
 638                 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa,
 639                                         vpap->dirty);
 640         vpap->gpa = gpa;
 641         vpap->pinned_addr = va;
 642         vpap->dirty = false;
 643         if (va)
 644                 vpap->pinned_end = va + vpap->len;
 645 }
 646
 647 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu)
 648 {
 649         if (!(vcpu->arch.vpa.update_pending ||
 650               vcpu->arch.slb_shadow.update_pending ||
 651               vcpu->arch.dtl.update_pending))
 652                 return;
 653
 654         spin_lock(&vcpu->arch.vpa_update_lock);
 655         if (vcpu->arch.vpa.update_pending) {
 656                 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa);
 657                 if (vcpu->arch.vpa.pinned_addr)
 658                         init_vpa(vcpu, vcpu->arch.vpa.pinned_addr);
 659         }
 660         if (vcpu->arch.dtl.update_pending) {
 661                 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl);
 662                 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr;
 663                 vcpu->arch.dtl_index = 0;
 664         }
 665         if (vcpu->arch.slb_shadow.update_pending)
 666                 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow);
 667         spin_unlock(&vcpu->arch.vpa_update_lock);
 668 }
 669
 670 /*
 671  * Return the accumulated stolen time for the vcore up until `now'.
 672  * The caller should hold the vcore lock.
 673  */
 674 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now)
 675 {
 676         u64 p;
 677         unsigned long flags;
 678
 679         spin_lock_irqsave(&vc->stoltb_lock, flags);
 680         p = vc->stolen_tb;
 681         if (vc->vcore_state != VCORE_INACTIVE &&
 682             vc->preempt_tb != TB_NIL)
 683                 p += now - vc->preempt_tb;
 684         spin_unlock_irqrestore(&vc->stoltb_lock, flags);
 685         return p;
 686 }
 687
 688 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu,
 689                                     struct kvmppc_vcore *vc)
 690 {
 691         struct dtl_entry *dt;
 692         struct lppaca *vpa;
 693         unsigned long stolen;
 694         unsigned long core_stolen;
 695         u64 now;
 696         unsigned long flags;
 697
 698         dt = vcpu->arch.dtl_ptr;
 699         vpa = vcpu->arch.vpa.pinned_addr;
 700         now = mftb();
 701         core_stolen = vcore_stolen_time(vc, now);
 702         stolen = core_stolen - vcpu->arch.stolen_logged;
 703         vcpu->arch.stolen_logged = core_stolen;
 704         spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags);
 705         stolen += vcpu->arch.busy_stolen;
 706         vcpu->arch.busy_stolen = 0;
 707         spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags);
 708         if (!dt || !vpa)
 709                 return;
 710         memset(dt, 0, sizeof(struct dtl_entry));
 711         dt->dispatch_reason = 7;
 712         dt->processor_id = cpu_to_be16(vc->pcpu + vcpu->arch.ptid);
 713         dt->timebase = cpu_to_be64(now + vc->tb_offset);
 714         dt->enqueue_to_dispatch_time = cpu_to_be32(stolen);
 715         dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu));
 716         dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr);
 717         ++dt;
 718         if (dt == vcpu->arch.dtl.pinned_end)
 719                 dt = vcpu->arch.dtl.pinned_addr;
 720         vcpu->arch.dtl_ptr = dt;
 721         /* order writing *dt vs. writing vpa->dtl_idx */
 722         smp_wmb();
 723         vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index);
 724         vcpu->arch.dtl.dirty = true;
 725 }
 726
 727 /* See if there is a doorbell interrupt pending for a vcpu */
 728 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
 729 {
 730         int thr;
 731         struct kvmppc_vcore *vc;
 732
 733         if (vcpu->arch.doorbell_request)
 734                 return true;
 735         /*
 736          * Ensure that the read of vcore->dpdes comes after the read
 737          * of vcpu->doorbell_request.  This barrier matches the
 738          * smp_wmb() in kvmppc_guest_entry_inject().
 739          */
 740         smp_rmb();
 741         vc = vcpu->arch.vcore;
 742         thr = vcpu->vcpu_id - vc->first_vcpuid;
 743         return !!(vc->dpdes & (1 << thr));
 744 }
 745
 746 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu)
 747 {
 748         if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207)
 749                 return true;
 750         if ((!vcpu->arch.vcore->arch_compat) &&
 751             cpu_has_feature(CPU_FTR_ARCH_207S))
 752                 return true;
 753         return false;
 754 }
 755
 756 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags,
 757                              unsigned long resource, unsigned long value1,
 758                              unsigned long value2)
 759 {
 760         switch (resource) {
 761         case H_SET_MODE_RESOURCE_SET_CIABR:
 762                 if (!kvmppc_power8_compatible(vcpu))
 763                         return H_P2;
 764                 if (value2)
 765                         return H_P4;
 766                 if (mflags)
 767                         return H_UNSUPPORTED_FLAG_START;
 768                 /* Guests can't breakpoint the hypervisor */
 769                 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER)
 770                         return H_P3;
 771                 vcpu->arch.ciabr  = value1;
 772                 return H_SUCCESS;
 773         case H_SET_MODE_RESOURCE_SET_DAWR:
 774                 if (!kvmppc_power8_compatible(vcpu))
 775                         return H_P2;
 776                 if (!ppc_breakpoint_available())
 777                         return H_P2;
 778                 if (mflags)
 779                         return H_UNSUPPORTED_FLAG_START;
 780                 if (value2 & DABRX_HYP)
 781                         return H_P4;
 782                 vcpu->arch.dawr  = value1;
 783                 vcpu->arch.dawrx = value2;
 784                 return H_SUCCESS;
 785         case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE:
 786                 /* KVM does not support mflags=2 (AIL=2) */
 787                 if (mflags != 0 && mflags != 3)
 788                         return H_UNSUPPORTED_FLAG_START;
 789                 return H_TOO_HARD;
 790         default:
 791                 return H_TOO_HARD;
 792         }
 793 }
 794
 795 /* Copy guest memory in place - must reside within a single memslot */
 796 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from,
 797                                   unsigned long len)
 798 {
 799         struct kvm_memory_slot *to_memslot = NULL;
 800         struct kvm_memory_slot *from_memslot = NULL;
 801         unsigned long to_addr, from_addr;
 802         int r;
 803
 804         /* Get HPA for from address */
 805         from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT);
 806         if (!from_memslot)
 807                 return -EFAULT;
 808         if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages)
 809                              << PAGE_SHIFT))
 810                 return -EINVAL;
 811         from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT);
 812         if (kvm_is_error_hva(from_addr))
 813                 return -EFAULT;
 814         from_addr |= (from & (PAGE_SIZE - 1));
 815
 816         /* Get HPA for to address */
 817         to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT);
 818         if (!to_memslot)
 819                 return -EFAULT;
 820         if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages)
 821                            << PAGE_SHIFT))
 822                 return -EINVAL;
 823         to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT);
 824         if (kvm_is_error_hva(to_addr))
 825                 return -EFAULT;
 826         to_addr |= (to & (PAGE_SIZE - 1));
 827
 828         /* Perform copy */
 829         r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr,
 830                              len);
 831         if (r)
 832                 return -EFAULT;
 833         mark_page_dirty(kvm, to >> PAGE_SHIFT);
 834         return 0;
 835 }
 836
 837 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags,
 838                                unsigned long dest, unsigned long src)
 839 {
 840         u64 pg_sz = SZ_4K;              /* 4K page size */
 841         u64 pg_mask = SZ_4K - 1;
 842         int ret;
 843
 844         /* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */
 845         if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE |
 846                       H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED))
 847                 return H_PARAMETER;
 848
 849         /* dest (and src if copy_page flag set) must be page aligned */
 850         if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask)))
 851                 return H_PARAMETER;
 852
 853         /* zero and/or copy the page as determined by the flags */
 854         if (flags & H_COPY_PAGE) {
 855                 ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz);
 856                 if (ret < 0)
 857                         return H_PARAMETER;
 858         } else if (flags & H_ZERO_PAGE) {
 859                 ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz);
 860                 if (ret < 0)
 861                         return H_PARAMETER;
 862         }
 863
 864         /* We can ignore the remaining flags */
 865
 866         return H_SUCCESS;
 867 }
 868
 869 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target)
 870 {
 871         struct kvmppc_vcore *vcore = target->arch.vcore;
 872
 873         /*
 874          * We expect to have been called by the real mode handler
 875          * (kvmppc_rm_h_confer()) which would have directly returned
 876          * H_SUCCESS if the source vcore wasn't idle (e.g. if it may
 877          * have useful work to do and should not confer) so we don't
 878          * recheck that here.
 879          */
 880
 881         spin_lock(&vcore->lock);
 882         if (target->arch.state == KVMPPC_VCPU_RUNNABLE &&
 883             vcore->vcore_state != VCORE_INACTIVE &&
 884             vcore->runner)
 885                 target = vcore->runner;
 886         spin_unlock(&vcore->lock);
 887
 888         return kvm_vcpu_yield_to(target);
 889 }
 890
 891 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu)
 892 {
 893         int yield_count = 0;
 894         struct lppaca *lppaca;
 895
 896         spin_lock(&vcpu->arch.vpa_update_lock);
 897         lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr;
 898         if (lppaca)
 899                 yield_count = be32_to_cpu(lppaca->yield_count);
 900         spin_unlock(&vcpu->arch.vpa_update_lock);
 901         return yield_count;
 902 }
 903
 904 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu)
 905 {
 906         unsigned long req = kvmppc_get_gpr(vcpu, 3);
 907         unsigned long target, ret = H_SUCCESS;
 908         int yield_count;
 909         struct kvm_vcpu *tvcpu;
 910         int idx, rc;
 911
 912         if (req <= MAX_HCALL_OPCODE &&
 913             !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls))
 914                 return RESUME_HOST;
 915
 916         switch (req) {
 917         case H_CEDE:
 918                 break;
 919         case H_PROD:
 920                 target = kvmppc_get_gpr(vcpu, 4);
 921                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 922                 if (!tvcpu) {
 923                         ret = H_PARAMETER;
 924                         break;
 925                 }
 926                 tvcpu->arch.prodded = 1;
 927                 smp_mb();
 928                 if (tvcpu->arch.ceded)
 929                         kvmppc_fast_vcpu_kick_hv(tvcpu);
 930                 break;
 931         case H_CONFER:
 932                 target = kvmppc_get_gpr(vcpu, 4);
 933                 if (target == -1)
 934                         break;
 935                 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target);
 936                 if (!tvcpu) {
 937                         ret = H_PARAMETER;
 938                         break;
 939                 }
 940                 yield_count = kvmppc_get_gpr(vcpu, 5);
 941                 if (kvmppc_get_yield_count(tvcpu) != yield_count)
 942                         break;
 943                 kvm_arch_vcpu_yield_to(tvcpu);
 944                 break;
 945         case H_REGISTER_VPA:
 946                 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4),
 947                                         kvmppc_get_gpr(vcpu, 5),
 948                                         kvmppc_get_gpr(vcpu, 6));
 949                 break;
 950         case H_RTAS:
 951                 if (list_empty(&vcpu->kvm->arch.rtas_tokens))
 952                         return RESUME_HOST;
 953
 954                 idx = srcu_read_lock(&vcpu->kvm->srcu);
 955                 rc = kvmppc_rtas_hcall(vcpu);
 956                 srcu_read_unlock(&vcpu->kvm->srcu, idx);
 957
 958                 if (rc == -ENOENT)
 959                         return RESUME_HOST;
 960                 else if (rc == 0)
 961                         break;
 962
 963                 /* Send the error out to userspace via KVM_RUN */
 964                 return rc;
 965         case H_LOGICAL_CI_LOAD:
 966                 ret = kvmppc_h_logical_ci_load(vcpu);
 967                 if (ret == H_TOO_HARD)
 968                         return RESUME_HOST;
 969                 break;
 970         case H_LOGICAL_CI_STORE:
 971                 ret = kvmppc_h_logical_ci_store(vcpu);
 972                 if (ret == H_TOO_HARD)
 973                         return RESUME_HOST;
 974                 break;
 975         case H_SET_MODE:
 976                 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4),
 977                                         kvmppc_get_gpr(vcpu, 5),
 978                                         kvmppc_get_gpr(vcpu, 6),
 979                                         kvmppc_get_gpr(vcpu, 7));
 980                 if (ret == H_TOO_HARD)
 981                         return RESUME_HOST;
 982                 break;
 983         case H_XIRR:
 984         case H_CPPR:
 985         case H_EOI:
 986         case H_IPI:
 987         case H_IPOLL:
 988         case H_XIRR_X:
 989                 if (kvmppc_xics_enabled(vcpu)) {
 990                         if (xics_on_xive()) {
 991                                 ret = H_NOT_AVAILABLE;
 992                                 return RESUME_GUEST;
 993                         }
 994                         ret = kvmppc_xics_hcall(vcpu, req);
 995                         break;
 996                 }
 997                 return RESUME_HOST;
 998         case H_SET_DABR:
 999                 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4));
1000                 break;
1001         case H_SET_XDABR:
1002                 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4),
1003                                                 kvmppc_get_gpr(vcpu, 5));
1004                 break;
1005 #ifdef CONFIG_SPAPR_TCE_IOMMU
1006         case H_GET_TCE:
1007                 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1008                                                 kvmppc_get_gpr(vcpu, 5));
1009                 if (ret == H_TOO_HARD)
1010                         return RESUME_HOST;
1011                 break;
1012         case H_PUT_TCE:
1013                 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1014                                                 kvmppc_get_gpr(vcpu, 5),
1015                                                 kvmppc_get_gpr(vcpu, 6));
1016                 if (ret == H_TOO_HARD)
1017                         return RESUME_HOST;
1018                 break;
1019         case H_PUT_TCE_INDIRECT:
1020                 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4),
1021                                                 kvmppc_get_gpr(vcpu, 5),
1022                                                 kvmppc_get_gpr(vcpu, 6),
1023                                                 kvmppc_get_gpr(vcpu, 7));
1024                 if (ret == H_TOO_HARD)
1025                         return RESUME_HOST;
1026                 break;
1027         case H_STUFF_TCE:
1028                 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4),
1029                                                 kvmppc_get_gpr(vcpu, 5),
1030                                                 kvmppc_get_gpr(vcpu, 6),
1031                                                 kvmppc_get_gpr(vcpu, 7));
1032                 if (ret == H_TOO_HARD)
1033                         return RESUME_HOST;
1034                 break;
1035 #endif
1036         case H_RANDOM:
1037                 if (!powernv_get_random_long(&vcpu->arch.regs.gpr[4]))
1038                         ret = H_HARDWARE;
1039                 break;
1040
1041         case H_SET_PARTITION_TABLE:
1042                 ret = H_FUNCTION;
1043                 if (nesting_enabled(vcpu->kvm))
1044                         ret = kvmhv_set_partition_table(vcpu);
1045                 break;
1046         case H_ENTER_NESTED:
1047                 ret = H_FUNCTION;
1048                 if (!nesting_enabled(vcpu->kvm))
1049                         break;
1050                 ret = kvmhv_enter_nested_guest(vcpu);
1051                 if (ret == H_INTERRUPT) {
1052                         kvmppc_set_gpr(vcpu, 3, 0);
1053                         vcpu->arch.hcall_needed = 0;
1054                         return -EINTR;
1055                 } else if (ret == H_TOO_HARD) {
1056                         kvmppc_set_gpr(vcpu, 3, 0);
1057                         vcpu->arch.hcall_needed = 0;
1058                         return RESUME_HOST;
1059                 }
1060                 break;
1061         case H_TLB_INVALIDATE:
1062                 ret = H_FUNCTION;
1063                 if (nesting_enabled(vcpu->kvm))
1064                         ret = kvmhv_do_nested_tlbie(vcpu);
1065                 break;
1066         case H_COPY_TOFROM_GUEST:
1067                 ret = H_FUNCTION;
1068                 if (nesting_enabled(vcpu->kvm))
1069                         ret = kvmhv_copy_tofrom_guest_nested(vcpu);
1070                 break;
1071         case H_PAGE_INIT:
1072                 ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4),
1073                                          kvmppc_get_gpr(vcpu, 5),
1074                                          kvmppc_get_gpr(vcpu, 6));
1075                 break;
1076         case H_SVM_PAGE_IN:
1077                 ret = kvmppc_h_svm_page_in(vcpu->kvm,
1078                                            kvmppc_get_gpr(vcpu, 4),
1079                                            kvmppc_get_gpr(vcpu, 5),
1080                                            kvmppc_get_gpr(vcpu, 6));
1081                 break;
1082         case H_SVM_PAGE_OUT:
1083                 ret = kvmppc_h_svm_page_out(vcpu->kvm,
1084                                             kvmppc_get_gpr(vcpu, 4),
1085                                             kvmppc_get_gpr(vcpu, 5),
1086                                             kvmppc_get_gpr(vcpu, 6));
1087                 break;
1088         case H_SVM_INIT_START:
1089                 ret = kvmppc_h_svm_init_start(vcpu->kvm);
1090                 break;
1091         case H_SVM_INIT_DONE:
1092                 ret = kvmppc_h_svm_init_done(vcpu->kvm);
1093                 break;
1094         case H_SVM_INIT_ABORT:
1095                 ret = kvmppc_h_svm_init_abort(vcpu->kvm);
1096                 break;
1097
1098         default:
1099                 return RESUME_HOST;
1100         }
1101         kvmppc_set_gpr(vcpu, 3, ret);
1102         vcpu->arch.hcall_needed = 0;
1103         return RESUME_GUEST;
1104 }
1105
1106 /*
1107  * Handle H_CEDE in the nested virtualization case where we haven't
1108  * called the real-mode hcall handlers in book3s_hv_rmhandlers.S.
1109  * This has to be done early, not in kvmppc_pseries_do_hcall(), so
1110  * that the cede logic in kvmppc_run_single_vcpu() works properly.
1111  */
1112 static void kvmppc_nested_cede(struct kvm_vcpu *vcpu)
1113 {
1114         vcpu->arch.shregs.msr |= MSR_EE;
1115         vcpu->arch.ceded = 1;
1116         smp_mb();
1117         if (vcpu->arch.prodded) {
1118                 vcpu->arch.prodded = 0;
1119                 smp_mb();
1120                 vcpu->arch.ceded = 0;
1121         }
1122 }
1123
1124 static int kvmppc_hcall_impl_hv(unsigned long cmd)
1125 {
1126         switch (cmd) {
1127         case H_CEDE:
1128         case H_PROD:
1129         case H_CONFER:
1130         case H_REGISTER_VPA:
1131         case H_SET_MODE:
1132         case H_LOGICAL_CI_LOAD:
1133         case H_LOGICAL_CI_STORE:
1134 #ifdef CONFIG_KVM_XICS
1135         case H_XIRR:
1136         case H_CPPR:
1137         case H_EOI:
1138         case H_IPI:
1139         case H_IPOLL:
1140         case H_XIRR_X:
1141 #endif
1142         case H_PAGE_INIT:
1143                 return 1;
1144         }
1145
1146         /* See if it's in the real-mode table */
1147         return kvmppc_hcall_impl_hv_realmode(cmd);
1148 }
1149
1150 static int kvmppc_emulate_debug_inst(struct kvm_run *run,
1151                                         struct kvm_vcpu *vcpu)
1152 {
1153         u32 last_inst;
1154
1155         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
1156                                         EMULATE_DONE) {
1157                 /*
1158                  * Fetch failed, so return to guest and
1159                  * try executing it again.
1160                  */
1161                 return RESUME_GUEST;
1162         }
1163
1164         if (last_inst == KVMPPC_INST_SW_BREAKPOINT) {
1165                 run->exit_reason = KVM_EXIT_DEBUG;
1166                 run->debug.arch.address = kvmppc_get_pc(vcpu);
1167                 return RESUME_HOST;
1168         } else {
1169                 kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1170                 return RESUME_GUEST;
1171         }
1172 }
1173
1174 static void do_nothing(void *x)
1175 {
1176 }
1177
1178 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
1179 {
1180         int thr, cpu, pcpu, nthreads;
1181         struct kvm_vcpu *v;
1182         unsigned long dpdes;
1183
1184         nthreads = vcpu->kvm->arch.emul_smt_mode;
1185         dpdes = 0;
1186         cpu = vcpu->vcpu_id & ~(nthreads - 1);
1187         for (thr = 0; thr < nthreads; ++thr, ++cpu) {
1188                 v = kvmppc_find_vcpu(vcpu->kvm, cpu);
1189                 if (!v)
1190                         continue;
1191                 /*
1192                  * If the vcpu is currently running on a physical cpu thread,
1193                  * interrupt it in order to pull it out of the guest briefly,
1194                  * which will update its vcore->dpdes value.
1195                  */
1196                 pcpu = READ_ONCE(v->cpu);
1197                 if (pcpu >= 0)
1198                         smp_call_function_single(pcpu, do_nothing, NULL, 1);
1199                 if (kvmppc_doorbell_pending(v))
1200                         dpdes |= 1 << thr;
1201         }
1202         return dpdes;
1203 }
1204
1205 /*
1206  * On POWER9, emulate doorbell-related instructions in order to
1207  * give the guest the illusion of running on a multi-threaded core.
1208  * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
1209  * and mfspr DPDES.
1210  */
1211 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
1212 {
1213         u32 inst, rb, thr;
1214         unsigned long arg;
1215         struct kvm *kvm = vcpu->kvm;
1216         struct kvm_vcpu *tvcpu;
1217
1218         if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
1219                 return RESUME_GUEST;
1220         if (get_op(inst) != 31)
1221                 return EMULATE_FAIL;
1222         rb = get_rb(inst);
1223         thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
1224         switch (get_xop(inst)) {
1225         case OP_31_XOP_MSGSNDP:
1226                 arg = kvmppc_get_gpr(vcpu, rb);
1227                 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1228                         break;
1229                 arg &= 0x3f;
1230                 if (arg >= kvm->arch.emul_smt_mode)
1231                         break;
1232                 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
1233                 if (!tvcpu)
1234                         break;
1235                 if (!tvcpu->arch.doorbell_request) {
1236                         tvcpu->arch.doorbell_request = 1;
1237                         kvmppc_fast_vcpu_kick_hv(tvcpu);
1238                 }
1239                 break;
1240         case OP_31_XOP_MSGCLRP:
1241                 arg = kvmppc_get_gpr(vcpu, rb);
1242                 if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
1243                         break;
1244                 vcpu->arch.vcore->dpdes = 0;
1245                 vcpu->arch.doorbell_request = 0;
1246                 break;
1247         case OP_31_XOP_MFSPR:
1248                 switch (get_sprn(inst)) {
1249                 case SPRN_TIR:
1250                         arg = thr;
1251                         break;
1252                 case SPRN_DPDES:
1253                         arg = kvmppc_read_dpdes(vcpu);
1254                         break;
1255                 default:
1256                         return EMULATE_FAIL;
1257                 }
1258                 kvmppc_set_gpr(vcpu, get_rt(inst), arg);
1259                 break;
1260         default:
1261                 return EMULATE_FAIL;
1262         }
1263         kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
1264         return RESUME_GUEST;
1265 }
1266
1267 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
1268                                  struct task_struct *tsk)
1269 {
1270         int r = RESUME_HOST;
1271
1272         vcpu->stat.sum_exits++;
1273
1274         /*
1275          * This can happen if an interrupt occurs in the last stages
1276          * of guest entry or the first stages of guest exit (i.e. after
1277          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1278          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1279          * That can happen due to a bug, or due to a machine check
1280          * occurring at just the wrong time.
1281          */
1282         if (vcpu->arch.shregs.msr & MSR_HV) {
1283                 printk(KERN_EMERG "KVM trap in HV mode!\n");
1284                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1285                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1286                         vcpu->arch.shregs.msr);
1287                 kvmppc_dump_regs(vcpu);
1288                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1289                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1290                 return RESUME_HOST;
1291         }
1292         run->exit_reason = KVM_EXIT_UNKNOWN;
1293         run->ready_for_interrupt_injection = 1;
1294         switch (vcpu->arch.trap) {
1295         /* We're good on these - the host merely wanted to get our attention */
1296         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1297                 vcpu->stat.dec_exits++;
1298                 r = RESUME_GUEST;
1299                 break;
1300         case BOOK3S_INTERRUPT_EXTERNAL:
1301         case BOOK3S_INTERRUPT_H_DOORBELL:
1302         case BOOK3S_INTERRUPT_H_VIRT:
1303                 vcpu->stat.ext_intr_exits++;
1304                 r = RESUME_GUEST;
1305                 break;
1306         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1307         case BOOK3S_INTERRUPT_HMI:
1308         case BOOK3S_INTERRUPT_PERFMON:
1309         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1310                 r = RESUME_GUEST;
1311                 break;
1312         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1313                 /* Print the MCE event to host console. */
1314                 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1315
1316                 /*
1317                  * If the guest can do FWNMI, exit to userspace so it can
1318                  * deliver a FWNMI to the guest.
1319                  * Otherwise we synthesize a machine check for the guest
1320                  * so that it knows that the machine check occurred.
1321                  */
1322                 if (!vcpu->kvm->arch.fwnmi_enabled) {
1323                         ulong flags = vcpu->arch.shregs.msr & 0x083c0000;
1324                         kvmppc_core_queue_machine_check(vcpu, flags);
1325                         r = RESUME_GUEST;
1326                         break;
1327                 }
1328
1329                 /* Exit to guest with KVM_EXIT_NMI as exit reason */
1330                 run->exit_reason = KVM_EXIT_NMI;
1331                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1332                 /* Clear out the old NMI status from run->flags */
1333                 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK;
1334                 /* Now set the NMI status */
1335                 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED)
1336                         run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV;
1337                 else
1338                         run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV;
1339
1340                 r = RESUME_HOST;
1341                 break;
1342         case BOOK3S_INTERRUPT_PROGRAM:
1343         {
1344                 ulong flags;
1345                 /*
1346                  * Normally program interrupts are delivered directly
1347                  * to the guest by the hardware, but we can get here
1348                  * as a result of a hypervisor emulation interrupt
1349                  * (e40) getting turned into a 700 by BML RTAS.
1350                  */
1351                 flags = vcpu->arch.shregs.msr & 0x1f0000ull;
1352                 kvmppc_core_queue_program(vcpu, flags);
1353                 r = RESUME_GUEST;
1354                 break;
1355         }
1356         case BOOK3S_INTERRUPT_SYSCALL:
1357         {
1358                 /* hcall - punt to userspace */
1359                 int i;
1360
1361                 /* hypercall with MSR_PR has already been handled in rmode,
1362                  * and never reaches here.
1363                  */
1364
1365                 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3);
1366                 for (i = 0; i < 9; ++i)
1367                         run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i);
1368                 run->exit_reason = KVM_EXIT_PAPR_HCALL;
1369                 vcpu->arch.hcall_needed = 1;
1370                 r = RESUME_HOST;
1371                 break;
1372         }
1373         /*
1374          * We get these next two if the guest accesses a page which it thinks
1375          * it has mapped but which is not actually present, either because
1376          * it is for an emulated I/O device or because the corresonding
1377          * host page has been paged out.  Any other HDSI/HISI interrupts
1378          * have been handled already.
1379          */
1380         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1381                 r = RESUME_PAGE_FAULT;
1382                 break;
1383         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1384                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1385                 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr &
1386                         DSISR_SRR1_MATCH_64S;
1387                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1388                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1389                 r = RESUME_PAGE_FAULT;
1390                 break;
1391         /*
1392          * This occurs if the guest executes an illegal instruction.
1393          * If the guest debug is disabled, generate a program interrupt
1394          * to the guest. If guest debug is enabled, we need to check
1395          * whether the instruction is a software breakpoint instruction.
1396          * Accordingly return to Guest or Host.
1397          */
1398         case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
1399                 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED)
1400                         vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ?
1401                                 swab32(vcpu->arch.emul_inst) :
1402                                 vcpu->arch.emul_inst;
1403                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) {
1404                         r = kvmppc_emulate_debug_inst(run, vcpu);
1405                 } else {
1406                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1407                         r = RESUME_GUEST;
1408                 }
1409                 break;
1410         /*
1411          * This occurs if the guest (kernel or userspace), does something that
1412          * is prohibited by HFSCR.
1413          * On POWER9, this could be a doorbell instruction that we need
1414          * to emulate.
1415          * Otherwise, we just generate a program interrupt to the guest.
1416          */
1417         case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
1418                 r = EMULATE_FAIL;
1419                 if (((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG) &&
1420                     cpu_has_feature(CPU_FTR_ARCH_300))
1421                         r = kvmppc_emulate_doorbell_instr(vcpu);
1422                 if (r == EMULATE_FAIL) {
1423                         kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
1424                         r = RESUME_GUEST;
1425                 }
1426                 break;
1427
1428 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1429         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1430                 /*
1431                  * This occurs for various TM-related instructions that
1432                  * we need to emulate on POWER9 DD2.2.  We have already
1433                  * handled the cases where the guest was in real-suspend
1434                  * mode and was transitioning to transactional state.
1435                  */
1436                 r = kvmhv_p9_tm_emulation(vcpu);
1437                 break;
1438 #endif
1439
1440         case BOOK3S_INTERRUPT_HV_RM_HARD:
1441                 r = RESUME_PASSTHROUGH;
1442                 break;
1443         default:
1444                 kvmppc_dump_regs(vcpu);
1445                 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1446                         vcpu->arch.trap, kvmppc_get_pc(vcpu),
1447                         vcpu->arch.shregs.msr);
1448                 run->hw.hardware_exit_reason = vcpu->arch.trap;
1449                 r = RESUME_HOST;
1450                 break;
1451         }
1452
1453         return r;
1454 }
1455
1456 static int kvmppc_handle_nested_exit(struct kvm_run *run, struct kvm_vcpu *vcpu)
1457 {
1458         int r;
1459         int srcu_idx;
1460
1461         vcpu->stat.sum_exits++;
1462
1463         /*
1464          * This can happen if an interrupt occurs in the last stages
1465          * of guest entry or the first stages of guest exit (i.e. after
1466          * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV
1467          * and before setting it to KVM_GUEST_MODE_HOST_HV).
1468          * That can happen due to a bug, or due to a machine check
1469          * occurring at just the wrong time.
1470          */
1471         if (vcpu->arch.shregs.msr & MSR_HV) {
1472                 pr_emerg("KVM trap in HV mode while nested!\n");
1473                 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n",
1474                          vcpu->arch.trap, kvmppc_get_pc(vcpu),
1475                          vcpu->arch.shregs.msr);
1476                 kvmppc_dump_regs(vcpu);
1477                 return RESUME_HOST;
1478         }
1479         switch (vcpu->arch.trap) {
1480         /* We're good on these - the host merely wanted to get our attention */
1481         case BOOK3S_INTERRUPT_HV_DECREMENTER:
1482                 vcpu->stat.dec_exits++;
1483                 r = RESUME_GUEST;
1484                 break;
1485         case BOOK3S_INTERRUPT_EXTERNAL:
1486                 vcpu->stat.ext_intr_exits++;
1487                 r = RESUME_HOST;
1488                 break;
1489         case BOOK3S_INTERRUPT_H_DOORBELL:
1490         case BOOK3S_INTERRUPT_H_VIRT:
1491                 vcpu->stat.ext_intr_exits++;
1492                 r = RESUME_GUEST;
1493                 break;
1494         /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/
1495         case BOOK3S_INTERRUPT_HMI:
1496         case BOOK3S_INTERRUPT_PERFMON:
1497         case BOOK3S_INTERRUPT_SYSTEM_RESET:
1498                 r = RESUME_GUEST;
1499                 break;
1500         case BOOK3S_INTERRUPT_MACHINE_CHECK:
1501                 /* Pass the machine check to the L1 guest */
1502                 r = RESUME_HOST;
1503                 /* Print the MCE event to host console. */
1504                 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true);
1505                 break;
1506         /*
1507          * We get these next two if the guest accesses a page which it thinks
1508          * it has mapped but which is not actually present, either because
1509          * it is for an emulated I/O device or because the corresonding
1510          * host page has been paged out.
1511          */
1512         case BOOK3S_INTERRUPT_H_DATA_STORAGE:
1513                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1514                 r = kvmhv_nested_page_fault(run, vcpu);
1515                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1516                 break;
1517         case BOOK3S_INTERRUPT_H_INST_STORAGE:
1518                 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu);
1519                 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) &
1520                                          DSISR_SRR1_MATCH_64S;
1521                 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE)
1522                         vcpu->arch.fault_dsisr |= DSISR_ISSTORE;
1523                 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1524                 r = kvmhv_nested_page_fault(run, vcpu);
1525                 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
1526                 break;
1527
1528 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1529         case BOOK3S_INTERRUPT_HV_SOFTPATCH:
1530                 /*
1531                  * This occurs for various TM-related instructions that
1532                  * we need to emulate on POWER9 DD2.2.  We have already
1533                  * handled the cases where the guest was in real-suspend
1534                  * mode and was transitioning to transactional state.
1535                  */
1536                 r = kvmhv_p9_tm_emulation(vcpu);
1537                 break;
1538 #endif
1539
1540         case BOOK3S_INTERRUPT_HV_RM_HARD:
1541                 vcpu->arch.trap = 0;
1542                 r = RESUME_GUEST;
1543                 if (!xics_on_xive())
1544                         kvmppc_xics_rm_complete(vcpu, 0);
1545                 break;
1546         default:
1547                 r = RESUME_HOST;
1548                 break;
1549         }
1550
1551         return r;
1552 }
1553
1554 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu,
1555                                             struct kvm_sregs *sregs)
1556 {
1557         int i;
1558
1559         memset(sregs, 0, sizeof(struct kvm_sregs));
1560         sregs->pvr = vcpu->arch.pvr;
1561         for (i = 0; i < vcpu->arch.slb_max; i++) {
1562                 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige;
1563                 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv;
1564         }
1565
1566         return 0;
1567 }
1568
1569 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu,
1570                                             struct kvm_sregs *sregs)
1571 {
1572         int i, j;
1573
1574         /* Only accept the same PVR as the host's, since we can't spoof it */
1575         if (sregs->pvr != vcpu->arch.pvr)
1576                 return -EINVAL;
1577
1578         j = 0;
1579         for (i = 0; i < vcpu->arch.slb_nr; i++) {
1580                 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) {
1581                         vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe;
1582                         vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv;
1583                         ++j;
1584                 }
1585         }
1586         vcpu->arch.slb_max = j;
1587
1588         return 0;
1589 }
1590
1591 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr,
1592                 bool preserve_top32)
1593 {
1594         struct kvm *kvm = vcpu->kvm;
1595         struct kvmppc_vcore *vc = vcpu->arch.vcore;
1596         u64 mask;
1597
1598         spin_lock(&vc->lock);
1599         /*
1600          * If ILE (interrupt little-endian) has changed, update the
1601          * MSR_LE bit in the intr_msr for each vcpu in this vcore.
1602          */
1603         if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) {
1604                 struct kvm_vcpu *vcpu;
1605                 int i;
1606
1607                 kvm_for_each_vcpu(i, vcpu, kvm) {
1608                         if (vcpu->arch.vcore != vc)
1609                                 continue;
1610                         if (new_lpcr & LPCR_ILE)
1611                                 vcpu->arch.intr_msr |= MSR_LE;
1612                         else
1613                                 vcpu->arch.intr_msr &= ~MSR_LE;
1614                 }
1615         }
1616
1617         /*
1618          * Userspace can only modify DPFD (default prefetch depth),
1619          * ILE (interrupt little-endian) and TC (translation control).
1620          * On POWER8 and POWER9 userspace can also modify AIL (alt. interrupt loc.).
1621          */
1622         mask = LPCR_DPFD | LPCR_ILE | LPCR_TC;
1623         if (cpu_has_feature(CPU_FTR_ARCH_207S))
1624                 mask |= LPCR_AIL;
1625         /*
1626          * On POWER9, allow userspace to enable large decrementer for the
1627          * guest, whether or not the host has it enabled.
1628          */
1629         if (cpu_has_feature(CPU_FTR_ARCH_300))
1630                 mask |= LPCR_LD;
1631
1632         /* Broken 32-bit version of LPCR must not clear top bits */
1633         if (preserve_top32)
1634                 mask &= 0xFFFFFFFF;
1635         vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask);
1636         spin_unlock(&vc->lock);
1637 }
1638
1639 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1640                                  union kvmppc_one_reg *val)
1641 {
1642         int r = 0;
1643         long int i;
1644
1645         switch (id) {
1646         case KVM_REG_PPC_DEBUG_INST:
1647                 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT);
1648                 break;
1649         case KVM_REG_PPC_HIOR:
1650                 *val = get_reg_val(id, 0);
1651                 break;
1652         case KVM_REG_PPC_DABR:
1653                 *val = get_reg_val(id, vcpu->arch.dabr);
1654                 break;
1655         case KVM_REG_PPC_DABRX:
1656                 *val = get_reg_val(id, vcpu->arch.dabrx);
1657                 break;
1658         case KVM_REG_PPC_DSCR:
1659                 *val = get_reg_val(id, vcpu->arch.dscr);
1660                 break;
1661         case KVM_REG_PPC_PURR:
1662                 *val = get_reg_val(id, vcpu->arch.purr);
1663                 break;
1664         case KVM_REG_PPC_SPURR:
1665                 *val = get_reg_val(id, vcpu->arch.spurr);
1666                 break;
1667         case KVM_REG_PPC_AMR:
1668                 *val = get_reg_val(id, vcpu->arch.amr);
1669                 break;
1670         case KVM_REG_PPC_UAMOR:
1671                 *val = get_reg_val(id, vcpu->arch.uamor);
1672                 break;
1673         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1674                 i = id - KVM_REG_PPC_MMCR0;
1675                 *val = get_reg_val(id, vcpu->arch.mmcr[i]);
1676                 break;
1677         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1678                 i = id - KVM_REG_PPC_PMC1;
1679                 *val = get_reg_val(id, vcpu->arch.pmc[i]);
1680                 break;
1681         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1682                 i = id - KVM_REG_PPC_SPMC1;
1683                 *val = get_reg_val(id, vcpu->arch.spmc[i]);
1684                 break;
1685         case KVM_REG_PPC_SIAR:
1686                 *val = get_reg_val(id, vcpu->arch.siar);
1687                 break;
1688         case KVM_REG_PPC_SDAR:
1689                 *val = get_reg_val(id, vcpu->arch.sdar);
1690                 break;
1691         case KVM_REG_PPC_SIER:
1692                 *val = get_reg_val(id, vcpu->arch.sier);
1693                 break;
1694         case KVM_REG_PPC_IAMR:
1695                 *val = get_reg_val(id, vcpu->arch.iamr);
1696                 break;
1697         case KVM_REG_PPC_PSPB:
1698                 *val = get_reg_val(id, vcpu->arch.pspb);
1699                 break;
1700         case KVM_REG_PPC_DPDES:
1701                 /*
1702                  * On POWER9, where we are emulating msgsndp etc.,
1703                  * we return 1 bit for each vcpu, which can come from
1704                  * either vcore->dpdes or doorbell_request.
1705                  * On POWER8, doorbell_request is 0.
1706                  */
1707                 *val = get_reg_val(id, vcpu->arch.vcore->dpdes |
1708                                    vcpu->arch.doorbell_request);
1709                 break;
1710         case KVM_REG_PPC_VTB:
1711                 *val = get_reg_val(id, vcpu->arch.vcore->vtb);
1712                 break;
1713         case KVM_REG_PPC_DAWR:
1714                 *val = get_reg_val(id, vcpu->arch.dawr);
1715                 break;
1716         case KVM_REG_PPC_DAWRX:
1717                 *val = get_reg_val(id, vcpu->arch.dawrx);
1718                 break;
1719         case KVM_REG_PPC_CIABR:
1720                 *val = get_reg_val(id, vcpu->arch.ciabr);
1721                 break;
1722         case KVM_REG_PPC_CSIGR:
1723                 *val = get_reg_val(id, vcpu->arch.csigr);
1724                 break;
1725         case KVM_REG_PPC_TACR:
1726                 *val = get_reg_val(id, vcpu->arch.tacr);
1727                 break;
1728         case KVM_REG_PPC_TCSCR:
1729                 *val = get_reg_val(id, vcpu->arch.tcscr);
1730                 break;
1731         case KVM_REG_PPC_PID:
1732                 *val = get_reg_val(id, vcpu->arch.pid);
1733                 break;
1734         case KVM_REG_PPC_ACOP:
1735                 *val = get_reg_val(id, vcpu->arch.acop);
1736                 break;
1737         case KVM_REG_PPC_WORT:
1738                 *val = get_reg_val(id, vcpu->arch.wort);
1739                 break;
1740         case KVM_REG_PPC_TIDR:
1741                 *val = get_reg_val(id, vcpu->arch.tid);
1742                 break;
1743         case KVM_REG_PPC_PSSCR:
1744                 *val = get_reg_val(id, vcpu->arch.psscr);
1745                 break;
1746         case KVM_REG_PPC_VPA_ADDR:
1747                 spin_lock(&vcpu->arch.vpa_update_lock);
1748                 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa);
1749                 spin_unlock(&vcpu->arch.vpa_update_lock);
1750                 break;
1751         case KVM_REG_PPC_VPA_SLB:
1752                 spin_lock(&vcpu->arch.vpa_update_lock);
1753                 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa;
1754                 val->vpaval.length = vcpu->arch.slb_shadow.len;
1755                 spin_unlock(&vcpu->arch.vpa_update_lock);
1756                 break;
1757         case KVM_REG_PPC_VPA_DTL:
1758                 spin_lock(&vcpu->arch.vpa_update_lock);
1759                 val->vpaval.addr = vcpu->arch.dtl.next_gpa;
1760                 val->vpaval.length = vcpu->arch.dtl.len;
1761                 spin_unlock(&vcpu->arch.vpa_update_lock);
1762                 break;
1763         case KVM_REG_PPC_TB_OFFSET:
1764                 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset);
1765                 break;
1766         case KVM_REG_PPC_LPCR:
1767         case KVM_REG_PPC_LPCR_64:
1768                 *val = get_reg_val(id, vcpu->arch.vcore->lpcr);
1769                 break;
1770         case KVM_REG_PPC_PPR:
1771                 *val = get_reg_val(id, vcpu->arch.ppr);
1772                 break;
1773 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1774         case KVM_REG_PPC_TFHAR:
1775                 *val = get_reg_val(id, vcpu->arch.tfhar);
1776                 break;
1777         case KVM_REG_PPC_TFIAR:
1778                 *val = get_reg_val(id, vcpu->arch.tfiar);
1779                 break;
1780         case KVM_REG_PPC_TEXASR:
1781                 *val = get_reg_val(id, vcpu->arch.texasr);
1782                 break;
1783         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
1784                 i = id - KVM_REG_PPC_TM_GPR0;
1785                 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]);
1786                 break;
1787         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
1788         {
1789                 int j;
1790                 i = id - KVM_REG_PPC_TM_VSR0;
1791                 if (i < 32)
1792                         for (j = 0; j < TS_FPRWIDTH; j++)
1793                                 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j];
1794                 else {
1795                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1796                                 val->vval = vcpu->arch.vr_tm.vr[i-32];
1797                         else
1798                                 r = -ENXIO;
1799                 }
1800                 break;
1801         }
1802         case KVM_REG_PPC_TM_CR:
1803                 *val = get_reg_val(id, vcpu->arch.cr_tm);
1804                 break;
1805         case KVM_REG_PPC_TM_XER:
1806                 *val = get_reg_val(id, vcpu->arch.xer_tm);
1807                 break;
1808         case KVM_REG_PPC_TM_LR:
1809                 *val = get_reg_val(id, vcpu->arch.lr_tm);
1810                 break;
1811         case KVM_REG_PPC_TM_CTR:
1812                 *val = get_reg_val(id, vcpu->arch.ctr_tm);
1813                 break;
1814         case KVM_REG_PPC_TM_FPSCR:
1815                 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr);
1816                 break;
1817         case KVM_REG_PPC_TM_AMR:
1818                 *val = get_reg_val(id, vcpu->arch.amr_tm);
1819                 break;
1820         case KVM_REG_PPC_TM_PPR:
1821                 *val = get_reg_val(id, vcpu->arch.ppr_tm);
1822                 break;
1823         case KVM_REG_PPC_TM_VRSAVE:
1824                 *val = get_reg_val(id, vcpu->arch.vrsave_tm);
1825                 break;
1826         case KVM_REG_PPC_TM_VSCR:
1827                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
1828                         *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]);
1829                 else
1830                         r = -ENXIO;
1831                 break;
1832         case KVM_REG_PPC_TM_DSCR:
1833                 *val = get_reg_val(id, vcpu->arch.dscr_tm);
1834                 break;
1835         case KVM_REG_PPC_TM_TAR:
1836                 *val = get_reg_val(id, vcpu->arch.tar_tm);
1837                 break;
1838 #endif
1839         case KVM_REG_PPC_ARCH_COMPAT:
1840                 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat);
1841                 break;
1842         case KVM_REG_PPC_DEC_EXPIRY:
1843                 *val = get_reg_val(id, vcpu->arch.dec_expires +
1844                                    vcpu->arch.vcore->tb_offset);
1845                 break;
1846         case KVM_REG_PPC_ONLINE:
1847                 *val = get_reg_val(id, vcpu->arch.online);
1848                 break;
1849         case KVM_REG_PPC_PTCR:
1850                 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr);
1851                 break;
1852         default:
1853                 r = -EINVAL;
1854                 break;
1855         }
1856
1857         return r;
1858 }
1859
1860 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id,
1861                                  union kvmppc_one_reg *val)
1862 {
1863         int r = 0;
1864         long int i;
1865         unsigned long addr, len;
1866
1867         switch (id) {
1868         case KVM_REG_PPC_HIOR:
1869                 /* Only allow this to be set to zero */
1870                 if (set_reg_val(id, *val))
1871                         r = -EINVAL;
1872                 break;
1873         case KVM_REG_PPC_DABR:
1874                 vcpu->arch.dabr = set_reg_val(id, *val);
1875                 break;
1876         case KVM_REG_PPC_DABRX:
1877                 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP;
1878                 break;
1879         case KVM_REG_PPC_DSCR:
1880                 vcpu->arch.dscr = set_reg_val(id, *val);
1881                 break;
1882         case KVM_REG_PPC_PURR:
1883                 vcpu->arch.purr = set_reg_val(id, *val);
1884                 break;
1885         case KVM_REG_PPC_SPURR:
1886                 vcpu->arch.spurr = set_reg_val(id, *val);
1887                 break;
1888         case KVM_REG_PPC_AMR:
1889                 vcpu->arch.amr = set_reg_val(id, *val);
1890                 break;
1891         case KVM_REG_PPC_UAMOR:
1892                 vcpu->arch.uamor = set_reg_val(id, *val);
1893                 break;
1894         case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS:
1895                 i = id - KVM_REG_PPC_MMCR0;
1896                 vcpu->arch.mmcr[i] = set_reg_val(id, *val);
1897                 break;
1898         case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8:
1899                 i = id - KVM_REG_PPC_PMC1;
1900                 vcpu->arch.pmc[i] = set_reg_val(id, *val);
1901                 break;
1902         case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2:
1903                 i = id - KVM_REG_PPC_SPMC1;
1904                 vcpu->arch.spmc[i] = set_reg_val(id, *val);
1905                 break;
1906         case KVM_REG_PPC_SIAR:
1907                 vcpu->arch.siar = set_reg_val(id, *val);
1908                 break;
1909         case KVM_REG_PPC_SDAR:
1910                 vcpu->arch.sdar = set_reg_val(id, *val);
1911                 break;
1912         case KVM_REG_PPC_SIER:
1913                 vcpu->arch.sier = set_reg_val(id, *val);
1914                 break;
1915         case KVM_REG_PPC_IAMR:
1916                 vcpu->arch.iamr = set_reg_val(id, *val);
1917                 break;
1918         case KVM_REG_PPC_PSPB:
1919                 vcpu->arch.pspb = set_reg_val(id, *val);
1920                 break;
1921         case KVM_REG_PPC_DPDES:
1922                 vcpu->arch.vcore->dpdes = set_reg_val(id, *val);
1923                 break;
1924         case KVM_REG_PPC_VTB:
1925                 vcpu->arch.vcore->vtb = set_reg_val(id, *val);
1926                 break;
1927         case KVM_REG_PPC_DAWR:
1928                 vcpu->arch.dawr = set_reg_val(id, *val);
1929                 break;
1930         case KVM_REG_PPC_DAWRX:
1931                 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP;
1932                 break;
1933         case KVM_REG_PPC_CIABR:
1934                 vcpu->arch.ciabr = set_reg_val(id, *val);
1935                 /* Don't allow setting breakpoints in hypervisor code */
1936                 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER)
1937                         vcpu->arch.ciabr &= ~CIABR_PRIV;        /* disable */
1938                 break;
1939         case KVM_REG_PPC_CSIGR:
1940                 vcpu->arch.csigr = set_reg_val(id, *val);
1941                 break;
1942         case KVM_REG_PPC_TACR:
1943                 vcpu->arch.tacr = set_reg_val(id, *val);
1944                 break;
1945         case KVM_REG_PPC_TCSCR:
1946                 vcpu->arch.tcscr = set_reg_val(id, *val);
1947                 break;
1948         case KVM_REG_PPC_PID:
1949                 vcpu->arch.pid = set_reg_val(id, *val);
1950                 break;
1951         case KVM_REG_PPC_ACOP:
1952                 vcpu->arch.acop = set_reg_val(id, *val);
1953                 break;
1954         case KVM_REG_PPC_WORT:
1955                 vcpu->arch.wort = set_reg_val(id, *val);
1956                 break;
1957         case KVM_REG_PPC_TIDR:
1958                 vcpu->arch.tid = set_reg_val(id, *val);
1959                 break;
1960         case KVM_REG_PPC_PSSCR:
1961                 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS;
1962                 break;
1963         case KVM_REG_PPC_VPA_ADDR:
1964                 addr = set_reg_val(id, *val);
1965                 r = -EINVAL;
1966                 if (!addr && (vcpu->arch.slb_shadow.next_gpa ||
1967                               vcpu->arch.dtl.next_gpa))
1968                         break;
1969                 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca));
1970                 break;
1971         case KVM_REG_PPC_VPA_SLB:
1972                 addr = val->vpaval.addr;
1973                 len = val->vpaval.length;
1974                 r = -EINVAL;
1975                 if (addr && !vcpu->arch.vpa.next_gpa)
1976                         break;
1977                 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len);
1978                 break;
1979         case KVM_REG_PPC_VPA_DTL:
1980                 addr = val->vpaval.addr;
1981                 len = val->vpaval.length;
1982                 r = -EINVAL;
1983                 if (addr && (len < sizeof(struct dtl_entry) ||
1984                              !vcpu->arch.vpa.next_gpa))
1985                         break;
1986                 len -= len % sizeof(struct dtl_entry);
1987                 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len);
1988                 break;
1989         case KVM_REG_PPC_TB_OFFSET:
1990                 /* round up to multiple of 2^24 */
1991                 vcpu->arch.vcore->tb_offset =
1992                         ALIGN(set_reg_val(id, *val), 1UL << 24);
1993                 break;
1994         case KVM_REG_PPC_LPCR:
1995                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true);
1996                 break;
1997         case KVM_REG_PPC_LPCR_64:
1998                 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false);
1999                 break;
2000         case KVM_REG_PPC_PPR:
2001                 vcpu->arch.ppr = set_reg_val(id, *val);
2002                 break;
2003 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2004         case KVM_REG_PPC_TFHAR:
2005                 vcpu->arch.tfhar = set_reg_val(id, *val);
2006                 break;
2007         case KVM_REG_PPC_TFIAR:
2008                 vcpu->arch.tfiar = set_reg_val(id, *val);
2009                 break;
2010         case KVM_REG_PPC_TEXASR:
2011                 vcpu->arch.texasr = set_reg_val(id, *val);
2012                 break;
2013         case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31:
2014                 i = id - KVM_REG_PPC_TM_GPR0;
2015                 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val);
2016                 break;
2017         case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63:
2018         {
2019                 int j;
2020                 i = id - KVM_REG_PPC_TM_VSR0;
2021                 if (i < 32)
2022                         for (j = 0; j < TS_FPRWIDTH; j++)
2023                                 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j];
2024                 else
2025                         if (cpu_has_feature(CPU_FTR_ALTIVEC))
2026                                 vcpu->arch.vr_tm.vr[i-32] = val->vval;
2027                         else
2028                                 r = -ENXIO;
2029                 break;
2030         }
2031         case KVM_REG_PPC_TM_CR:
2032                 vcpu->arch.cr_tm = set_reg_val(id, *val);
2033                 break;
2034         case KVM_REG_PPC_TM_XER:
2035                 vcpu->arch.xer_tm = set_reg_val(id, *val);
2036                 break;
2037         case KVM_REG_PPC_TM_LR:
2038                 vcpu->arch.lr_tm = set_reg_val(id, *val);
2039                 break;
2040         case KVM_REG_PPC_TM_CTR:
2041                 vcpu->arch.ctr_tm = set_reg_val(id, *val);
2042                 break;
2043         case KVM_REG_PPC_TM_FPSCR:
2044                 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val);
2045                 break;
2046         case KVM_REG_PPC_TM_AMR:
2047                 vcpu->arch.amr_tm = set_reg_val(id, *val);
2048                 break;
2049         case KVM_REG_PPC_TM_PPR:
2050                 vcpu->arch.ppr_tm = set_reg_val(id, *val);
2051                 break;
2052         case KVM_REG_PPC_TM_VRSAVE:
2053                 vcpu->arch.vrsave_tm = set_reg_val(id, *val);
2054                 break;
2055         case KVM_REG_PPC_TM_VSCR:
2056                 if (cpu_has_feature(CPU_FTR_ALTIVEC))
2057                         vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val);
2058                 else
2059                         r = - ENXIO;
2060                 break;
2061         case KVM_REG_PPC_TM_DSCR:
2062                 vcpu->arch.dscr_tm = set_reg_val(id, *val);
2063                 break;
2064         case KVM_REG_PPC_TM_TAR:
2065                 vcpu->arch.tar_tm = set_reg_val(id, *val);
2066                 break;
2067 #endif
2068         case KVM_REG_PPC_ARCH_COMPAT:
2069                 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val));
2070                 break;
2071         case KVM_REG_PPC_DEC_EXPIRY:
2072                 vcpu->arch.dec_expires = set_reg_val(id, *val) -
2073                         vcpu->arch.vcore->tb_offset;
2074                 break;
2075         case KVM_REG_PPC_ONLINE:
2076                 i = set_reg_val(id, *val);
2077                 if (i && !vcpu->arch.online)
2078                         atomic_inc(&vcpu->arch.vcore->online_count);
2079                 else if (!i && vcpu->arch.online)
2080                         atomic_dec(&vcpu->arch.vcore->online_count);
2081                 vcpu->arch.online = i;
2082                 break;
2083         case KVM_REG_PPC_PTCR:
2084                 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val);
2085                 break;
2086         default:
2087                 r = -EINVAL;
2088                 break;
2089         }
2090
2091         return r;
2092 }
2093
2094 /*
2095  * On POWER9, threads are independent and can be in different partitions.
2096  * Therefore we consider each thread to be a subcore.
2097  * There is a restriction that all threads have to be in the same
2098  * MMU mode (radix or HPT), unfortunately, but since we only support
2099  * HPT guests on a HPT host so far, that isn't an impediment yet.
2100  */
2101 static int threads_per_vcore(struct kvm *kvm)
2102 {
2103         if (kvm->arch.threads_indep)
2104                 return 1;
2105         return threads_per_subcore;
2106 }
2107
2108 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
2109 {
2110         struct kvmppc_vcore *vcore;
2111
2112         vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL);
2113
2114         if (vcore == NULL)
2115                 return NULL;
2116
2117         spin_lock_init(&vcore->lock);
2118         spin_lock_init(&vcore->stoltb_lock);
2119         init_swait_queue_head(&vcore->wq);
2120         vcore->preempt_tb = TB_NIL;
2121         vcore->lpcr = kvm->arch.lpcr;
2122         vcore->first_vcpuid = id;
2123         vcore->kvm = kvm;
2124         INIT_LIST_HEAD(&vcore->preempt_list);
2125
2126         return vcore;
2127 }
2128
2129 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2130 static struct debugfs_timings_element {
2131         const char *name;
2132         size_t offset;
2133 } timings[] = {
2134         {"rm_entry",    offsetof(struct kvm_vcpu, arch.rm_entry)},
2135         {"rm_intr",     offsetof(struct kvm_vcpu, arch.rm_intr)},
2136         {"rm_exit",     offsetof(struct kvm_vcpu, arch.rm_exit)},
2137         {"guest",       offsetof(struct kvm_vcpu, arch.guest_time)},
2138         {"cede",        offsetof(struct kvm_vcpu, arch.cede_time)},
2139 };
2140
2141 #define N_TIMINGS       (ARRAY_SIZE(timings))
2142
2143 struct debugfs_timings_state {
2144         struct kvm_vcpu *vcpu;
2145         unsigned int    buflen;
2146         char            buf[N_TIMINGS * 100];
2147 };
2148
2149 static int debugfs_timings_open(struct inode *inode, struct file *file)
2150 {
2151         struct kvm_vcpu *vcpu = inode->i_private;
2152         struct debugfs_timings_state *p;
2153
2154         p = kzalloc(sizeof(*p), GFP_KERNEL);
2155         if (!p)
2156                 return -ENOMEM;
2157
2158         kvm_get_kvm(vcpu->kvm);
2159         p->vcpu = vcpu;
2160         file->private_data = p;
2161
2162         return nonseekable_open(inode, file);
2163 }
2164
2165 static int debugfs_timings_release(struct inode *inode, struct file *file)
2166 {
2167         struct debugfs_timings_state *p = file->private_data;
2168
2169         kvm_put_kvm(p->vcpu->kvm);
2170         kfree(p);
2171         return 0;
2172 }
2173
2174 static ssize_t debugfs_timings_read(struct file *file, char __user *buf,
2175                                     size_t len, loff_t *ppos)
2176 {
2177         struct debugfs_timings_state *p = file->private_data;
2178         struct kvm_vcpu *vcpu = p->vcpu;
2179         char *s, *buf_end;
2180         struct kvmhv_tb_accumulator tb;
2181         u64 count;
2182         loff_t pos;
2183         ssize_t n;
2184         int i, loops;
2185         bool ok;
2186
2187         if (!p->buflen) {
2188                 s = p->buf;
2189                 buf_end = s + sizeof(p->buf);
2190                 for (i = 0; i < N_TIMINGS; ++i) {
2191                         struct kvmhv_tb_accumulator *acc;
2192
2193                         acc = (struct kvmhv_tb_accumulator *)
2194                                 ((unsigned long)vcpu + timings[i].offset);
2195                         ok = false;
2196                         for (loops = 0; loops < 1000; ++loops) {
2197                                 count = acc->seqcount;
2198                                 if (!(count & 1)) {
2199                                         smp_rmb();
2200                                         tb = *acc;
2201                                         smp_rmb();
2202                                         if (count == acc->seqcount) {
2203                                                 ok = true;
2204                                                 break;
2205                                         }
2206                                 }
2207                                 udelay(1);
2208                         }
2209                         if (!ok)
2210                                 snprintf(s, buf_end - s, "%s: stuck\n",
2211                                         timings[i].name);
2212                         else
2213                                 snprintf(s, buf_end - s,
2214                                         "%s: %llu %llu %llu %llu\n",
2215                                         timings[i].name, count / 2,
2216                                         tb_to_ns(tb.tb_total),
2217                                         tb_to_ns(tb.tb_min),
2218                                         tb_to_ns(tb.tb_max));
2219                         s += strlen(s);
2220                 }
2221                 p->buflen = s - p->buf;
2222         }
2223
2224         pos = *ppos;
2225         if (pos >= p->buflen)
2226                 return 0;
2227         if (len > p->buflen - pos)
2228                 len = p->buflen - pos;
2229         n = copy_to_user(buf, p->buf + pos, len);
2230         if (n) {
2231                 if (n == len)
2232                         return -EFAULT;
2233                 len -= n;
2234         }
2235         *ppos = pos + len;
2236         return len;
2237 }
2238
2239 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf,
2240                                      size_t len, loff_t *ppos)
2241 {
2242         return -EACCES;
2243 }
2244
2245 static const struct file_operations debugfs_timings_ops = {
2246         .owner   = THIS_MODULE,
2247         .open    = debugfs_timings_open,
2248         .release = debugfs_timings_release,
2249         .read    = debugfs_timings_read,
2250         .write   = debugfs_timings_write,
2251         .llseek  = generic_file_llseek,
2252 };
2253
2254 /* Create a debugfs directory for the vcpu */
2255 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2256 {
2257         char buf[16];
2258         struct kvm *kvm = vcpu->kvm;
2259
2260         snprintf(buf, sizeof(buf), "vcpu%u", id);
2261         if (IS_ERR_OR_NULL(kvm->arch.debugfs_dir))
2262                 return;
2263         vcpu->arch.debugfs_dir = debugfs_create_dir(buf, kvm->arch.debugfs_dir);
2264         if (IS_ERR_OR_NULL(vcpu->arch.debugfs_dir))
2265                 return;
2266         vcpu->arch.debugfs_timings =
2267                 debugfs_create_file("timings", 0444, vcpu->arch.debugfs_dir,
2268                                     vcpu, &debugfs_timings_ops);
2269 }
2270
2271 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2272 static void debugfs_vcpu_init(struct kvm_vcpu *vcpu, unsigned int id)
2273 {
2274 }
2275 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */
2276
2277 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
2278                                                    unsigned int id)
2279 {
2280         struct kvm_vcpu *vcpu;
2281         int err;
2282         int core;
2283         struct kvmppc_vcore *vcore;
2284
2285         err = -ENOMEM;
2286         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2287         if (!vcpu)
2288                 goto out;
2289
2290         err = kvm_vcpu_init(vcpu, kvm, id);
2291         if (err)
2292                 goto free_vcpu;
2293
2294         vcpu->arch.shared = &vcpu->arch.shregs;
2295 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
2296         /*
2297          * The shared struct is never shared on HV,
2298          * so we can always use host endianness
2299          */
2300 #ifdef __BIG_ENDIAN__
2301         vcpu->arch.shared_big_endian = true;
2302 #else
2303         vcpu->arch.shared_big_endian = false;
2304 #endif
2305 #endif
2306         vcpu->arch.mmcr[0] = MMCR0_FC;
2307         vcpu->arch.ctrl = CTRL_RUNLATCH;
2308         /* default to host PVR, since we can't spoof it */
2309         kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR));
2310         spin_lock_init(&vcpu->arch.vpa_update_lock);
2311         spin_lock_init(&vcpu->arch.tbacct_lock);
2312         vcpu->arch.busy_preempt = TB_NIL;
2313         vcpu->arch.intr_msr = MSR_SF | MSR_ME;
2314
2315         /*
2316          * Set the default HFSCR for the guest from the host value.
2317          * This value is only used on POWER9.
2318          * On POWER9, we want to virtualize the doorbell facility, so we
2319          * don't set the HFSCR_MSGP bit, and that causes those instructions
2320          * to trap and then we emulate them.
2321          */
2322         vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB |
2323                 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP;
2324         if (cpu_has_feature(CPU_FTR_HVMODE)) {
2325                 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR);
2326                 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
2327                         vcpu->arch.hfscr |= HFSCR_TM;
2328         }
2329         if (cpu_has_feature(CPU_FTR_TM_COMP))
2330                 vcpu->arch.hfscr |= HFSCR_TM;
2331
2332         kvmppc_mmu_book3s_hv_init(vcpu);
2333
2334         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
2335
2336         init_waitqueue_head(&vcpu->arch.cpu_run);
2337
2338         mutex_lock(&kvm->lock);
2339         vcore = NULL;
2340         err = -EINVAL;
2341         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
2342                 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) {
2343                         pr_devel("KVM: VCPU ID too high\n");
2344                         core = KVM_MAX_VCORES;
2345                 } else {
2346                         BUG_ON(kvm->arch.smt_mode != 1);
2347                         core = kvmppc_pack_vcpu_id(kvm, id);
2348                 }
2349         } else {
2350                 core = id / kvm->arch.smt_mode;
2351         }
2352         if (core < KVM_MAX_VCORES) {
2353                 vcore = kvm->arch.vcores[core];
2354                 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
2355                         pr_devel("KVM: collision on id %u", id);
2356                         vcore = NULL;
2357                 } else if (!vcore) {
2358                         /*
2359                          * Take mmu_setup_lock for mutual exclusion
2360                          * with kvmppc_update_lpcr().
2361                          */
2362                         err = -ENOMEM;
2363                         vcore = kvmppc_vcore_create(kvm,
2364                                         id & ~(kvm->arch.smt_mode - 1));
2365                         mutex_lock(&kvm->arch.mmu_setup_lock);
2366                         kvm->arch.vcores[core] = vcore;
2367                         kvm->arch.online_vcores++;
2368                         mutex_unlock(&kvm->arch.mmu_setup_lock);
2369                 }
2370         }
2371         mutex_unlock(&kvm->lock);
2372
2373         if (!vcore)
2374                 goto free_vcpu;
2375
2376         spin_lock(&vcore->lock);
2377         ++vcore->num_threads;
2378         spin_unlock(&vcore->lock);
2379         vcpu->arch.vcore = vcore;
2380         vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid;
2381         vcpu->arch.thread_cpu = -1;
2382         vcpu->arch.prev_cpu = -1;
2383
2384         vcpu->arch.cpu_type = KVM_CPU_3S_64;
2385         kvmppc_sanity_check(vcpu);
2386
2387         debugfs_vcpu_init(vcpu, id);
2388
2389         return vcpu;
2390
2391 free_vcpu:
2392         kmem_cache_free(kvm_vcpu_cache, vcpu);
2393 out:
2394         return ERR_PTR(err);
2395 }
2396
2397 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
2398                               unsigned long flags)
2399 {
2400         int err;
2401         int esmt = 0;
2402
2403         if (flags)
2404                 return -EINVAL;
2405         if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode))
2406                 return -EINVAL;
2407         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
2408                 /*
2409                  * On POWER8 (or POWER7), the threading mode is "strict",
2410                  * so we pack smt_mode vcpus per vcore.
2411                  */
2412                 if (smt_mode > threads_per_subcore)
2413                         return -EINVAL;
2414         } else {
2415                 /*
2416                  * On POWER9, the threading mode is "loose",
2417                  * so each vcpu gets its own vcore.
2418                  */
2419                 esmt = smt_mode;
2420                 smt_mode = 1;
2421         }
2422         mutex_lock(&kvm->lock);
2423         err = -EBUSY;
2424         if (!kvm->arch.online_vcores) {
2425                 kvm->arch.smt_mode = smt_mode;
2426                 kvm->arch.emul_smt_mode = esmt;
2427                 err = 0;
2428         }
2429         mutex_unlock(&kvm->lock);
2430
2431         return err;
2432 }
2433
2434 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa)
2435 {
2436         if (vpa->pinned_addr)
2437                 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa,
2438                                         vpa->dirty);
2439 }
2440
2441 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu)
2442 {
2443         spin_lock(&vcpu->arch.vpa_update_lock);
2444         unpin_vpa(vcpu->kvm, &vcpu->arch.dtl);
2445         unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow);
2446         unpin_vpa(vcpu->kvm, &vcpu->arch.vpa);
2447         spin_unlock(&vcpu->arch.vpa_update_lock);
2448         kvm_vcpu_uninit(vcpu);
2449         kmem_cache_free(kvm_vcpu_cache, vcpu);
2450 }
2451
2452 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu)
2453 {
2454         /* Indicate we want to get back into the guest */
2455         return 1;
2456 }
2457
2458 static void kvmppc_set_timer(struct kvm_vcpu *vcpu)
2459 {
2460         unsigned long dec_nsec, now;
2461
2462         now = get_tb();
2463         if (now > vcpu->arch.dec_expires) {
2464                 /* decrementer has already gone negative */
2465                 kvmppc_core_queue_dec(vcpu);
2466                 kvmppc_core_prepare_to_enter(vcpu);
2467                 return;
2468         }
2469         dec_nsec = tb_to_ns(vcpu->arch.dec_expires - now);
2470         hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL);
2471         vcpu->arch.timer_running = 1;
2472 }
2473
2474 extern int __kvmppc_vcore_entry(void);
2475
2476 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc,
2477                                    struct kvm_vcpu *vcpu)
2478 {
2479         u64 now;
2480
2481         if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
2482                 return;
2483         spin_lock_irq(&vcpu->arch.tbacct_lock);
2484         now = mftb();
2485         vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) -
2486                 vcpu->arch.stolen_logged;
2487         vcpu->arch.busy_preempt = now;
2488         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
2489         spin_unlock_irq(&vcpu->arch.tbacct_lock);
2490         --vc->n_runnable;
2491         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL);
2492 }
2493
2494 static int kvmppc_grab_hwthread(int cpu)
2495 {
2496         struct paca_struct *tpaca;
2497         long timeout = 10000;
2498
2499         tpaca = paca_ptrs[cpu];
2500
2501         /* Ensure the thread won't go into the kernel if it wakes */
2502         tpaca->kvm_hstate.kvm_vcpu = NULL;
2503         tpaca->kvm_hstate.kvm_vcore = NULL;
2504         tpaca->kvm_hstate.napping = 0;
2505         smp_wmb();
2506         tpaca->kvm_hstate.hwthread_req = 1;
2507
2508         /*
2509          * If the thread is already executing in the kernel (e.g. handling
2510          * a stray interrupt), wait for it to get back to nap mode.
2511          * The smp_mb() is to ensure that our setting of hwthread_req
2512          * is visible before we look at hwthread_state, so if this
2513          * races with the code at system_reset_pSeries and the thread
2514          * misses our setting of hwthread_req, we are sure to see its
2515          * setting of hwthread_state, and vice versa.
2516          */
2517         smp_mb();
2518         while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) {
2519                 if (--timeout <= 0) {
2520                         pr_err("KVM: couldn't grab cpu %d\n", cpu);
2521                         return -EBUSY;
2522                 }
2523                 udelay(1);
2524         }
2525         return 0;
2526 }
2527
2528 static void kvmppc_release_hwthread(int cpu)
2529 {
2530         struct paca_struct *tpaca;
2531
2532         tpaca = paca_ptrs[cpu];
2533         tpaca->kvm_hstate.hwthread_req = 0;
2534         tpaca->kvm_hstate.kvm_vcpu = NULL;
2535         tpaca->kvm_hstate.kvm_vcore = NULL;
2536         tpaca->kvm_hstate.kvm_split_mode = NULL;
2537 }
2538
2539 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
2540 {
2541         struct kvm_nested_guest *nested = vcpu->arch.nested;
2542         cpumask_t *cpu_in_guest;
2543         int i;
2544
2545         cpu = cpu_first_thread_sibling(cpu);
2546         if (nested) {
2547                 cpumask_set_cpu(cpu, &nested->need_tlb_flush);
2548                 cpu_in_guest = &nested->cpu_in_guest;
2549         } else {
2550                 cpumask_set_cpu(cpu, &kvm->arch.need_tlb_flush);
2551                 cpu_in_guest = &kvm->arch.cpu_in_guest;
2552         }
2553         /*
2554          * Make sure setting of bit in need_tlb_flush precedes
2555          * testing of cpu_in_guest bits.  The matching barrier on
2556          * the other side is the first smp_mb() in kvmppc_run_core().
2557          */
2558         smp_mb();
2559         for (i = 0; i < threads_per_core; ++i)
2560                 if (cpumask_test_cpu(cpu + i, cpu_in_guest))
2561                         smp_call_function_single(cpu + i, do_nothing, NULL, 1);
2562 }
2563
2564 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
2565 {
2566         struct kvm_nested_guest *nested = vcpu->arch.nested;
2567         struct kvm *kvm = vcpu->kvm;
2568         int prev_cpu;
2569
2570         if (!cpu_has_feature(CPU_FTR_HVMODE))
2571                 return;
2572
2573         if (nested)
2574                 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id];
2575         else
2576                 prev_cpu = vcpu->arch.prev_cpu;
2577
2578         /*
2579          * With radix, the guest can do TLB invalidations itself,
2580          * and it could choose to use the local form (tlbiel) if
2581          * it is invalidating a translation that has only ever been
2582          * used on one vcpu.  However, that doesn't mean it has
2583          * only ever been used on one physical cpu, since vcpus
2584          * can move around between pcpus.  To cope with this, when
2585          * a vcpu moves from one pcpu to another, we need to tell
2586          * any vcpus running on the same core as this vcpu previously
2587          * ran to flush the TLB.  The TLB is shared between threads,
2588          * so we use a single bit in .need_tlb_flush for all 4 threads.
2589          */
2590         if (prev_cpu != pcpu) {
2591                 if (prev_cpu >= 0 &&
2592                     cpu_first_thread_sibling(prev_cpu) !=
2593                     cpu_first_thread_sibling(pcpu))
2594                         radix_flush_cpu(kvm, prev_cpu, vcpu);
2595                 if (nested)
2596                         nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
2597                 else
2598                         vcpu->arch.prev_cpu = pcpu;
2599         }
2600 }
2601
2602 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
2603 {
2604         int cpu;
2605         struct paca_struct *tpaca;
2606         struct kvm *kvm = vc->kvm;
2607
2608         cpu = vc->pcpu;
2609         if (vcpu) {
2610                 if (vcpu->arch.timer_running) {
2611                         hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
2612                         vcpu->arch.timer_running = 0;
2613                 }
2614                 cpu += vcpu->arch.ptid;
2615                 vcpu->cpu = vc->pcpu;
2616                 vcpu->arch.thread_cpu = cpu;
2617                 cpumask_set_cpu(cpu, &kvm->arch.cpu_in_guest);
2618         }
2619         tpaca = paca_ptrs[cpu];
2620         tpaca->kvm_hstate.kvm_vcpu = vcpu;
2621         tpaca->kvm_hstate.ptid = cpu - vc->pcpu;
2622         tpaca->kvm_hstate.fake_suspend = 0;
2623         /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */
2624         smp_wmb();
2625         tpaca->kvm_hstate.kvm_vcore = vc;
2626         if (cpu != smp_processor_id())
2627                 kvmppc_ipi_thread(cpu);
2628 }
2629
2630 static void kvmppc_wait_for_nap(int n_threads)
2631 {
2632         int cpu = smp_processor_id();
2633         int i, loops;
2634
2635         if (n_threads <= 1)
2636                 return;
2637         for (loops = 0; loops < 1000000; ++loops) {
2638                 /*
2639                  * Check if all threads are finished.
2640                  * We set the vcore pointer when starting a thread
2641                  * and the thread clears it when finished, so we look
2642                  * for any threads that still have a non-NULL vcore ptr.
2643                  */
2644                 for (i = 1; i < n_threads; ++i)
2645                         if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2646                                 break;
2647                 if (i == n_threads) {
2648                         HMT_medium();
2649                         return;
2650                 }
2651                 HMT_low();
2652         }
2653         HMT_medium();
2654         for (i = 1; i < n_threads; ++i)
2655                 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore)
2656                         pr_err("KVM: CPU %d seems to be stuck\n", cpu + i);
2657 }
2658
2659 /*
2660  * Check that we are on thread 0 and that any other threads in
2661  * this core are off-line.  Then grab the threads so they can't
2662  * enter the kernel.
2663  */
2664 static int on_primary_thread(void)
2665 {
2666         int cpu = smp_processor_id();
2667         int thr;
2668
2669         /* Are we on a primary subcore? */
2670         if (cpu_thread_in_subcore(cpu))
2671                 return 0;
2672
2673         thr = 0;
2674         while (++thr < threads_per_subcore)
2675                 if (cpu_online(cpu + thr))
2676                         return 0;
2677
2678         /* Grab all hw threads so they can't go into the kernel */
2679         for (thr = 1; thr < threads_per_subcore; ++thr) {
2680                 if (kvmppc_grab_hwthread(cpu + thr)) {
2681                         /* Couldn't grab one; let the others go */
2682                         do {
2683                                 kvmppc_release_hwthread(cpu + thr);
2684                         } while (--thr > 0);
2685                         return 0;
2686                 }
2687         }
2688         return 1;
2689 }
2690
2691 /*
2692  * A list of virtual cores for each physical CPU.
2693  * These are vcores that could run but their runner VCPU tasks are
2694  * (or may be) preempted.
2695  */
2696 struct preempted_vcore_list {
2697         struct list_head        list;
2698         spinlock_t              lock;
2699 };
2700
2701 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores);
2702
2703 static void init_vcore_lists(void)
2704 {
2705         int cpu;
2706
2707         for_each_possible_cpu(cpu) {
2708                 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu);
2709                 spin_lock_init(&lp->lock);
2710                 INIT_LIST_HEAD(&lp->list);
2711         }
2712 }
2713
2714 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc)
2715 {
2716         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2717
2718         vc->vcore_state = VCORE_PREEMPT;
2719         vc->pcpu = smp_processor_id();
2720         if (vc->num_threads < threads_per_vcore(vc->kvm)) {
2721                 spin_lock(&lp->lock);
2722                 list_add_tail(&vc->preempt_list, &lp->list);
2723                 spin_unlock(&lp->lock);
2724         }
2725
2726         /* Start accumulating stolen time */
2727         kvmppc_core_start_stolen(vc);
2728 }
2729
2730 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc)
2731 {
2732         struct preempted_vcore_list *lp;
2733
2734         kvmppc_core_end_stolen(vc);
2735         if (!list_empty(&vc->preempt_list)) {
2736                 lp = &per_cpu(preempted_vcores, vc->pcpu);
2737                 spin_lock(&lp->lock);
2738                 list_del_init(&vc->preempt_list);
2739                 spin_unlock(&lp->lock);
2740         }
2741         vc->vcore_state = VCORE_INACTIVE;
2742 }
2743
2744 /*
2745  * This stores information about the virtual cores currently
2746  * assigned to a physical core.
2747  */
2748 struct core_info {
2749         int             n_subcores;
2750         int             max_subcore_threads;
2751         int             total_threads;
2752         int             subcore_threads[MAX_SUBCORES];
2753         struct kvmppc_vcore *vc[MAX_SUBCORES];
2754 };
2755
2756 /*
2757  * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7
2758  * respectively in 2-way micro-threading (split-core) mode on POWER8.
2759  */
2760 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 };
2761
2762 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc)
2763 {
2764         memset(cip, 0, sizeof(*cip));
2765         cip->n_subcores = 1;
2766         cip->max_subcore_threads = vc->num_threads;
2767         cip->total_threads = vc->num_threads;
2768         cip->subcore_threads[0] = vc->num_threads;
2769         cip->vc[0] = vc;
2770 }
2771
2772 static bool subcore_config_ok(int n_subcores, int n_threads)
2773 {
2774         /*
2775          * POWER9 "SMT4" cores are permanently in what is effectively a 4-way
2776          * split-core mode, with one thread per subcore.
2777          */
2778         if (cpu_has_feature(CPU_FTR_ARCH_300))
2779                 return n_subcores <= 4 && n_threads == 1;
2780
2781         /* On POWER8, can only dynamically split if unsplit to begin with */
2782         if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS)
2783                 return false;
2784         if (n_subcores > MAX_SUBCORES)
2785                 return false;
2786         if (n_subcores > 1) {
2787                 if (!(dynamic_mt_modes & 2))
2788                         n_subcores = 4;
2789                 if (n_subcores > 2 && !(dynamic_mt_modes & 4))
2790                         return false;
2791         }
2792
2793         return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS;
2794 }
2795
2796 static void init_vcore_to_run(struct kvmppc_vcore *vc)
2797 {
2798         vc->entry_exit_map = 0;
2799         vc->in_guest = 0;
2800         vc->napping_threads = 0;
2801         vc->conferring_threads = 0;
2802         vc->tb_offset_applied = 0;
2803 }
2804
2805 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip)
2806 {
2807         int n_threads = vc->num_threads;
2808         int sub;
2809
2810         if (!cpu_has_feature(CPU_FTR_ARCH_207S))
2811                 return false;
2812
2813         /* In one_vm_per_core mode, require all vcores to be from the same vm */
2814         if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm)
2815                 return false;
2816
2817         /* Some POWER9 chips require all threads to be in the same MMU mode */
2818         if (no_mixing_hpt_and_radix &&
2819             kvm_is_radix(vc->kvm) != kvm_is_radix(cip->vc[0]->kvm))
2820                 return false;
2821
2822         if (n_threads < cip->max_subcore_threads)
2823                 n_threads = cip->max_subcore_threads;
2824         if (!subcore_config_ok(cip->n_subcores + 1, n_threads))
2825                 return false;
2826         cip->max_subcore_threads = n_threads;
2827
2828         sub = cip->n_subcores;
2829         ++cip->n_subcores;
2830         cip->total_threads += vc->num_threads;
2831         cip->subcore_threads[sub] = vc->num_threads;
2832         cip->vc[sub] = vc;
2833         init_vcore_to_run(vc);
2834         list_del_init(&vc->preempt_list);
2835
2836         return true;
2837 }
2838
2839 /*
2840  * Work out whether it is possible to piggyback the execution of
2841  * vcore *pvc onto the execution of the other vcores described in *cip.
2842  */
2843 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip,
2844                           int target_threads)
2845 {
2846         if (cip->total_threads + pvc->num_threads > target_threads)
2847                 return false;
2848
2849         return can_dynamic_split(pvc, cip);
2850 }
2851
2852 static void prepare_threads(struct kvmppc_vcore *vc)
2853 {
2854         int i;
2855         struct kvm_vcpu *vcpu;
2856
2857         for_each_runnable_thread(i, vcpu, vc) {
2858                 if (signal_pending(vcpu->arch.run_task))
2859                         vcpu->arch.ret = -EINTR;
2860                 else if (vcpu->arch.vpa.update_pending ||
2861                          vcpu->arch.slb_shadow.update_pending ||
2862                          vcpu->arch.dtl.update_pending)
2863                         vcpu->arch.ret = RESUME_GUEST;
2864                 else
2865                         continue;
2866                 kvmppc_remove_runnable(vc, vcpu);
2867                 wake_up(&vcpu->arch.cpu_run);
2868         }
2869 }
2870
2871 static void collect_piggybacks(struct core_info *cip, int target_threads)
2872 {
2873         struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores);
2874         struct kvmppc_vcore *pvc, *vcnext;
2875
2876         spin_lock(&lp->lock);
2877         list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) {
2878                 if (!spin_trylock(&pvc->lock))
2879                         continue;
2880                 prepare_threads(pvc);
2881                 if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) {
2882                         list_del_init(&pvc->preempt_list);
2883                         if (pvc->runner == NULL) {
2884                                 pvc->vcore_state = VCORE_INACTIVE;
2885                                 kvmppc_core_end_stolen(pvc);
2886                         }
2887                         spin_unlock(&pvc->lock);
2888                         continue;
2889                 }
2890                 if (!can_piggyback(pvc, cip, target_threads)) {
2891                         spin_unlock(&pvc->lock);
2892                         continue;
2893                 }
2894                 kvmppc_core_end_stolen(pvc);
2895                 pvc->vcore_state = VCORE_PIGGYBACK;
2896                 if (cip->total_threads >= target_threads)
2897                         break;
2898         }
2899         spin_unlock(&lp->lock);
2900 }
2901
2902 static bool recheck_signals_and_mmu(struct core_info *cip)
2903 {
2904         int sub, i;
2905         struct kvm_vcpu *vcpu;
2906         struct kvmppc_vcore *vc;
2907
2908         for (sub = 0; sub < cip->n_subcores; ++sub) {
2909                 vc = cip->vc[sub];
2910                 if (!vc->kvm->arch.mmu_ready)
2911                         return true;
2912                 for_each_runnable_thread(i, vcpu, vc)
2913                         if (signal_pending(vcpu->arch.run_task))
2914                                 return true;
2915         }
2916         return false;
2917 }
2918
2919 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master)
2920 {
2921         int still_running = 0, i;
2922         u64 now;
2923         long ret;
2924         struct kvm_vcpu *vcpu;
2925
2926         spin_lock(&vc->lock);
2927         now = get_tb();
2928         for_each_runnable_thread(i, vcpu, vc) {
2929                 /*
2930                  * It's safe to unlock the vcore in the loop here, because
2931                  * for_each_runnable_thread() is safe against removal of
2932                  * the vcpu, and the vcore state is VCORE_EXITING here,
2933                  * so any vcpus becoming runnable will have their arch.trap
2934                  * set to zero and can't actually run in the guest.
2935                  */
2936                 spin_unlock(&vc->lock);
2937                 /* cancel pending dec exception if dec is positive */
2938                 if (now < vcpu->arch.dec_expires &&
2939                     kvmppc_core_pending_dec(vcpu))
2940                         kvmppc_core_dequeue_dec(vcpu);
2941
2942                 trace_kvm_guest_exit(vcpu);
2943
2944                 ret = RESUME_GUEST;
2945                 if (vcpu->arch.trap)
2946                         ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu,
2947                                                     vcpu->arch.run_task);
2948
2949                 vcpu->arch.ret = ret;
2950                 vcpu->arch.trap = 0;
2951
2952                 spin_lock(&vc->lock);
2953                 if (is_kvmppc_resume_guest(vcpu->arch.ret)) {
2954                         if (vcpu->arch.pending_exceptions)
2955                                 kvmppc_core_prepare_to_enter(vcpu);
2956                         if (vcpu->arch.ceded)
2957                                 kvmppc_set_timer(vcpu);
2958                         else
2959                                 ++still_running;
2960                 } else {
2961                         kvmppc_remove_runnable(vc, vcpu);
2962                         wake_up(&vcpu->arch.cpu_run);
2963                 }
2964         }
2965         if (!is_master) {
2966                 if (still_running > 0) {
2967                         kvmppc_vcore_preempt(vc);
2968                 } else if (vc->runner) {
2969                         vc->vcore_state = VCORE_PREEMPT;
2970                         kvmppc_core_start_stolen(vc);
2971                 } else {
2972                         vc->vcore_state = VCORE_INACTIVE;
2973                 }
2974                 if (vc->n_runnable > 0 && vc->runner == NULL) {
2975                         /* make sure there's a candidate runner awake */
2976                         i = -1;
2977                         vcpu = next_runnable_thread(vc, &i);
2978                         wake_up(&vcpu->arch.cpu_run);
2979                 }
2980         }
2981         spin_unlock(&vc->lock);
2982 }
2983
2984 /*
2985  * Clear core from the list of active host cores as we are about to
2986  * enter the guest. Only do this if it is the primary thread of the
2987  * core (not if a subcore) that is entering the guest.
2988  */
2989 static inline int kvmppc_clear_host_core(unsigned int cpu)
2990 {
2991         int core;
2992
2993         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
2994                 return 0;
2995         /*
2996          * Memory barrier can be omitted here as we will do a smp_wmb()
2997          * later in kvmppc_start_thread and we need ensure that state is
2998          * visible to other CPUs only after we enter guest.
2999          */
3000         core = cpu >> threads_shift;
3001         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0;
3002         return 0;
3003 }
3004
3005 /*
3006  * Advertise this core as an active host core since we exited the guest
3007  * Only need to do this if it is the primary thread of the core that is
3008  * exiting.
3009  */
3010 static inline int kvmppc_set_host_core(unsigned int cpu)
3011 {
3012         int core;
3013
3014         if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu))
3015                 return 0;
3016
3017         /*
3018          * Memory barrier can be omitted here because we do a spin_unlock
3019          * immediately after this which provides the memory barrier.
3020          */
3021         core = cpu >> threads_shift;
3022         kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1;
3023         return 0;
3024 }
3025
3026 static void set_irq_happened(int trap)
3027 {
3028         switch (trap) {
3029         case BOOK3S_INTERRUPT_EXTERNAL:
3030                 local_paca->irq_happened |= PACA_IRQ_EE;
3031                 break;
3032         case BOOK3S_INTERRUPT_H_DOORBELL:
3033                 local_paca->irq_happened |= PACA_IRQ_DBELL;
3034                 break;
3035         case BOOK3S_INTERRUPT_HMI:
3036                 local_paca->irq_happened |= PACA_IRQ_HMI;
3037                 break;
3038         case BOOK3S_INTERRUPT_SYSTEM_RESET:
3039                 replay_system_reset();
3040                 break;
3041         }
3042 }
3043
3044 /*
3045  * Run a set of guest threads on a physical core.
3046  * Called with vc->lock held.
3047  */
3048 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
3049 {
3050         struct kvm_vcpu *vcpu;
3051         int i;
3052         int srcu_idx;
3053         struct core_info core_info;
3054         struct kvmppc_vcore *pvc;
3055         struct kvm_split_mode split_info, *sip;
3056         int split, subcore_size, active;
3057         int sub;
3058         bool thr0_done;
3059         unsigned long cmd_bit, stat_bit;
3060         int pcpu, thr;
3061         int target_threads;
3062         int controlled_threads;
3063         int trap;
3064         bool is_power8;
3065         bool hpt_on_radix;
3066
3067         /*
3068          * Remove from the list any threads that have a signal pending
3069          * or need a VPA update done
3070          */
3071         prepare_threads(vc);
3072
3073         /* if the runner is no longer runnable, let the caller pick a new one */
3074         if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE)
3075                 return;
3076
3077         /*
3078          * Initialize *vc.
3079          */
3080         init_vcore_to_run(vc);
3081         vc->preempt_tb = TB_NIL;
3082
3083         /*
3084          * Number of threads that we will be controlling: the same as
3085          * the number of threads per subcore, except on POWER9,
3086          * where it's 1 because the threads are (mostly) independent.
3087          */
3088         controlled_threads = threads_per_vcore(vc->kvm);
3089
3090         /*
3091          * Make sure we are running on primary threads, and that secondary
3092          * threads are offline.  Also check if the number of threads in this
3093          * guest are greater than the current system threads per guest.
3094          * On POWER9, we need to be not in independent-threads mode if
3095          * this is a HPT guest on a radix host machine where the
3096          * CPU threads may not be in different MMU modes.
3097          */
3098         hpt_on_radix = no_mixing_hpt_and_radix && radix_enabled() &&
3099                 !kvm_is_radix(vc->kvm);
3100         if (((controlled_threads > 1) &&
3101              ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) ||
3102             (hpt_on_radix && vc->kvm->arch.threads_indep)) {
3103                 for_each_runnable_thread(i, vcpu, vc) {
3104                         vcpu->arch.ret = -EBUSY;
3105                         kvmppc_remove_runnable(vc, vcpu);
3106                         wake_up(&vcpu->arch.cpu_run);
3107                 }
3108                 goto out;
3109         }
3110
3111         /*
3112          * See if we could run any other vcores on the physical core
3113          * along with this one.
3114          */
3115         init_core_info(&core_info, vc);
3116         pcpu = smp_processor_id();
3117         target_threads = controlled_threads;
3118         if (target_smt_mode && target_smt_mode < target_threads)
3119                 target_threads = target_smt_mode;
3120         if (vc->num_threads < target_threads)
3121                 collect_piggybacks(&core_info, target_threads);
3122
3123         /*
3124          * On radix, arrange for TLB flushing if necessary.
3125          * This has to be done before disabling interrupts since
3126          * it uses smp_call_function().
3127          */
3128         pcpu = smp_processor_id();
3129         if (kvm_is_radix(vc->kvm)) {
3130                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3131                         for_each_runnable_thread(i, vcpu, core_info.vc[sub])
3132                                 kvmppc_prepare_radix_vcpu(vcpu, pcpu);
3133         }
3134
3135         /*
3136          * Hard-disable interrupts, and check resched flag and signals.
3137          * If we need to reschedule or deliver a signal, clean up
3138          * and return without going into the guest(s).
3139          * If the mmu_ready flag has been cleared, don't go into the
3140          * guest because that means a HPT resize operation is in progress.
3141          */
3142         local_irq_disable();
3143         hard_irq_disable();
3144         if (lazy_irq_pending() || need_resched() ||
3145             recheck_signals_and_mmu(&core_info)) {
3146                 local_irq_enable();
3147                 vc->vcore_state = VCORE_INACTIVE;
3148                 /* Unlock all except the primary vcore */
3149                 for (sub = 1; sub < core_info.n_subcores; ++sub) {
3150                         pvc = core_info.vc[sub];
3151                         /* Put back on to the preempted vcores list */
3152                         kvmppc_vcore_preempt(pvc);
3153                         spin_unlock(&pvc->lock);
3154                 }
3155                 for (i = 0; i < controlled_threads; ++i)
3156                         kvmppc_release_hwthread(pcpu + i);
3157                 return;
3158         }
3159
3160         kvmppc_clear_host_core(pcpu);
3161
3162         /* Decide on micro-threading (split-core) mode */
3163         subcore_size = threads_per_subcore;
3164         cmd_bit = stat_bit = 0;
3165         split = core_info.n_subcores;
3166         sip = NULL;
3167         is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S)
3168                 && !cpu_has_feature(CPU_FTR_ARCH_300);
3169
3170         if (split > 1 || hpt_on_radix) {
3171                 sip = &split_info;
3172                 memset(&split_info, 0, sizeof(split_info));
3173                 for (sub = 0; sub < core_info.n_subcores; ++sub)
3174                         split_info.vc[sub] = core_info.vc[sub];
3175
3176                 if (is_power8) {
3177                         if (split == 2 && (dynamic_mt_modes & 2)) {
3178                                 cmd_bit = HID0_POWER8_1TO2LPAR;
3179                                 stat_bit = HID0_POWER8_2LPARMODE;
3180                         } else {
3181                                 split = 4;
3182                                 cmd_bit = HID0_POWER8_1TO4LPAR;
3183                                 stat_bit = HID0_POWER8_4LPARMODE;
3184                         }
3185                         subcore_size = MAX_SMT_THREADS / split;
3186                         split_info.rpr = mfspr(SPRN_RPR);
3187                         split_info.pmmar = mfspr(SPRN_PMMAR);
3188                         split_info.ldbar = mfspr(SPRN_LDBAR);
3189                         split_info.subcore_size = subcore_size;
3190                 } else {
3191                         split_info.subcore_size = 1;
3192                         if (hpt_on_radix) {
3193                                 /* Use the split_info for LPCR/LPIDR changes */
3194                                 split_info.lpcr_req = vc->lpcr;
3195                                 split_info.lpidr_req = vc->kvm->arch.lpid;
3196                                 split_info.host_lpcr = vc->kvm->arch.host_lpcr;
3197                                 split_info.do_set = 1;
3198                         }
3199                 }
3200
3201                 /* order writes to split_info before kvm_split_mode pointer */
3202                 smp_wmb();
3203         }
3204
3205         for (thr = 0; thr < controlled_threads; ++thr) {
3206                 struct paca_struct *paca = paca_ptrs[pcpu + thr];
3207
3208                 paca->kvm_hstate.tid = thr;
3209                 paca->kvm_hstate.napping = 0;
3210                 paca->kvm_hstate.kvm_split_mode = sip;
3211         }
3212
3213         /* Initiate micro-threading (split-core) on POWER8 if required */
3214         if (cmd_bit) {
3215                 unsigned long hid0 = mfspr(SPRN_HID0);
3216
3217                 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS;
3218                 mb();
3219                 mtspr(SPRN_HID0, hid0);
3220                 isync();
3221                 for (;;) {
3222                         hid0 = mfspr(SPRN_HID0);
3223                         if (hid0 & stat_bit)
3224                                 break;
3225                         cpu_relax();
3226                 }
3227         }
3228
3229         /*
3230          * On POWER8, set RWMR register.
3231          * Since it only affects PURR and SPURR, it doesn't affect
3232          * the host, so we don't save/restore the host value.
3233          */
3234         if (is_power8) {
3235                 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD;
3236                 int n_online = atomic_read(&vc->online_count);
3237
3238                 /*
3239                  * Use the 8-thread value if we're doing split-core
3240                  * or if the vcore's online count looks bogus.
3241                  */
3242                 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS &&
3243                     n_online >= 1 && n_online <= MAX_SMT_THREADS)
3244                         rwmr_val = p8_rwmr_values[n_online];
3245                 mtspr(SPRN_RWMR, rwmr_val);
3246         }
3247
3248         /* Start all the threads */
3249         active = 0;
3250         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3251                 thr = is_power8 ? subcore_thread_map[sub] : sub;
3252                 thr0_done = false;
3253                 active |= 1 << thr;
3254                 pvc = core_info.vc[sub];
3255                 pvc->pcpu = pcpu + thr;
3256                 for_each_runnable_thread(i, vcpu, pvc) {
3257                         kvmppc_start_thread(vcpu, pvc);
3258                         kvmppc_create_dtl_entry(vcpu, pvc);
3259                         trace_kvm_guest_enter(vcpu);
3260                         if (!vcpu->arch.ptid)
3261                                 thr0_done = true;
3262                         active |= 1 << (thr + vcpu->arch.ptid);
3263                 }
3264                 /*
3265                  * We need to start the first thread of each subcore
3266                  * even if it doesn't have a vcpu.
3267                  */
3268                 if (!thr0_done)
3269                         kvmppc_start_thread(NULL, pvc);
3270         }
3271
3272         /*
3273          * Ensure that split_info.do_nap is set after setting
3274          * the vcore pointer in the PACA of the secondaries.
3275          */
3276         smp_mb();
3277
3278         /*
3279          * When doing micro-threading, poke the inactive threads as well.
3280          * This gets them to the nap instruction after kvm_do_nap,
3281          * which reduces the time taken to unsplit later.
3282          * For POWER9 HPT guest on radix host, we need all the secondary
3283          * threads woken up so they can do the LPCR/LPIDR change.
3284          */
3285         if (cmd_bit || hpt_on_radix) {
3286                 split_info.do_nap = 1;  /* ask secondaries to nap when done */
3287                 for (thr = 1; thr < threads_per_subcore; ++thr)
3288                         if (!(active & (1 << thr)))
3289                                 kvmppc_ipi_thread(pcpu + thr);
3290         }
3291
3292         vc->vcore_state = VCORE_RUNNING;
3293         preempt_disable();
3294
3295         trace_kvmppc_run_core(vc, 0);
3296
3297         for (sub = 0; sub < core_info.n_subcores; ++sub)
3298                 spin_unlock(&core_info.vc[sub]->lock);
3299
3300         guest_enter_irqoff();
3301
3302         srcu_idx = srcu_read_lock(&vc->kvm->srcu);
3303
3304         this_cpu_disable_ftrace();
3305
3306         /*
3307          * Interrupts will be enabled once we get into the guest,
3308          * so tell lockdep that we're about to enable interrupts.
3309          */
3310         trace_hardirqs_on();
3311
3312         trap = __kvmppc_vcore_entry();
3313
3314         trace_hardirqs_off();
3315
3316         this_cpu_enable_ftrace();
3317
3318         srcu_read_unlock(&vc->kvm->srcu, srcu_idx);
3319
3320         set_irq_happened(trap);
3321
3322         spin_lock(&vc->lock);
3323         /* prevent other vcpu threads from doing kvmppc_start_thread() now */
3324         vc->vcore_state = VCORE_EXITING;
3325
3326         /* wait for secondary threads to finish writing their state to memory */
3327         kvmppc_wait_for_nap(controlled_threads);
3328
3329         /* Return to whole-core mode if we split the core earlier */
3330         if (cmd_bit) {
3331                 unsigned long hid0 = mfspr(SPRN_HID0);
3332                 unsigned long loops = 0;
3333
3334                 hid0 &= ~HID0_POWER8_DYNLPARDIS;
3335                 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE;
3336                 mb();
3337                 mtspr(SPRN_HID0, hid0);
3338                 isync();
3339                 for (;;) {
3340                         hid0 = mfspr(SPRN_HID0);
3341                         if (!(hid0 & stat_bit))
3342                                 break;
3343                         cpu_relax();
3344                         ++loops;
3345                 }
3346         } else if (hpt_on_radix) {
3347                 /* Wait for all threads to have seen final sync */
3348                 for (thr = 1; thr < controlled_threads; ++thr) {
3349                         struct paca_struct *paca = paca_ptrs[pcpu + thr];
3350
3351                         while (paca->kvm_hstate.kvm_split_mode) {
3352                                 HMT_low();
3353                                 barrier();
3354                         }
3355                         HMT_medium();
3356                 }
3357         }
3358         split_info.do_nap = 0;
3359
3360         kvmppc_set_host_core(pcpu);
3361
3362         local_irq_enable();
3363         guest_exit();
3364
3365         /* Let secondaries go back to the offline loop */
3366         for (i = 0; i < controlled_threads; ++i) {
3367                 kvmppc_release_hwthread(pcpu + i);
3368                 if (sip && sip->napped[i])
3369                         kvmppc_ipi_thread(pcpu + i);
3370                 cpumask_clear_cpu(pcpu + i, &vc->kvm->arch.cpu_in_guest);
3371         }
3372
3373         spin_unlock(&vc->lock);
3374
3375         /* make sure updates to secondary vcpu structs are visible now */
3376         smp_mb();
3377
3378         preempt_enable();
3379
3380         for (sub = 0; sub < core_info.n_subcores; ++sub) {
3381                 pvc = core_info.vc[sub];
3382                 post_guest_process(pvc, pvc == vc);
3383         }
3384
3385         spin_lock(&vc->lock);
3386
3387  out:
3388         vc->vcore_state = VCORE_INACTIVE;
3389         trace_kvmppc_run_core(vc, 1);
3390 }
3391
3392 /*
3393  * Load up hypervisor-mode registers on P9.
3394  */
3395 static int kvmhv_load_hv_regs_and_go(struct kvm_vcpu *vcpu, u64 time_limit,
3396                                      unsigned long lpcr)
3397 {
3398         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3399         s64 hdec;
3400         u64 tb, purr, spurr;
3401         int trap;
3402         unsigned long host_hfscr = mfspr(SPRN_HFSCR);
3403         unsigned long host_ciabr = mfspr(SPRN_CIABR);
3404         unsigned long host_dawr = mfspr(SPRN_DAWR);
3405         unsigned long host_dawrx = mfspr(SPRN_DAWRX);
3406         unsigned long host_psscr = mfspr(SPRN_PSSCR);
3407         unsigned long host_pidr = mfspr(SPRN_PID);
3408
3409         hdec = time_limit - mftb();
3410         if (hdec < 0)
3411                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3412         mtspr(SPRN_HDEC, hdec);
3413
3414         if (vc->tb_offset) {
3415                 u64 new_tb = mftb() + vc->tb_offset;
3416                 mtspr(SPRN_TBU40, new_tb);
3417                 tb = mftb();
3418                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3419                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3420                 vc->tb_offset_applied = vc->tb_offset;
3421         }
3422
3423         if (vc->pcr)
3424                 mtspr(SPRN_PCR, vc->pcr | PCR_MASK);
3425         mtspr(SPRN_DPDES, vc->dpdes);
3426         mtspr(SPRN_VTB, vc->vtb);
3427
3428         local_paca->kvm_hstate.host_purr = mfspr(SPRN_PURR);
3429         local_paca->kvm_hstate.host_spurr = mfspr(SPRN_SPURR);
3430         mtspr(SPRN_PURR, vcpu->arch.purr);
3431         mtspr(SPRN_SPURR, vcpu->arch.spurr);
3432
3433         if (dawr_enabled()) {
3434                 mtspr(SPRN_DAWR, vcpu->arch.dawr);
3435                 mtspr(SPRN_DAWRX, vcpu->arch.dawrx);
3436         }
3437         mtspr(SPRN_CIABR, vcpu->arch.ciabr);
3438         mtspr(SPRN_IC, vcpu->arch.ic);
3439         mtspr(SPRN_PID, vcpu->arch.pid);
3440
3441         mtspr(SPRN_PSSCR, vcpu->arch.psscr | PSSCR_EC |
3442               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3443
3444         mtspr(SPRN_HFSCR, vcpu->arch.hfscr);
3445
3446         mtspr(SPRN_SPRG0, vcpu->arch.shregs.sprg0);
3447         mtspr(SPRN_SPRG1, vcpu->arch.shregs.sprg1);
3448         mtspr(SPRN_SPRG2, vcpu->arch.shregs.sprg2);
3449         mtspr(SPRN_SPRG3, vcpu->arch.shregs.sprg3);
3450
3451         mtspr(SPRN_AMOR, ~0UL);
3452
3453         mtspr(SPRN_LPCR, lpcr);
3454         isync();
3455
3456         kvmppc_xive_push_vcpu(vcpu);
3457
3458         mtspr(SPRN_SRR0, vcpu->arch.shregs.srr0);
3459         mtspr(SPRN_SRR1, vcpu->arch.shregs.srr1);
3460
3461         trap = __kvmhv_vcpu_entry_p9(vcpu);
3462
3463         /* Advance host PURR/SPURR by the amount used by guest */
3464         purr = mfspr(SPRN_PURR);
3465         spurr = mfspr(SPRN_SPURR);
3466         mtspr(SPRN_PURR, local_paca->kvm_hstate.host_purr +
3467               purr - vcpu->arch.purr);
3468         mtspr(SPRN_SPURR, local_paca->kvm_hstate.host_spurr +
3469               spurr - vcpu->arch.spurr);
3470         vcpu->arch.purr = purr;
3471         vcpu->arch.spurr = spurr;
3472
3473         vcpu->arch.ic = mfspr(SPRN_IC);
3474         vcpu->arch.pid = mfspr(SPRN_PID);
3475         vcpu->arch.psscr = mfspr(SPRN_PSSCR) & PSSCR_GUEST_VIS;
3476
3477         vcpu->arch.shregs.sprg0 = mfspr(SPRN_SPRG0);
3478         vcpu->arch.shregs.sprg1 = mfspr(SPRN_SPRG1);
3479         vcpu->arch.shregs.sprg2 = mfspr(SPRN_SPRG2);
3480         vcpu->arch.shregs.sprg3 = mfspr(SPRN_SPRG3);
3481
3482         /* Preserve PSSCR[FAKE_SUSPEND] until we've called kvmppc_save_tm_hv */
3483         mtspr(SPRN_PSSCR, host_psscr |
3484               (local_paca->kvm_hstate.fake_suspend << PSSCR_FAKE_SUSPEND_LG));
3485         mtspr(SPRN_HFSCR, host_hfscr);
3486         mtspr(SPRN_CIABR, host_ciabr);
3487         mtspr(SPRN_DAWR, host_dawr);
3488         mtspr(SPRN_DAWRX, host_dawrx);
3489         mtspr(SPRN_PID, host_pidr);
3490
3491         /*
3492          * Since this is radix, do a eieio; tlbsync; ptesync sequence in
3493          * case we interrupted the guest between a tlbie and a ptesync.
3494          */
3495         asm volatile("eieio; tlbsync; ptesync");
3496
3497         mtspr(SPRN_LPID, vcpu->kvm->arch.host_lpid);    /* restore host LPID */
3498         isync();
3499
3500         vc->dpdes = mfspr(SPRN_DPDES);
3501         vc->vtb = mfspr(SPRN_VTB);
3502         mtspr(SPRN_DPDES, 0);
3503         if (vc->pcr)
3504                 mtspr(SPRN_PCR, PCR_MASK);
3505
3506         if (vc->tb_offset_applied) {
3507                 u64 new_tb = mftb() - vc->tb_offset_applied;
3508                 mtspr(SPRN_TBU40, new_tb);
3509                 tb = mftb();
3510                 if ((tb & 0xffffff) < (new_tb & 0xffffff))
3511                         mtspr(SPRN_TBU40, new_tb + 0x1000000);
3512                 vc->tb_offset_applied = 0;
3513         }
3514
3515         mtspr(SPRN_HDEC, 0x7fffffff);
3516         mtspr(SPRN_LPCR, vcpu->kvm->arch.host_lpcr);
3517
3518         return trap;
3519 }
3520
3521 /*
3522  * Virtual-mode guest entry for POWER9 and later when the host and
3523  * guest are both using the radix MMU.  The LPIDR has already been set.
3524  */
3525 int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
3526                          unsigned long lpcr)
3527 {
3528         struct kvmppc_vcore *vc = vcpu->arch.vcore;
3529         unsigned long host_dscr = mfspr(SPRN_DSCR);
3530         unsigned long host_tidr = mfspr(SPRN_TIDR);
3531         unsigned long host_iamr = mfspr(SPRN_IAMR);
3532         unsigned long host_amr = mfspr(SPRN_AMR);
3533         s64 dec;
3534         u64 tb;
3535         int trap, save_pmu;
3536
3537         dec = mfspr(SPRN_DEC);
3538         tb = mftb();
3539         if (dec < 512)
3540                 return BOOK3S_INTERRUPT_HV_DECREMENTER;
3541         local_paca->kvm_hstate.dec_expires = dec + tb;
3542         if (local_paca->kvm_hstate.dec_expires < time_limit)
3543                 time_limit = local_paca->kvm_hstate.dec_expires;
3544
3545         vcpu->arch.ceded = 0;
3546
3547         kvmhv_save_host_pmu();          /* saves it to PACA kvm_hstate */
3548
3549         kvmppc_subcore_enter_guest();
3550
3551         vc->entry_exit_map = 1;
3552         vc->in_guest = 1;
3553
3554         if (vcpu->arch.vpa.pinned_addr) {
3555                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3556                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3557                 lp->yield_count = cpu_to_be32(yield_count);
3558                 vcpu->arch.vpa.dirty = 1;
3559         }
3560
3561         if (cpu_has_feature(CPU_FTR_TM) ||
3562             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3563                 kvmppc_restore_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3564
3565         kvmhv_load_guest_pmu(vcpu);
3566
3567         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3568         load_fp_state(&vcpu->arch.fp);
3569 #ifdef CONFIG_ALTIVEC
3570         load_vr_state(&vcpu->arch.vr);
3571 #endif
3572         mtspr(SPRN_VRSAVE, vcpu->arch.vrsave);
3573
3574         mtspr(SPRN_DSCR, vcpu->arch.dscr);
3575         mtspr(SPRN_IAMR, vcpu->arch.iamr);
3576         mtspr(SPRN_PSPB, vcpu->arch.pspb);
3577         mtspr(SPRN_FSCR, vcpu->arch.fscr);
3578         mtspr(SPRN_TAR, vcpu->arch.tar);
3579         mtspr(SPRN_EBBHR, vcpu->arch.ebbhr);
3580         mtspr(SPRN_EBBRR, vcpu->arch.ebbrr);
3581         mtspr(SPRN_BESCR, vcpu->arch.bescr);
3582         mtspr(SPRN_WORT, vcpu->arch.wort);
3583         mtspr(SPRN_TIDR, vcpu->arch.tid);
3584         mtspr(SPRN_DAR, vcpu->arch.shregs.dar);
3585         mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr);
3586         mtspr(SPRN_AMR, vcpu->arch.amr);
3587         mtspr(SPRN_UAMOR, vcpu->arch.uamor);
3588
3589         if (!(vcpu->arch.ctrl & 1))
3590                 mtspr(SPRN_CTRLT, mfspr(SPRN_CTRLF) & ~1);
3591
3592         mtspr(SPRN_DEC, vcpu->arch.dec_expires - mftb());
3593
3594         if (kvmhv_on_pseries()) {
3595                 /*
3596                  * We need to save and restore the guest visible part of the
3597                  * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor
3598                  * doesn't do this for us. Note only required if pseries since
3599                  * this is done in kvmhv_load_hv_regs_and_go() below otherwise.
3600                  */
3601                 unsigned long host_psscr;
3602                 /* call our hypervisor to load up HV regs and go */
3603                 struct hv_guest_state hvregs;
3604
3605                 host_psscr = mfspr(SPRN_PSSCR_PR);
3606                 mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr);
3607                 kvmhv_save_hv_regs(vcpu, &hvregs);
3608                 hvregs.lpcr = lpcr;
3609                 vcpu->arch.regs.msr = vcpu->arch.shregs.msr;
3610                 hvregs.version = HV_GUEST_STATE_VERSION;
3611                 if (vcpu->arch.nested) {
3612                         hvregs.lpid = vcpu->arch.nested->shadow_lpid;
3613                         hvregs.vcpu_token = vcpu->arch.nested_vcpu_id;
3614                 } else {
3615                         hvregs.lpid = vcpu->kvm->arch.lpid;
3616                         hvregs.vcpu_token = vcpu->vcpu_id;
3617                 }
3618                 hvregs.hdec_expiry = time_limit;
3619                 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs),
3620                                           __pa(&vcpu->arch.regs));
3621                 kvmhv_restore_hv_return_state(vcpu, &hvregs);
3622                 vcpu->arch.shregs.msr = vcpu->arch.regs.msr;
3623                 vcpu->arch.shregs.dar = mfspr(SPRN_DAR);
3624                 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR);
3625                 vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR);
3626                 mtspr(SPRN_PSSCR_PR, host_psscr);
3627
3628                 /* H_CEDE has to be handled now, not later */
3629                 if (trap == BOOK3S_INTERRUPT_SYSCALL && !vcpu->arch.nested &&
3630                     kvmppc_get_gpr(vcpu, 3) == H_CEDE) {
3631                         kvmppc_nested_cede(vcpu);
3632                         trap = 0;
3633                 }
3634         } else {
3635                 trap = kvmhv_load_hv_regs_and_go(vcpu, time_limit, lpcr);
3636         }
3637
3638         vcpu->arch.slb_max = 0;
3639         dec = mfspr(SPRN_DEC);
3640         if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */
3641                 dec = (s32) dec;
3642         tb = mftb();
3643         vcpu->arch.dec_expires = dec + tb;
3644         vcpu->cpu = -1;
3645         vcpu->arch.thread_cpu = -1;
3646         vcpu->arch.ctrl = mfspr(SPRN_CTRLF);
3647
3648         vcpu->arch.iamr = mfspr(SPRN_IAMR);
3649         vcpu->arch.pspb = mfspr(SPRN_PSPB);
3650         vcpu->arch.fscr = mfspr(SPRN_FSCR);
3651         vcpu->arch.tar = mfspr(SPRN_TAR);
3652         vcpu->arch.ebbhr = mfspr(SPRN_EBBHR);
3653         vcpu->arch.ebbrr = mfspr(SPRN_EBBRR);
3654         vcpu->arch.bescr = mfspr(SPRN_BESCR);
3655         vcpu->arch.wort = mfspr(SPRN_WORT);
3656         vcpu->arch.tid = mfspr(SPRN_TIDR);
3657         vcpu->arch.amr = mfspr(SPRN_AMR);
3658         vcpu->arch.uamor = mfspr(SPRN_UAMOR);
3659         vcpu->arch.dscr = mfspr(SPRN_DSCR);
3660
3661         mtspr(SPRN_PSPB, 0);
3662         mtspr(SPRN_WORT, 0);
3663         mtspr(SPRN_UAMOR, 0);
3664         mtspr(SPRN_DSCR, host_dscr);
3665         mtspr(SPRN_TIDR, host_tidr);
3666         mtspr(SPRN_IAMR, host_iamr);
3667         mtspr(SPRN_PSPB, 0);
3668
3669         if (host_amr != vcpu->arch.amr)
3670                 mtspr(SPRN_AMR, host_amr);
3671
3672         msr_check_and_set(MSR_FP | MSR_VEC | MSR_VSX);
3673         store_fp_state(&vcpu->arch.fp);
3674 #ifdef CONFIG_ALTIVEC
3675         store_vr_state(&vcpu->arch.vr);
3676 #endif
3677         vcpu->arch.vrsave = mfspr(SPRN_VRSAVE);
3678
3679         if (cpu_has_feature(CPU_FTR_TM) ||
3680             cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST))
3681                 kvmppc_save_tm_hv(vcpu, vcpu->arch.shregs.msr, true);
3682
3683         save_pmu = 1;
3684         if (vcpu->arch.vpa.pinned_addr) {
3685                 struct lppaca *lp = vcpu->arch.vpa.pinned_addr;
3686                 u32 yield_count = be32_to_cpu(lp->yield_count) + 1;
3687                 lp->yield_count = cpu_to_be32(yield_count);
3688                 vcpu->arch.vpa.dirty = 1;
3689                 save_pmu = lp->pmcregs_in_use;
3690         }
3691         /* Must save pmu if this guest is capable of running nested guests */
3692         save_pmu |= nesting_enabled(vcpu->kvm);
3693
3694         kvmhv_save_guest_pmu(vcpu, save_pmu);
3695
3696         vc->entry_exit_map = 0x101;
3697         vc->in_guest = 0;
3698
3699         mtspr(SPRN_DEC, local_paca->kvm_hstate.dec_expires - mftb());
3700         mtspr(SPRN_SPRG_VDSO_WRITE, local_paca->sprg_vdso);
3701
3702         kvmhv_load_host_pmu();
3703
3704         kvmppc_subcore_exit_guest();
3705
3706         return trap;
3707 }
3708
3709 /*
3710  * Wait for some other vcpu thread to execute us, and
3711  * wake us up when we need to handle something in the host.
3712  */
3713 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc,
3714                                  struct kvm_vcpu *vcpu, int wait_state)
3715 {
3716         DEFINE_WAIT(wait);
3717
3718         prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state);
3719         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
3720                 spin_unlock(&vc->lock);
3721                 schedule();
3722                 spin_lock(&vc->lock);
3723         }
3724         finish_wait(&vcpu->arch.cpu_run, &wait);
3725 }
3726
3727 static void grow_halt_poll_ns(struct kvmppc_vcore *vc)
3728 {
3729         if (!halt_poll_ns_grow)
3730                 return;
3731
3732         vc->halt_poll_ns *= halt_poll_ns_grow;
3733         if (vc->halt_poll_ns < halt_poll_ns_grow_start)
3734                 vc->halt_poll_ns = halt_poll_ns_grow_start;
3735 }
3736
3737 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc)
3738 {
3739         if (halt_poll_ns_shrink == 0)
3740                 vc->halt_poll_ns = 0;
3741         else
3742                 vc->halt_poll_ns /= halt_poll_ns_shrink;
3743 }
3744
3745 #ifdef CONFIG_KVM_XICS
3746 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3747 {
3748         if (!xics_on_xive())
3749                 return false;
3750         return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr <
3751                 vcpu->arch.xive_saved_state.cppr;
3752 }
3753 #else
3754 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu)
3755 {
3756         return false;
3757 }
3758 #endif /* CONFIG_KVM_XICS */
3759
3760 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu)
3761 {
3762         if (vcpu->arch.pending_exceptions || vcpu->arch.prodded ||
3763             kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu))
3764                 return true;
3765
3766         return false;
3767 }
3768
3769 /*
3770  * Check to see if any of the runnable vcpus on the vcore have pending
3771  * exceptions or are no longer ceded
3772  */
3773 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc)
3774 {
3775         struct kvm_vcpu *vcpu;
3776         int i;
3777
3778         for_each_runnable_thread(i, vcpu, vc) {
3779                 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu))
3780                         return 1;
3781         }
3782
3783         return 0;
3784 }
3785
3786 /*
3787  * All the vcpus in this vcore are idle, so wait for a decrementer
3788  * or external interrupt to one of the vcpus.  vc->lock is held.
3789  */
3790 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc)
3791 {
3792         ktime_t cur, start_poll, start_wait;
3793         int do_sleep = 1;
3794         u64 block_ns;
3795         DECLARE_SWAITQUEUE(wait);
3796
3797         /* Poll for pending exceptions and ceded state */
3798         cur = start_poll = ktime_get();
3799         if (vc->halt_poll_ns) {
3800                 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns);
3801                 ++vc->runner->stat.halt_attempted_poll;
3802
3803                 vc->vcore_state = VCORE_POLLING;
3804                 spin_unlock(&vc->lock);
3805
3806                 do {
3807                         if (kvmppc_vcore_check_block(vc)) {
3808                                 do_sleep = 0;
3809                                 break;
3810                         }
3811                         cur = ktime_get();
3812                 } while (single_task_running() && ktime_before(cur, stop));
3813
3814                 spin_lock(&vc->lock);
3815                 vc->vcore_state = VCORE_INACTIVE;
3816
3817                 if (!do_sleep) {
3818                         ++vc->runner->stat.halt_successful_poll;
3819                         goto out;
3820                 }
3821         }
3822
3823         prepare_to_swait_exclusive(&vc->wq, &wait, TASK_INTERRUPTIBLE);
3824
3825         if (kvmppc_vcore_check_block(vc)) {
3826                 finish_swait(&vc->wq, &wait);
3827                 do_sleep = 0;
3828                 /* If we polled, count this as a successful poll */
3829                 if (vc->halt_poll_ns)
3830                         ++vc->runner->stat.halt_successful_poll;
3831                 goto out;
3832         }
3833
3834         start_wait = ktime_get();
3835
3836         vc->vcore_state = VCORE_SLEEPING;
3837         trace_kvmppc_vcore_blocked(vc, 0);
3838         spin_unlock(&vc->lock);
3839         schedule();
3840         finish_swait(&vc->wq, &wait);
3841         spin_lock(&vc->lock);
3842         vc->vcore_state = VCORE_INACTIVE;
3843         trace_kvmppc_vcore_blocked(vc, 1);
3844         ++vc->runner->stat.halt_successful_wait;
3845
3846         cur = ktime_get();
3847
3848 out:
3849         block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll);
3850
3851         /* Attribute wait time */
3852         if (do_sleep) {
3853                 vc->runner->stat.halt_wait_ns +=
3854                         ktime_to_ns(cur) - ktime_to_ns(start_wait);
3855                 /* Attribute failed poll time */
3856                 if (vc->halt_poll_ns)
3857                         vc->runner->stat.halt_poll_fail_ns +=
3858                                 ktime_to_ns(start_wait) -
3859                                 ktime_to_ns(start_poll);
3860         } else {
3861                 /* Attribute successful poll time */
3862                 if (vc->halt_poll_ns)
3863                         vc->runner->stat.halt_poll_success_ns +=
3864                                 ktime_to_ns(cur) -
3865                                 ktime_to_ns(start_poll);
3866         }
3867
3868         /* Adjust poll time */
3869         if (halt_poll_ns) {
3870                 if (block_ns <= vc->halt_poll_ns)
3871                         ;
3872                 /* We slept and blocked for longer than the max halt time */
3873                 else if (vc->halt_poll_ns && block_ns > halt_poll_ns)
3874                         shrink_halt_poll_ns(vc);
3875                 /* We slept and our poll time is too small */
3876                 else if (vc->halt_poll_ns < halt_poll_ns &&
3877                                 block_ns < halt_poll_ns)
3878                         grow_halt_poll_ns(vc);
3879                 if (vc->halt_poll_ns > halt_poll_ns)
3880                         vc->halt_poll_ns = halt_poll_ns;
3881         } else
3882                 vc->halt_poll_ns = 0;
3883
3884         trace_kvmppc_vcore_wakeup(do_sleep, block_ns);
3885 }
3886
3887 /*
3888  * This never fails for a radix guest, as none of the operations it does
3889  * for a radix guest can fail or have a way to report failure.
3890  * kvmhv_run_single_vcpu() relies on this fact.
3891  */
3892 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu)
3893 {
3894         int r = 0;
3895         struct kvm *kvm = vcpu->kvm;
3896
3897         mutex_lock(&kvm->arch.mmu_setup_lock);
3898         if (!kvm->arch.mmu_ready) {
3899                 if (!kvm_is_radix(kvm))
3900                         r = kvmppc_hv_setup_htab_rma(vcpu);
3901                 if (!r) {
3902                         if (cpu_has_feature(CPU_FTR_ARCH_300))
3903                                 kvmppc_setup_partition_table(kvm);
3904                         kvm->arch.mmu_ready = 1;
3905                 }
3906         }
3907         mutex_unlock(&kvm->arch.mmu_setup_lock);
3908         return r;
3909 }
3910
3911 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3912 {
3913         int n_ceded, i, r;
3914         struct kvmppc_vcore *vc;
3915         struct kvm_vcpu *v;
3916
3917         trace_kvmppc_run_vcpu_enter(vcpu);
3918
3919         kvm_run->exit_reason = 0;
3920         vcpu->arch.ret = RESUME_GUEST;
3921         vcpu->arch.trap = 0;
3922         kvmppc_update_vpas(vcpu);
3923
3924         /*
3925          * Synchronize with other threads in this virtual core
3926          */
3927         vc = vcpu->arch.vcore;
3928         spin_lock(&vc->lock);
3929         vcpu->arch.ceded = 0;
3930         vcpu->arch.run_task = current;
3931         vcpu->arch.kvm_run = kvm_run;
3932         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
3933         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
3934         vcpu->arch.busy_preempt = TB_NIL;
3935         WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu);
3936         ++vc->n_runnable;
3937
3938         /*
3939          * This happens the first time this is called for a vcpu.
3940          * If the vcore is already running, we may be able to start
3941          * this thread straight away and have it join in.
3942          */
3943         if (!signal_pending(current)) {
3944                 if ((vc->vcore_state == VCORE_PIGGYBACK ||
3945                      vc->vcore_state == VCORE_RUNNING) &&
3946                            !VCORE_IS_EXITING(vc)) {
3947                         kvmppc_create_dtl_entry(vcpu, vc);
3948                         kvmppc_start_thread(vcpu, vc);
3949                         trace_kvm_guest_enter(vcpu);
3950                 } else if (vc->vcore_state == VCORE_SLEEPING) {
3951                         swake_up_one(&vc->wq);
3952                 }
3953
3954         }
3955
3956         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
3957                !signal_pending(current)) {
3958                 /* See if the MMU is ready to go */
3959                 if (!vcpu->kvm->arch.mmu_ready) {
3960                         spin_unlock(&vc->lock);
3961                         r = kvmhv_setup_mmu(vcpu);
3962                         spin_lock(&vc->lock);
3963                         if (r) {
3964                                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3965                                 kvm_run->fail_entry.
3966                                         hardware_entry_failure_reason = 0;
3967                                 vcpu->arch.ret = r;
3968                                 break;
3969                         }
3970                 }
3971
3972                 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
3973                         kvmppc_vcore_end_preempt(vc);
3974
3975                 if (vc->vcore_state != VCORE_INACTIVE) {
3976                         kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE);
3977                         continue;
3978                 }
3979                 for_each_runnable_thread(i, v, vc) {
3980                         kvmppc_core_prepare_to_enter(v);
3981                         if (signal_pending(v->arch.run_task)) {
3982                                 kvmppc_remove_runnable(vc, v);
3983                                 v->stat.signal_exits++;
3984                                 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR;
3985                                 v->arch.ret = -EINTR;
3986                                 wake_up(&v->arch.cpu_run);
3987                         }
3988                 }
3989                 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE)
3990                         break;
3991                 n_ceded = 0;
3992                 for_each_runnable_thread(i, v, vc) {
3993                         if (!kvmppc_vcpu_woken(v))
3994                                 n_ceded += v->arch.ceded;
3995                         else
3996                                 v->arch.ceded = 0;
3997                 }
3998                 vc->runner = vcpu;
3999                 if (n_ceded == vc->n_runnable) {
4000                         kvmppc_vcore_blocked(vc);
4001                 } else if (need_resched()) {
4002                         kvmppc_vcore_preempt(vc);
4003                         /* Let something else run */
4004                         cond_resched_lock(&vc->lock);
4005                         if (vc->vcore_state == VCORE_PREEMPT)
4006                                 kvmppc_vcore_end_preempt(vc);
4007                 } else {
4008                         kvmppc_run_core(vc);
4009                 }
4010                 vc->runner = NULL;
4011         }
4012
4013         while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE &&
4014                (vc->vcore_state == VCORE_RUNNING ||
4015                 vc->vcore_state == VCORE_EXITING ||
4016                 vc->vcore_state == VCORE_PIGGYBACK))
4017                 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE);
4018
4019         if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL)
4020                 kvmppc_vcore_end_preempt(vc);
4021
4022         if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) {
4023                 kvmppc_remove_runnable(vc, vcpu);
4024                 vcpu->stat.signal_exits++;
4025                 kvm_run->exit_reason = KVM_EXIT_INTR;
4026                 vcpu->arch.ret = -EINTR;
4027         }
4028
4029         if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) {
4030                 /* Wake up some vcpu to run the core */
4031                 i = -1;
4032                 v = next_runnable_thread(vc, &i);
4033                 wake_up(&v->arch.cpu_run);
4034         }
4035
4036         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4037         spin_unlock(&vc->lock);
4038         return vcpu->arch.ret;
4039 }
4040
4041 int kvmhv_run_single_vcpu(struct kvm_run *kvm_run,
4042                           struct kvm_vcpu *vcpu, u64 time_limit,
4043                           unsigned long lpcr)
4044 {
4045         int trap, r, pcpu;
4046         int srcu_idx, lpid;
4047         struct kvmppc_vcore *vc;
4048         struct kvm *kvm = vcpu->kvm;
4049         struct kvm_nested_guest *nested = vcpu->arch.nested;
4050
4051         trace_kvmppc_run_vcpu_enter(vcpu);
4052
4053         kvm_run->exit_reason = 0;
4054         vcpu->arch.ret = RESUME_GUEST;
4055         vcpu->arch.trap = 0;
4056
4057         vc = vcpu->arch.vcore;
4058         vcpu->arch.ceded = 0;
4059         vcpu->arch.run_task = current;
4060         vcpu->arch.kvm_run = kvm_run;
4061         vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb());
4062         vcpu->arch.state = KVMPPC_VCPU_RUNNABLE;
4063         vcpu->arch.busy_preempt = TB_NIL;
4064         vcpu->arch.last_inst = KVM_INST_FETCH_FAILED;
4065         vc->runnable_threads[0] = vcpu;
4066         vc->n_runnable = 1;
4067         vc->runner = vcpu;
4068
4069         /* See if the MMU is ready to go */
4070         if (!kvm->arch.mmu_ready)
4071                 kvmhv_setup_mmu(vcpu);
4072
4073         if (need_resched())
4074                 cond_resched();
4075
4076         kvmppc_update_vpas(vcpu);
4077
4078         init_vcore_to_run(vc);
4079         vc->preempt_tb = TB_NIL;
4080
4081         preempt_disable();
4082         pcpu = smp_processor_id();
4083         vc->pcpu = pcpu;
4084         kvmppc_prepare_radix_vcpu(vcpu, pcpu);
4085
4086         local_irq_disable();
4087         hard_irq_disable();
4088         if (signal_pending(current))
4089                 goto sigpend;
4090         if (lazy_irq_pending() || need_resched() || !kvm->arch.mmu_ready)
4091                 goto out;
4092
4093         if (!nested) {
4094                 kvmppc_core_prepare_to_enter(vcpu);
4095                 if (vcpu->arch.doorbell_request) {
4096                         vc->dpdes = 1;
4097                         smp_wmb();
4098                         vcpu->arch.doorbell_request = 0;
4099                 }
4100                 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL,
4101                              &vcpu->arch.pending_exceptions))
4102                         lpcr |= LPCR_MER;
4103         } else if (vcpu->arch.pending_exceptions ||
4104                    vcpu->arch.doorbell_request ||
4105                    xive_interrupt_pending(vcpu)) {
4106                 vcpu->arch.ret = RESUME_HOST;
4107                 goto out;
4108         }
4109
4110         kvmppc_clear_host_core(pcpu);
4111
4112         local_paca->kvm_hstate.tid = 0;
4113         local_paca->kvm_hstate.napping = 0;
4114         local_paca->kvm_hstate.kvm_split_mode = NULL;
4115         kvmppc_start_thread(vcpu, vc);
4116         kvmppc_create_dtl_entry(vcpu, vc);
4117         trace_kvm_guest_enter(vcpu);
4118
4119         vc->vcore_state = VCORE_RUNNING;
4120         trace_kvmppc_run_core(vc, 0);
4121
4122         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4123                 lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
4124                 mtspr(SPRN_LPID, lpid);
4125                 isync();
4126                 kvmppc_check_need_tlb_flush(kvm, pcpu, nested);
4127         }
4128
4129         guest_enter_irqoff();
4130
4131         srcu_idx = srcu_read_lock(&kvm->srcu);
4132
4133         this_cpu_disable_ftrace();
4134
4135         /* Tell lockdep that we're about to enable interrupts */
4136         trace_hardirqs_on();
4137
4138         trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr);
4139         vcpu->arch.trap = trap;
4140
4141         trace_hardirqs_off();
4142
4143         this_cpu_enable_ftrace();
4144
4145         srcu_read_unlock(&kvm->srcu, srcu_idx);
4146
4147         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4148                 mtspr(SPRN_LPID, kvm->arch.host_lpid);
4149                 isync();
4150         }
4151
4152         set_irq_happened(trap);
4153
4154         kvmppc_set_host_core(pcpu);
4155
4156         local_irq_enable();
4157         guest_exit();
4158
4159         cpumask_clear_cpu(pcpu, &kvm->arch.cpu_in_guest);
4160
4161         preempt_enable();
4162
4163         /*
4164          * cancel pending decrementer exception if DEC is now positive, or if
4165          * entering a nested guest in which case the decrementer is now owned
4166          * by L2 and the L1 decrementer is provided in hdec_expires
4167          */
4168         if (kvmppc_core_pending_dec(vcpu) &&
4169                         ((get_tb() < vcpu->arch.dec_expires) ||
4170                          (trap == BOOK3S_INTERRUPT_SYSCALL &&
4171                           kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED)))
4172                 kvmppc_core_dequeue_dec(vcpu);
4173
4174         trace_kvm_guest_exit(vcpu);
4175         r = RESUME_GUEST;
4176         if (trap) {
4177                 if (!nested)
4178                         r = kvmppc_handle_exit_hv(kvm_run, vcpu, current);
4179                 else
4180                         r = kvmppc_handle_nested_exit(kvm_run, vcpu);
4181         }
4182         vcpu->arch.ret = r;
4183
4184         if (is_kvmppc_resume_guest(r) && vcpu->arch.ceded &&
4185             !kvmppc_vcpu_woken(vcpu)) {
4186                 kvmppc_set_timer(vcpu);
4187                 while (vcpu->arch.ceded && !kvmppc_vcpu_woken(vcpu)) {
4188                         if (signal_pending(current)) {
4189                                 vcpu->stat.signal_exits++;
4190                                 kvm_run->exit_reason = KVM_EXIT_INTR;
4191                                 vcpu->arch.ret = -EINTR;
4192                                 break;
4193                         }
4194                         spin_lock(&vc->lock);
4195                         kvmppc_vcore_blocked(vc);
4196                         spin_unlock(&vc->lock);
4197                 }
4198         }
4199         vcpu->arch.ceded = 0;
4200
4201         vc->vcore_state = VCORE_INACTIVE;
4202         trace_kvmppc_run_core(vc, 1);
4203
4204  done:
4205         kvmppc_remove_runnable(vc, vcpu);
4206         trace_kvmppc_run_vcpu_exit(vcpu, kvm_run);
4207
4208         return vcpu->arch.ret;
4209
4210  sigpend:
4211         vcpu->stat.signal_exits++;
4212         kvm_run->exit_reason = KVM_EXIT_INTR;
4213         vcpu->arch.ret = -EINTR;
4214  out:
4215         local_irq_enable();
4216         preempt_enable();
4217         goto done;
4218 }
4219
4220 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
4221 {
4222         int r;
4223         int srcu_idx;
4224         unsigned long ebb_regs[3] = {}; /* shut up GCC */
4225         unsigned long user_tar = 0;
4226         unsigned int user_vrsave;
4227         struct kvm *kvm;
4228
4229         if (!vcpu->arch.sane) {
4230                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4231                 return -EINVAL;
4232         }
4233
4234         /*
4235          * Don't allow entry with a suspended transaction, because
4236          * the guest entry/exit code will lose it.
4237          * If the guest has TM enabled, save away their TM-related SPRs
4238          * (they will get restored by the TM unavailable interrupt).
4239          */
4240 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
4241         if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
4242             (current->thread.regs->msr & MSR_TM)) {
4243                 if (MSR_TM_ACTIVE(current->thread.regs->msr)) {
4244                         run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4245                         run->fail_entry.hardware_entry_failure_reason = 0;
4246                         return -EINVAL;
4247                 }
4248                 /* Enable TM so we can read the TM SPRs */
4249                 mtmsr(mfmsr() | MSR_TM);
4250                 current->thread.tm_tfhar = mfspr(SPRN_TFHAR);
4251                 current->thread.tm_tfiar = mfspr(SPRN_TFIAR);
4252                 current->thread.tm_texasr = mfspr(SPRN_TEXASR);
4253                 current->thread.regs->msr &= ~MSR_TM;
4254         }
4255 #endif
4256
4257         /*
4258          * Force online to 1 for the sake of old userspace which doesn't
4259          * set it.
4260          */
4261         if (!vcpu->arch.online) {
4262                 atomic_inc(&vcpu->arch.vcore->online_count);
4263                 vcpu->arch.online = 1;
4264         }
4265
4266         kvmppc_core_prepare_to_enter(vcpu);
4267
4268         /* No need to go into the guest when all we'll do is come back out */
4269         if (signal_pending(current)) {
4270                 run->exit_reason = KVM_EXIT_INTR;
4271                 return -EINTR;
4272         }
4273
4274         kvm = vcpu->kvm;
4275         atomic_inc(&kvm->arch.vcpus_running);
4276         /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */
4277         smp_mb();
4278
4279         flush_all_to_thread(current);
4280
4281         /* Save userspace EBB and other register values */
4282         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4283                 ebb_regs[0] = mfspr(SPRN_EBBHR);
4284                 ebb_regs[1] = mfspr(SPRN_EBBRR);
4285                 ebb_regs[2] = mfspr(SPRN_BESCR);
4286                 user_tar = mfspr(SPRN_TAR);
4287         }
4288         user_vrsave = mfspr(SPRN_VRSAVE);
4289
4290         vcpu->arch.wqp = &vcpu->arch.vcore->wq;
4291         vcpu->arch.pgdir = kvm->mm->pgd;
4292         vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
4293
4294         do {
4295                 /*
4296                  * The early POWER9 chips that can't mix radix and HPT threads
4297                  * on the same core also need the workaround for the problem
4298                  * where the TLB would prefetch entries in the guest exit path
4299                  * for radix guests using the guest PIDR value and LPID 0.
4300                  * The workaround is in the old path (kvmppc_run_vcpu())
4301                  * but not the new path (kvmhv_run_single_vcpu()).
4302                  */
4303                 if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
4304                     !no_mixing_hpt_and_radix)
4305                         r = kvmhv_run_single_vcpu(run, vcpu, ~(u64)0,
4306                                                   vcpu->arch.vcore->lpcr);
4307                 else
4308                         r = kvmppc_run_vcpu(run, vcpu);
4309
4310                 if (run->exit_reason == KVM_EXIT_PAPR_HCALL &&
4311                     !(vcpu->arch.shregs.msr & MSR_PR)) {
4312                         trace_kvm_hcall_enter(vcpu);
4313                         r = kvmppc_pseries_do_hcall(vcpu);
4314                         trace_kvm_hcall_exit(vcpu, r);
4315                         kvmppc_core_prepare_to_enter(vcpu);
4316                 } else if (r == RESUME_PAGE_FAULT) {
4317                         srcu_idx = srcu_read_lock(&kvm->srcu);
4318                         r = kvmppc_book3s_hv_page_fault(run, vcpu,
4319                                 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr);
4320                         srcu_read_unlock(&kvm->srcu, srcu_idx);
4321                 } else if (r == RESUME_PASSTHROUGH) {
4322                         if (WARN_ON(xics_on_xive()))
4323                                 r = H_SUCCESS;
4324                         else
4325                                 r = kvmppc_xics_rm_complete(vcpu, 0);
4326                 }
4327         } while (is_kvmppc_resume_guest(r));
4328
4329         /* Restore userspace EBB and other register values */
4330         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
4331                 mtspr(SPRN_EBBHR, ebb_regs[0]);
4332                 mtspr(SPRN_EBBRR, ebb_regs[1]);
4333                 mtspr(SPRN_BESCR, ebb_regs[2]);
4334                 mtspr(SPRN_TAR, user_tar);
4335                 mtspr(SPRN_FSCR, current->thread.fscr);
4336         }
4337         mtspr(SPRN_VRSAVE, user_vrsave);
4338
4339         vcpu->arch.state = KVMPPC_VCPU_NOTREADY;
4340         atomic_dec(&kvm->arch.vcpus_running);
4341         return r;
4342 }
4343
4344 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps,
4345                                      int shift, int sllp)
4346 {
4347         (*sps)->page_shift = shift;
4348         (*sps)->slb_enc = sllp;
4349         (*sps)->enc[0].page_shift = shift;
4350         (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift);
4351         /*
4352          * Add 16MB MPSS support (may get filtered out by userspace)
4353          */
4354         if (shift != 24) {
4355                 int penc = kvmppc_pgsize_lp_encoding(shift, 24);
4356                 if (penc != -1) {
4357                         (*sps)->enc[1].page_shift = 24;
4358                         (*sps)->enc[1].pte_enc = penc;
4359                 }
4360         }
4361         (*sps)++;
4362 }
4363
4364 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm,
4365                                          struct kvm_ppc_smmu_info *info)
4366 {
4367         struct kvm_ppc_one_seg_page_size *sps;
4368
4369         /*
4370          * POWER7, POWER8 and POWER9 all support 32 storage keys for data.
4371          * POWER7 doesn't support keys for instruction accesses,
4372          * POWER8 and POWER9 do.
4373          */
4374         info->data_keys = 32;
4375         info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0;
4376
4377         /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */
4378         info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS;
4379         info->slb_size = 32;
4380
4381         /* We only support these sizes for now, and no muti-size segments */
4382         sps = &info->sps[0];
4383         kvmppc_add_seg_page_size(&sps, 12, 0);
4384         kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01);
4385         kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L);
4386
4387         /* If running as a nested hypervisor, we don't support HPT guests */
4388         if (kvmhv_on_pseries())
4389                 info->flags |= KVM_PPC_NO_HASH;
4390
4391         return 0;
4392 }
4393
4394 /*
4395  * Get (and clear) the dirty memory log for a memory slot.
4396  */
4397 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm,
4398                                          struct kvm_dirty_log *log)
4399 {
4400         struct kvm_memslots *slots;
4401         struct kvm_memory_slot *memslot;
4402         int i, r;
4403         unsigned long n;
4404         unsigned long *buf, *p;
4405         struct kvm_vcpu *vcpu;
4406
4407         mutex_lock(&kvm->slots_lock);
4408
4409         r = -EINVAL;
4410         if (log->slot >= KVM_USER_MEM_SLOTS)
4411                 goto out;
4412
4413         slots = kvm_memslots(kvm);
4414         memslot = id_to_memslot(slots, log->slot);
4415         r = -ENOENT;
4416         if (!memslot->dirty_bitmap)
4417                 goto out;
4418
4419         /*
4420          * Use second half of bitmap area because both HPT and radix
4421          * accumulate bits in the first half.
4422          */
4423         n = kvm_dirty_bitmap_bytes(memslot);
4424         buf = memslot->dirty_bitmap + n / sizeof(long);
4425         memset(buf, 0, n);
4426
4427         if (kvm_is_radix(kvm))
4428                 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf);
4429         else
4430                 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf);
4431         if (r)
4432                 goto out;
4433
4434         /*
4435          * We accumulate dirty bits in the first half of the
4436          * memslot's dirty_bitmap area, for when pages are paged
4437          * out or modified by the host directly.  Pick up these
4438          * bits and add them to the map.
4439          */
4440         p = memslot->dirty_bitmap;
4441         for (i = 0; i < n / sizeof(long); ++i)
4442                 buf[i] |= xchg(&p[i], 0);
4443
4444         /* Harvest dirty bits from VPA and DTL updates */
4445         /* Note: we never modify the SLB shadow buffer areas */
4446         kvm_for_each_vcpu(i, vcpu, kvm) {
4447                 spin_lock(&vcpu->arch.vpa_update_lock);
4448                 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf);
4449                 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf);
4450                 spin_unlock(&vcpu->arch.vpa_update_lock);
4451         }
4452
4453         r = -EFAULT;
4454         if (copy_to_user(log->dirty_bitmap, buf, n))
4455                 goto out;
4456
4457         r = 0;
4458 out:
4459         mutex_unlock(&kvm->slots_lock);
4460         return r;
4461 }
4462
4463 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free,
4464                                         struct kvm_memory_slot *dont)
4465 {
4466         if (!dont || free->arch.rmap != dont->arch.rmap) {
4467                 vfree(free->arch.rmap);
4468                 free->arch.rmap = NULL;
4469         }
4470 }
4471
4472 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot,
4473                                          unsigned long npages)
4474 {
4475         slot->arch.rmap = vzalloc(array_size(npages, sizeof(*slot->arch.rmap)));
4476         if (!slot->arch.rmap)
4477                 return -ENOMEM;
4478
4479         return 0;
4480 }
4481
4482 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm,
4483                                         struct kvm_memory_slot *memslot,
4484                                         const struct kvm_userspace_memory_region *mem)
4485 {
4486         return 0;
4487 }
4488
4489 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm,
4490                                 const struct kvm_userspace_memory_region *mem,
4491                                 const struct kvm_memory_slot *old,
4492                                 const struct kvm_memory_slot *new,
4493                                 enum kvm_mr_change change)
4494 {
4495         unsigned long npages = mem->memory_size >> PAGE_SHIFT;
4496
4497         /*
4498          * If we are making a new memslot, it might make
4499          * some address that was previously cached as emulated
4500          * MMIO be no longer emulated MMIO, so invalidate
4501          * all the caches of emulated MMIO translations.
4502          */
4503         if (npages)
4504                 atomic64_inc(&kvm->arch.mmio_update);
4505
4506         /*
4507          * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels
4508          * have already called kvm_arch_flush_shadow_memslot() to
4509          * flush shadow mappings.  For KVM_MR_CREATE we have no
4510          * previous mappings.  So the only case to handle is
4511          * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit
4512          * has been changed.
4513          * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES
4514          * to get rid of any THP PTEs in the partition-scoped page tables
4515          * so we can track dirtiness at the page level; we flush when
4516          * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to
4517          * using THP PTEs.
4518          */
4519         if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) &&
4520             ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES))
4521                 kvmppc_radix_flush_memslot(kvm, old);
4522         /*
4523          * If UV hasn't yet called H_SVM_INIT_START, don't register memslots.
4524          */
4525         if (!kvm->arch.secure_guest)
4526                 return;
4527
4528         switch (change) {
4529         case KVM_MR_CREATE:
4530                 if (kvmppc_uvmem_slot_init(kvm, new))
4531                         return;
4532                 uv_register_mem_slot(kvm->arch.lpid,
4533                                      new->base_gfn << PAGE_SHIFT,
4534                                      new->npages * PAGE_SIZE,
4535                                      0, new->id);
4536                 break;
4537         case KVM_MR_DELETE:
4538                 uv_unregister_mem_slot(kvm->arch.lpid, old->id);
4539                 kvmppc_uvmem_slot_free(kvm, old);
4540                 break;
4541         default:
4542                 /* TODO: Handle KVM_MR_MOVE */
4543                 break;
4544         }
4545 }
4546
4547 /*
4548  * Update LPCR values in kvm->arch and in vcores.
4549  * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion
4550  * of kvm->arch.lpcr update).
4551  */
4552 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask)
4553 {
4554         long int i;
4555         u32 cores_done = 0;
4556
4557         if ((kvm->arch.lpcr & mask) == lpcr)
4558                 return;
4559
4560         kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr;
4561
4562         for (i = 0; i < KVM_MAX_VCORES; ++i) {
4563                 struct kvmppc_vcore *vc = kvm->arch.vcores[i];
4564                 if (!vc)
4565                         continue;
4566                 spin_lock(&vc->lock);
4567                 vc->lpcr = (vc->lpcr & ~mask) | lpcr;
4568                 spin_unlock(&vc->lock);
4569                 if (++cores_done >= kvm->arch.online_vcores)
4570                         break;
4571         }
4572 }
4573
4574 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu)
4575 {
4576         return;
4577 }
4578
4579 void kvmppc_setup_partition_table(struct kvm *kvm)
4580 {
4581         unsigned long dw0, dw1;
4582
4583         if (!kvm_is_radix(kvm)) {
4584                 /* PS field - page size for VRMA */
4585                 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) |
4586                         ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1);
4587                 /* HTABSIZE and HTABORG fields */
4588                 dw0 |= kvm->arch.sdr1;
4589
4590                 /* Second dword as set by userspace */
4591                 dw1 = kvm->arch.process_table;
4592         } else {
4593                 dw0 = PATB_HR | radix__get_tree_size() |
4594                         __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE;
4595                 dw1 = PATB_GR | kvm->arch.process_table;
4596         }
4597         kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1);
4598 }
4599
4600 /*
4601  * Set up HPT (hashed page table) and RMA (real-mode area).
4602  * Must be called with kvm->arch.mmu_setup_lock held.
4603  */
4604 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu)
4605 {
4606         int err = 0;
4607         struct kvm *kvm = vcpu->kvm;
4608         unsigned long hva;
4609         struct kvm_memory_slot *memslot;
4610         struct vm_area_struct *vma;
4611         unsigned long lpcr = 0, senc;
4612         unsigned long psize, porder;
4613         int srcu_idx;
4614
4615         /* Allocate hashed page table (if not done already) and reset it */
4616         if (!kvm->arch.hpt.virt) {
4617                 int order = KVM_DEFAULT_HPT_ORDER;
4618                 struct kvm_hpt_info info;
4619
4620                 err = kvmppc_allocate_hpt(&info, order);
4621                 /* If we get here, it means userspace didn't specify a
4622                  * size explicitly.  So, try successively smaller
4623                  * sizes if the default failed. */
4624                 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER)
4625                         err  = kvmppc_allocate_hpt(&info, order);
4626
4627                 if (err < 0) {
4628                         pr_err("KVM: Couldn't alloc HPT\n");
4629                         goto out;
4630                 }
4631
4632                 kvmppc_set_hpt(kvm, &info);
4633         }
4634
4635         /* Look up the memslot for guest physical address 0 */
4636         srcu_idx = srcu_read_lock(&kvm->srcu);
4637         memslot = gfn_to_memslot(kvm, 0);
4638
4639         /* We must have some memory at 0 by now */
4640         err = -EINVAL;
4641         if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
4642                 goto out_srcu;
4643
4644         /* Look up the VMA for the start of this memory slot */
4645         hva = memslot->userspace_addr;
4646         down_read(&kvm->mm->mmap_sem);
4647         vma = find_vma(kvm->mm, hva);
4648         if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO))
4649                 goto up_out;
4650
4651         psize = vma_kernel_pagesize(vma);
4652
4653         up_read(&kvm->mm->mmap_sem);
4654
4655         /* We can handle 4k, 64k or 16M pages in the VRMA */
4656         if (psize >= 0x1000000)
4657                 psize = 0x1000000;
4658         else if (psize >= 0x10000)
4659                 psize = 0x10000;
4660         else
4661                 psize = 0x1000;
4662         porder = __ilog2(psize);
4663
4664         senc = slb_pgsize_encoding(psize);
4665         kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
4666                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4667         /* Create HPTEs in the hash page table for the VRMA */
4668         kvmppc_map_vrma(vcpu, memslot, porder);
4669
4670         /* Update VRMASD field in the LPCR */
4671         if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
4672                 /* the -4 is to account for senc values starting at 0x10 */
4673                 lpcr = senc << (LPCR_VRMASD_SH - 4);
4674                 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
4675         }
4676
4677         /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */
4678         smp_wmb();
4679         err = 0;
4680  out_srcu:
4681         srcu_read_unlock(&kvm->srcu, srcu_idx);
4682  out:
4683         return err;
4684
4685  up_out:
4686         up_read(&kvm->mm->mmap_sem);
4687         goto out_srcu;
4688 }
4689
4690 /*
4691  * Must be called with kvm->arch.mmu_setup_lock held and
4692  * mmu_ready = 0 and no vcpus running.
4693  */
4694 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm)
4695 {
4696         if (nesting_enabled(kvm))
4697                 kvmhv_release_all_nested(kvm);
4698         kvmppc_rmap_reset(kvm);
4699         kvm->arch.process_table = 0;
4700         /* Mutual exclusion with kvm_unmap_hva_range etc. */
4701         spin_lock(&kvm->mmu_lock);
4702         kvm->arch.radix = 0;
4703         spin_unlock(&kvm->mmu_lock);
4704         kvmppc_free_radix(kvm);
4705         kvmppc_update_lpcr(kvm, LPCR_VPM1,
4706                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4707         return 0;
4708 }
4709
4710 /*
4711  * Must be called with kvm->arch.mmu_setup_lock held and
4712  * mmu_ready = 0 and no vcpus running.
4713  */
4714 int kvmppc_switch_mmu_to_radix(struct kvm *kvm)
4715 {
4716         int err;
4717
4718         err = kvmppc_init_vm_radix(kvm);
4719         if (err)
4720                 return err;
4721         kvmppc_rmap_reset(kvm);
4722         /* Mutual exclusion with kvm_unmap_hva_range etc. */
4723         spin_lock(&kvm->mmu_lock);
4724         kvm->arch.radix = 1;
4725         spin_unlock(&kvm->mmu_lock);
4726         kvmppc_free_hpt(&kvm->arch.hpt);
4727         kvmppc_update_lpcr(kvm, LPCR_UPRT | LPCR_GTSE | LPCR_HR,
4728                            LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR);
4729         return 0;
4730 }
4731
4732 #ifdef CONFIG_KVM_XICS
4733 /*
4734  * Allocate a per-core structure for managing state about which cores are
4735  * running in the host versus the guest and for exchanging data between
4736  * real mode KVM and CPU running in the host.
4737  * This is only done for the first VM.
4738  * The allocated structure stays even if all VMs have stopped.
4739  * It is only freed when the kvm-hv module is unloaded.
4740  * It's OK for this routine to fail, we just don't support host
4741  * core operations like redirecting H_IPI wakeups.
4742  */
4743 void kvmppc_alloc_host_rm_ops(void)
4744 {
4745         struct kvmppc_host_rm_ops *ops;
4746         unsigned long l_ops;
4747         int cpu, core;
4748         int size;
4749
4750         /* Not the first time here ? */
4751         if (kvmppc_host_rm_ops_hv != NULL)
4752                 return;
4753
4754         ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL);
4755         if (!ops)
4756                 return;
4757
4758         size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core);
4759         ops->rm_core = kzalloc(size, GFP_KERNEL);
4760
4761         if (!ops->rm_core) {
4762                 kfree(ops);
4763                 return;
4764         }
4765
4766         cpus_read_lock();
4767
4768         for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) {
4769                 if (!cpu_online(cpu))
4770                         continue;
4771
4772                 core = cpu >> threads_shift;
4773                 ops->rm_core[core].rm_state.in_host = 1;
4774         }
4775
4776         ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv;
4777
4778         /*
4779          * Make the contents of the kvmppc_host_rm_ops structure visible
4780          * to other CPUs before we assign it to the global variable.
4781          * Do an atomic assignment (no locks used here), but if someone
4782          * beats us to it, just free our copy and return.
4783          */
4784         smp_wmb();
4785         l_ops = (unsigned long) ops;
4786
4787         if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) {
4788                 cpus_read_unlock();
4789                 kfree(ops->rm_core);
4790                 kfree(ops);
4791                 return;
4792         }
4793
4794         cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE,
4795                                              "ppc/kvm_book3s:prepare",
4796                                              kvmppc_set_host_core,
4797                                              kvmppc_clear_host_core);
4798         cpus_read_unlock();
4799 }
4800
4801 void kvmppc_free_host_rm_ops(void)
4802 {
4803         if (kvmppc_host_rm_ops_hv) {
4804                 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE);
4805                 kfree(kvmppc_host_rm_ops_hv->rm_core);
4806                 kfree(kvmppc_host_rm_ops_hv);
4807                 kvmppc_host_rm_ops_hv = NULL;
4808         }
4809 }
4810 #endif
4811
4812 static int kvmppc_core_init_vm_hv(struct kvm *kvm)
4813 {
4814         unsigned long lpcr, lpid;
4815         char buf[32];
4816         int ret;
4817
4818         mutex_init(&kvm->arch.uvmem_lock);
4819         INIT_LIST_HEAD(&kvm->arch.uvmem_pfns);
4820         mutex_init(&kvm->arch.mmu_setup_lock);
4821
4822         /* Allocate the guest's logical partition ID */
4823
4824         lpid = kvmppc_alloc_lpid();
4825         if ((long)lpid < 0)
4826                 return -ENOMEM;
4827         kvm->arch.lpid = lpid;
4828
4829         kvmppc_alloc_host_rm_ops();
4830
4831         kvmhv_vm_nested_init(kvm);
4832
4833         /*
4834          * Since we don't flush the TLB when tearing down a VM,
4835          * and this lpid might have previously been used,
4836          * make sure we flush on each core before running the new VM.
4837          * On POWER9, the tlbie in mmu_partition_table_set_entry()
4838          * does this flush for us.
4839          */
4840         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4841                 cpumask_setall(&kvm->arch.need_tlb_flush);
4842
4843         /* Start out with the default set of hcalls enabled */
4844         memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls,
4845                sizeof(kvm->arch.enabled_hcalls));
4846
4847         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4848                 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1);
4849
4850         /* Init LPCR for virtual RMA mode */
4851         if (cpu_has_feature(CPU_FTR_HVMODE)) {
4852                 kvm->arch.host_lpid = mfspr(SPRN_LPID);
4853                 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR);
4854                 lpcr &= LPCR_PECE | LPCR_LPES;
4855         } else {
4856                 lpcr = 0;
4857         }
4858         lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE |
4859                 LPCR_VPM0 | LPCR_VPM1;
4860         kvm->arch.vrma_slb_v = SLB_VSID_B_1T |
4861                 (VRMA_VSID << SLB_VSID_SHIFT_1T);
4862         /* On POWER8 turn on online bit to enable PURR/SPURR */
4863         if (cpu_has_feature(CPU_FTR_ARCH_207S))
4864                 lpcr |= LPCR_ONL;
4865         /*
4866          * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed)
4867          * Set HVICE bit to enable hypervisor virtualization interrupts.
4868          * Set HEIC to prevent OS interrupts to go to hypervisor (should
4869          * be unnecessary but better safe than sorry in case we re-enable
4870          * EE in HV mode with this LPCR still set)
4871          */
4872         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4873                 lpcr &= ~LPCR_VPM0;
4874                 lpcr |= LPCR_HVICE | LPCR_HEIC;
4875
4876                 /*
4877                  * If xive is enabled, we route 0x500 interrupts directly
4878                  * to the guest.
4879                  */
4880                 if (xics_on_xive())
4881                         lpcr |= LPCR_LPES;
4882         }
4883
4884         /*
4885          * If the host uses radix, the guest starts out as radix.
4886          */
4887         if (radix_enabled()) {
4888                 kvm->arch.radix = 1;
4889                 kvm->arch.mmu_ready = 1;
4890                 lpcr &= ~LPCR_VPM1;
4891                 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR;
4892                 ret = kvmppc_init_vm_radix(kvm);
4893                 if (ret) {
4894                         kvmppc_free_lpid(kvm->arch.lpid);
4895                         return ret;
4896                 }
4897                 kvmppc_setup_partition_table(kvm);
4898         }
4899
4900         kvm->arch.lpcr = lpcr;
4901
4902         /* Initialization for future HPT resizes */
4903         kvm->arch.resize_hpt = NULL;
4904
4905         /*
4906          * Work out how many sets the TLB has, for the use of
4907          * the TLB invalidation loop in book3s_hv_rmhandlers.S.
4908          */
4909         if (radix_enabled())
4910                 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX;     /* 128 */
4911         else if (cpu_has_feature(CPU_FTR_ARCH_300))
4912                 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH;      /* 256 */
4913         else if (cpu_has_feature(CPU_FTR_ARCH_207S))
4914                 kvm->arch.tlb_sets = POWER8_TLB_SETS;           /* 512 */
4915         else
4916                 kvm->arch.tlb_sets = POWER7_TLB_SETS;           /* 128 */
4917
4918         /*
4919          * Track that we now have a HV mode VM active. This blocks secondary
4920          * CPU threads from coming online.
4921          * On POWER9, we only need to do this if the "indep_threads_mode"
4922          * module parameter has been set to N.
4923          */
4924         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4925                 if (!indep_threads_mode && !cpu_has_feature(CPU_FTR_HVMODE)) {
4926                         pr_warn("KVM: Ignoring indep_threads_mode=N in nested hypervisor\n");
4927                         kvm->arch.threads_indep = true;
4928                 } else {
4929                         kvm->arch.threads_indep = indep_threads_mode;
4930                 }
4931         }
4932         if (!kvm->arch.threads_indep)
4933                 kvm_hv_vm_activated();
4934
4935         /*
4936          * Initialize smt_mode depending on processor.
4937          * POWER8 and earlier have to use "strict" threading, where
4938          * all vCPUs in a vcore have to run on the same (sub)core,
4939          * whereas on POWER9 the threads can each run a different
4940          * guest.
4941          */
4942         if (!cpu_has_feature(CPU_FTR_ARCH_300))
4943                 kvm->arch.smt_mode = threads_per_subcore;
4944         else
4945                 kvm->arch.smt_mode = 1;
4946         kvm->arch.emul_smt_mode = 1;
4947
4948         /*
4949          * Create a debugfs directory for the VM
4950          */
4951         snprintf(buf, sizeof(buf), "vm%d", current->pid);
4952         kvm->arch.debugfs_dir = debugfs_create_dir(buf, kvm_debugfs_dir);
4953         kvmppc_mmu_debugfs_init(kvm);
4954         if (radix_enabled())
4955                 kvmhv_radix_debugfs_init(kvm);
4956
4957         return 0;
4958 }
4959
4960 static void kvmppc_free_vcores(struct kvm *kvm)
4961 {
4962         long int i;
4963
4964         for (i = 0; i < KVM_MAX_VCORES; ++i)
4965                 kfree(kvm->arch.vcores[i]);
4966         kvm->arch.online_vcores = 0;
4967 }
4968
4969 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm)
4970 {
4971         debugfs_remove_recursive(kvm->arch.debugfs_dir);
4972
4973         if (!kvm->arch.threads_indep)
4974                 kvm_hv_vm_deactivated();
4975
4976         kvmppc_free_vcores(kvm);
4977
4978
4979         if (kvm_is_radix(kvm))
4980                 kvmppc_free_radix(kvm);
4981         else
4982                 kvmppc_free_hpt(&kvm->arch.hpt);
4983
4984         /* Perform global invalidation and return lpid to the pool */
4985         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
4986                 if (nesting_enabled(kvm))
4987                         kvmhv_release_all_nested(kvm);
4988                 kvm->arch.process_table = 0;
4989                 if (kvm->arch.secure_guest)
4990                         uv_svm_terminate(kvm->arch.lpid);
4991                 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0);
4992         }
4993
4994         kvmppc_free_lpid(kvm->arch.lpid);
4995
4996         kvmppc_free_pimap(kvm);
4997 }
4998
4999 /* We don't need to emulate any privileged instructions or dcbz */
5000 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
5001                                      unsigned int inst, int *advance)
5002 {
5003         return EMULATE_FAIL;
5004 }
5005
5006 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn,
5007                                         ulong spr_val)
5008 {
5009         return EMULATE_FAIL;
5010 }
5011
5012 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn,
5013                                         ulong *spr_val)
5014 {
5015         return EMULATE_FAIL;
5016 }
5017
5018 static int kvmppc_core_check_processor_compat_hv(void)
5019 {
5020         if (cpu_has_feature(CPU_FTR_HVMODE) &&
5021             cpu_has_feature(CPU_FTR_ARCH_206))
5022                 return 0;
5023
5024         /* POWER9 in radix mode is capable of being a nested hypervisor. */
5025         if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled())
5026                 return 0;
5027
5028         return -EIO;
5029 }
5030
5031 #ifdef CONFIG_KVM_XICS
5032
5033 void kvmppc_free_pimap(struct kvm *kvm)
5034 {
5035         kfree(kvm->arch.pimap);
5036 }
5037
5038 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void)
5039 {
5040         return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL);
5041 }
5042
5043 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5044 {
5045         struct irq_desc *desc;
5046         struct kvmppc_irq_map *irq_map;
5047         struct kvmppc_passthru_irqmap *pimap;
5048         struct irq_chip *chip;
5049         int i, rc = 0;
5050
5051         if (!kvm_irq_bypass)
5052                 return 1;
5053
5054         desc = irq_to_desc(host_irq);
5055         if (!desc)
5056                 return -EIO;
5057
5058         mutex_lock(&kvm->lock);
5059
5060         pimap = kvm->arch.pimap;
5061         if (pimap == NULL) {
5062                 /* First call, allocate structure to hold IRQ map */
5063                 pimap = kvmppc_alloc_pimap();
5064                 if (pimap == NULL) {
5065                         mutex_unlock(&kvm->lock);
5066                         return -ENOMEM;
5067                 }
5068                 kvm->arch.pimap = pimap;
5069         }
5070
5071         /*
5072          * For now, we only support interrupts for which the EOI operation
5073          * is an OPAL call followed by a write to XIRR, since that's
5074          * what our real-mode EOI code does, or a XIVE interrupt
5075          */
5076         chip = irq_data_get_irq_chip(&desc->irq_data);
5077         if (!chip || !(is_pnv_opal_msi(chip) || is_xive_irq(chip))) {
5078                 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n",
5079                         host_irq, guest_gsi);
5080                 mutex_unlock(&kvm->lock);
5081                 return -ENOENT;
5082         }
5083
5084         /*
5085          * See if we already have an entry for this guest IRQ number.
5086          * If it's mapped to a hardware IRQ number, that's an error,
5087          * otherwise re-use this entry.
5088          */
5089         for (i = 0; i < pimap->n_mapped; i++) {
5090                 if (guest_gsi == pimap->mapped[i].v_hwirq) {
5091                         if (pimap->mapped[i].r_hwirq) {
5092                                 mutex_unlock(&kvm->lock);
5093                                 return -EINVAL;
5094                         }
5095                         break;
5096                 }
5097         }
5098
5099         if (i == KVMPPC_PIRQ_MAPPED) {
5100                 mutex_unlock(&kvm->lock);
5101                 return -EAGAIN;         /* table is full */
5102         }
5103
5104         irq_map = &pimap->mapped[i];
5105
5106         irq_map->v_hwirq = guest_gsi;
5107         irq_map->desc = desc;
5108
5109         /*
5110          * Order the above two stores before the next to serialize with
5111          * the KVM real mode handler.
5112          */
5113         smp_wmb();
5114         irq_map->r_hwirq = desc->irq_data.hwirq;
5115
5116         if (i == pimap->n_mapped)
5117                 pimap->n_mapped++;
5118
5119         if (xics_on_xive())
5120                 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, desc);
5121         else
5122                 kvmppc_xics_set_mapped(kvm, guest_gsi, desc->irq_data.hwirq);
5123         if (rc)
5124                 irq_map->r_hwirq = 0;
5125
5126         mutex_unlock(&kvm->lock);
5127
5128         return 0;
5129 }
5130
5131 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi)
5132 {
5133         struct irq_desc *desc;
5134         struct kvmppc_passthru_irqmap *pimap;
5135         int i, rc = 0;
5136
5137         if (!kvm_irq_bypass)
5138                 return 0;
5139
5140         desc = irq_to_desc(host_irq);
5141         if (!desc)
5142                 return -EIO;
5143
5144         mutex_lock(&kvm->lock);
5145         if (!kvm->arch.pimap)
5146                 goto unlock;
5147
5148         pimap = kvm->arch.pimap;
5149
5150         for (i = 0; i < pimap->n_mapped; i++) {
5151                 if (guest_gsi == pimap->mapped[i].v_hwirq)
5152                         break;
5153         }
5154
5155         if (i == pimap->n_mapped) {
5156                 mutex_unlock(&kvm->lock);
5157                 return -ENODEV;
5158         }
5159
5160         if (xics_on_xive())
5161                 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, pimap->mapped[i].desc);
5162         else
5163                 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq);
5164
5165         /* invalidate the entry (what do do on error from the above ?) */
5166         pimap->mapped[i].r_hwirq = 0;
5167
5168         /*
5169          * We don't free this structure even when the count goes to
5170          * zero. The structure is freed when we destroy the VM.
5171          */
5172  unlock:
5173         mutex_unlock(&kvm->lock);
5174         return rc;
5175 }
5176
5177 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons,
5178                                              struct irq_bypass_producer *prod)
5179 {
5180         int ret = 0;
5181         struct kvm_kernel_irqfd *irqfd =
5182                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5183
5184         irqfd->producer = prod;
5185
5186         ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5187         if (ret)
5188                 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n",
5189                         prod->irq, irqfd->gsi, ret);
5190
5191         return ret;
5192 }
5193
5194 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons,
5195                                               struct irq_bypass_producer *prod)
5196 {
5197         int ret;
5198         struct kvm_kernel_irqfd *irqfd =
5199                 container_of(cons, struct kvm_kernel_irqfd, consumer);
5200
5201         irqfd->producer = NULL;
5202
5203         /*
5204          * When producer of consumer is unregistered, we change back to
5205          * default external interrupt handling mode - KVM real mode
5206          * will switch back to host.
5207          */
5208         ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi);
5209         if (ret)
5210                 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n",
5211                         prod->irq, irqfd->gsi, ret);
5212 }
5213 #endif
5214
5215 static long kvm_arch_vm_ioctl_hv(struct file *filp,
5216                                  unsigned int ioctl, unsigned long arg)
5217 {
5218         struct kvm *kvm __maybe_unused = filp->private_data;
5219         void __user *argp = (void __user *)arg;
5220         long r;
5221
5222         switch (ioctl) {
5223
5224         case KVM_PPC_ALLOCATE_HTAB: {
5225                 u32 htab_order;
5226
5227                 r = -EFAULT;
5228                 if (get_user(htab_order, (u32 __user *)argp))
5229                         break;
5230                 r = kvmppc_alloc_reset_hpt(kvm, htab_order);
5231                 if (r)
5232                         break;
5233                 r = 0;
5234                 break;
5235         }
5236
5237         case KVM_PPC_GET_HTAB_FD: {
5238                 struct kvm_get_htab_fd ghf;
5239
5240                 r = -EFAULT;
5241                 if (copy_from_user(&ghf, argp, sizeof(ghf)))
5242                         break;
5243                 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf);
5244                 break;
5245         }
5246
5247         case KVM_PPC_RESIZE_HPT_PREPARE: {
5248                 struct kvm_ppc_resize_hpt rhpt;
5249
5250                 r = -EFAULT;
5251                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5252                         break;
5253
5254                 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt);
5255                 break;
5256         }
5257
5258         case KVM_PPC_RESIZE_HPT_COMMIT: {
5259                 struct kvm_ppc_resize_hpt rhpt;
5260
5261                 r = -EFAULT;
5262                 if (copy_from_user(&rhpt, argp, sizeof(rhpt)))
5263                         break;
5264
5265                 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt);
5266                 break;
5267         }
5268
5269         default:
5270                 r = -ENOTTY;
5271         }
5272
5273         return r;
5274 }
5275
5276 /*
5277  * List of hcall numbers to enable by default.
5278  * For compatibility with old userspace, we enable by default
5279  * all hcalls that were implemented before the hcall-enabling
5280  * facility was added.  Note this list should not include H_RTAS.
5281  */
5282 static unsigned int default_hcall_list[] = {
5283         H_REMOVE,
5284         H_ENTER,
5285         H_READ,
5286         H_PROTECT,
5287         H_BULK_REMOVE,
5288         H_GET_TCE,
5289         H_PUT_TCE,
5290         H_SET_DABR,
5291         H_SET_XDABR,
5292         H_CEDE,
5293         H_PROD,
5294         H_CONFER,
5295         H_REGISTER_VPA,
5296 #ifdef CONFIG_KVM_XICS
5297         H_EOI,
5298         H_CPPR,
5299         H_IPI,
5300         H_IPOLL,
5301         H_XIRR,
5302         H_XIRR_X,
5303 #endif
5304         0
5305 };
5306
5307 static void init_default_hcalls(void)
5308 {
5309         int i;
5310         unsigned int hcall;
5311
5312         for (i = 0; default_hcall_list[i]; ++i) {
5313                 hcall = default_hcall_list[i];
5314                 WARN_ON(!kvmppc_hcall_impl_hv(hcall));
5315                 __set_bit(hcall / 4, default_enabled_hcalls);
5316         }
5317 }
5318
5319 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg)
5320 {
5321         unsigned long lpcr;
5322         int radix;
5323         int err;
5324
5325         /* If not on a POWER9, reject it */
5326         if (!cpu_has_feature(CPU_FTR_ARCH_300))
5327                 return -ENODEV;
5328
5329         /* If any unknown flags set, reject it */
5330         if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE))
5331                 return -EINVAL;
5332
5333         /* GR (guest radix) bit in process_table field must match */
5334         radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX);
5335         if (!!(cfg->process_table & PATB_GR) != radix)
5336                 return -EINVAL;
5337
5338         /* Process table size field must be reasonable, i.e. <= 24 */
5339         if ((cfg->process_table & PRTS_MASK) > 24)
5340                 return -EINVAL;
5341
5342         /* We can change a guest to/from radix now, if the host is radix */
5343         if (radix && !radix_enabled())
5344                 return -EINVAL;
5345
5346         /* If we're a nested hypervisor, we currently only support radix */
5347         if (kvmhv_on_pseries() && !radix)
5348                 return -EINVAL;
5349
5350         mutex_lock(&kvm->arch.mmu_setup_lock);
5351         if (radix != kvm_is_radix(kvm)) {
5352                 if (kvm->arch.mmu_ready) {
5353                         kvm->arch.mmu_ready = 0;
5354                         /* order mmu_ready vs. vcpus_running */
5355                         smp_mb();
5356                         if (atomic_read(&kvm->arch.vcpus_running)) {
5357                                 kvm->arch.mmu_ready = 1;
5358                                 err = -EBUSY;
5359                                 goto out_unlock;
5360                         }
5361                 }
5362                 if (radix)
5363                         err = kvmppc_switch_mmu_to_radix(kvm);
5364                 else
5365                         err = kvmppc_switch_mmu_to_hpt(kvm);
5366                 if (err)
5367                         goto out_unlock;
5368         }
5369
5370         kvm->arch.process_table = cfg->process_table;
5371         kvmppc_setup_partition_table(kvm);
5372
5373         lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0;
5374         kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE);
5375         err = 0;
5376
5377  out_unlock:
5378         mutex_unlock(&kvm->arch.mmu_setup_lock);
5379         return err;
5380 }
5381
5382 static int kvmhv_enable_nested(struct kvm *kvm)
5383 {
5384         if (!nested)
5385                 return -EPERM;
5386         if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
5387                 return -ENODEV;
5388
5389         /* kvm == NULL means the caller is testing if the capability exists */
5390         if (kvm)
5391                 kvm->arch.nested_enable = true;
5392         return 0;
5393 }
5394
5395 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5396                                  int size)
5397 {
5398         int rc = -EINVAL;
5399
5400         if (kvmhv_vcpu_is_radix(vcpu)) {
5401                 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size);
5402
5403                 if (rc > 0)
5404                         rc = -EINVAL;
5405         }
5406
5407         /* For now quadrants are the only way to access nested guest memory */
5408         if (rc && vcpu->arch.nested)
5409                 rc = -EAGAIN;
5410
5411         return rc;
5412 }
5413
5414 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr,
5415                                 int size)
5416 {
5417         int rc = -EINVAL;
5418
5419         if (kvmhv_vcpu_is_radix(vcpu)) {
5420                 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size);
5421
5422                 if (rc > 0)
5423                         rc = -EINVAL;
5424         }
5425
5426         /* For now quadrants are the only way to access nested guest memory */
5427         if (rc && vcpu->arch.nested)
5428                 rc = -EAGAIN;
5429
5430         return rc;
5431 }
5432
5433 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa)
5434 {
5435         unpin_vpa(kvm, vpa);
5436         vpa->gpa = 0;
5437         vpa->pinned_addr = NULL;
5438         vpa->dirty = false;
5439         vpa->update_pending = 0;
5440 }
5441
5442 /*
5443  *  IOCTL handler to turn off secure mode of guest
5444  *
5445  * - Release all device pages
5446  * - Issue ucall to terminate the guest on the UV side
5447  * - Unpin the VPA pages.
5448  * - Reinit the partition scoped page tables
5449  */
5450 static int kvmhv_svm_off(struct kvm *kvm)
5451 {
5452         struct kvm_vcpu *vcpu;
5453         int mmu_was_ready;
5454         int srcu_idx;
5455         int ret = 0;
5456         int i;
5457
5458         if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START))
5459                 return ret;
5460
5461         mutex_lock(&kvm->arch.mmu_setup_lock);
5462         mmu_was_ready = kvm->arch.mmu_ready;
5463         if (kvm->arch.mmu_ready) {
5464                 kvm->arch.mmu_ready = 0;
5465                 /* order mmu_ready vs. vcpus_running */
5466                 smp_mb();
5467                 if (atomic_read(&kvm->arch.vcpus_running)) {
5468                         kvm->arch.mmu_ready = 1;
5469                         ret = -EBUSY;
5470                         goto out;
5471                 }
5472         }
5473
5474         srcu_idx = srcu_read_lock(&kvm->srcu);
5475         for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
5476                 struct kvm_memory_slot *memslot;
5477                 struct kvm_memslots *slots = __kvm_memslots(kvm, i);
5478
5479                 if (!slots)
5480                         continue;
5481
5482                 kvm_for_each_memslot(memslot, slots) {
5483                         kvmppc_uvmem_drop_pages(memslot, kvm, true);
5484                         uv_unregister_mem_slot(kvm->arch.lpid, memslot->id);
5485                 }
5486         }
5487         srcu_read_unlock(&kvm->srcu, srcu_idx);
5488
5489         ret = uv_svm_terminate(kvm->arch.lpid);
5490         if (ret != U_SUCCESS) {
5491                 ret = -EINVAL;
5492                 goto out;
5493         }
5494
5495         /*
5496          * When secure guest is reset, all the guest pages are sent
5497          * to UV via UV_PAGE_IN before the non-boot vcpus get a
5498          * chance to run and unpin their VPA pages. Unpinning of all
5499          * VPA pages is done here explicitly so that VPA pages
5500          * can be migrated to the secure side.
5501          *
5502          * This is required to for the secure SMP guest to reboot
5503          * correctly.
5504          */
5505         kvm_for_each_vcpu(i, vcpu, kvm) {
5506                 spin_lock(&vcpu->arch.vpa_update_lock);
5507                 unpin_vpa_reset(kvm, &vcpu->arch.dtl);
5508                 unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow);
5509                 unpin_vpa_reset(kvm, &vcpu->arch.vpa);
5510                 spin_unlock(&vcpu->arch.vpa_update_lock);
5511         }
5512
5513         kvmppc_setup_partition_table(kvm);
5514         kvm->arch.secure_guest = 0;
5515         kvm->arch.mmu_ready = mmu_was_ready;
5516 out:
5517         mutex_unlock(&kvm->arch.mmu_setup_lock);
5518         return ret;
5519 }
5520
5521 static struct kvmppc_ops kvm_ops_hv = {
5522         .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv,
5523         .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv,
5524         .get_one_reg = kvmppc_get_one_reg_hv,
5525         .set_one_reg = kvmppc_set_one_reg_hv,
5526         .vcpu_load   = kvmppc_core_vcpu_load_hv,
5527         .vcpu_put    = kvmppc_core_vcpu_put_hv,
5528         .inject_interrupt = kvmppc_inject_interrupt_hv,
5529         .set_msr     = kvmppc_set_msr_hv,
5530         .vcpu_run    = kvmppc_vcpu_run_hv,
5531         .vcpu_create = kvmppc_core_vcpu_create_hv,
5532         .vcpu_free   = kvmppc_core_vcpu_free_hv,
5533         .check_requests = kvmppc_core_check_requests_hv,
5534         .get_dirty_log  = kvm_vm_ioctl_get_dirty_log_hv,
5535         .flush_memslot  = kvmppc_core_flush_memslot_hv,
5536         .prepare_memory_region = kvmppc_core_prepare_memory_region_hv,
5537         .commit_memory_region  = kvmppc_core_commit_memory_region_hv,
5538         .unmap_hva_range = kvm_unmap_hva_range_hv,
5539         .age_hva  = kvm_age_hva_hv,
5540         .test_age_hva = kvm_test_age_hva_hv,
5541         .set_spte_hva = kvm_set_spte_hva_hv,
5542         .mmu_destroy  = kvmppc_mmu_destroy_hv,
5543         .free_memslot = kvmppc_core_free_memslot_hv,
5544         .create_memslot = kvmppc_core_create_memslot_hv,
5545         .init_vm =  kvmppc_core_init_vm_hv,
5546         .destroy_vm = kvmppc_core_destroy_vm_hv,
5547         .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv,
5548         .emulate_op = kvmppc_core_emulate_op_hv,
5549         .emulate_mtspr = kvmppc_core_emulate_mtspr_hv,
5550         .emulate_mfspr = kvmppc_core_emulate_mfspr_hv,
5551         .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv,
5552         .arch_vm_ioctl  = kvm_arch_vm_ioctl_hv,
5553         .hcall_implemented = kvmppc_hcall_impl_hv,
5554 #ifdef CONFIG_KVM_XICS
5555         .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv,
5556         .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv,
5557 #endif
5558         .configure_mmu = kvmhv_configure_mmu,
5559         .get_rmmu_info = kvmhv_get_rmmu_info,
5560         .set_smt_mode = kvmhv_set_smt_mode,
5561         .enable_nested = kvmhv_enable_nested,
5562         .load_from_eaddr = kvmhv_load_from_eaddr,
5563         .store_to_eaddr = kvmhv_store_to_eaddr,
5564         .svm_off = kvmhv_svm_off,
5565 };
5566
5567 static int kvm_init_subcore_bitmap(void)
5568 {
5569         int i, j;
5570         int nr_cores = cpu_nr_cores();
5571         struct sibling_subcore_state *sibling_subcore_state;
5572
5573         for (i = 0; i < nr_cores; i++) {
5574                 int first_cpu = i * threads_per_core;
5575                 int node = cpu_to_node(first_cpu);
5576
5577                 /* Ignore if it is already allocated. */
5578                 if (paca_ptrs[first_cpu]->sibling_subcore_state)
5579                         continue;
5580
5581                 sibling_subcore_state =
5582                         kzalloc_node(sizeof(struct sibling_subcore_state),
5583                                                         GFP_KERNEL, node);
5584                 if (!sibling_subcore_state)
5585                         return -ENOMEM;
5586
5587
5588                 for (j = 0; j < threads_per_core; j++) {
5589                         int cpu = first_cpu + j;
5590
5591                         paca_ptrs[cpu]->sibling_subcore_state =
5592                                                 sibling_subcore_state;
5593                 }
5594         }
5595         return 0;
5596 }
5597
5598 static int kvmppc_radix_possible(void)
5599 {
5600         return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled();
5601 }
5602
5603 static int kvmppc_book3s_init_hv(void)
5604 {
5605         int r;
5606
5607         if (!tlbie_capable) {
5608                 pr_err("KVM-HV: Host does not support TLBIE\n");
5609                 return -ENODEV;
5610         }
5611
5612         /*
5613          * FIXME!! Do we need to check on all cpus ?
5614          */
5615         r = kvmppc_core_check_processor_compat_hv();
5616         if (r < 0)
5617                 return -ENODEV;
5618
5619         r = kvmhv_nested_init();
5620         if (r)
5621                 return r;
5622
5623         r = kvm_init_subcore_bitmap();
5624         if (r)
5625                 return r;
5626
5627         /*
5628          * We need a way of accessing the XICS interrupt controller,
5629          * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or
5630          * indirectly, via OPAL.
5631          */
5632 #ifdef CONFIG_SMP
5633         if (!xics_on_xive() && !kvmhv_on_pseries() &&
5634             !local_paca->kvm_hstate.xics_phys) {
5635                 struct device_node *np;
5636
5637                 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc");
5638                 if (!np) {
5639                         pr_err("KVM-HV: Cannot determine method for accessing XICS\n");
5640                         return -ENODEV;
5641                 }
5642                 /* presence of intc confirmed - node can be dropped again */
5643                 of_node_put(np);
5644         }
5645 #endif
5646
5647         kvm_ops_hv.owner = THIS_MODULE;
5648         kvmppc_hv_ops = &kvm_ops_hv;
5649
5650         init_default_hcalls();
5651
5652         init_vcore_lists();
5653
5654         r = kvmppc_mmu_hv_init();
5655         if (r)
5656                 return r;
5657
5658         if (kvmppc_radix_possible())
5659                 r = kvmppc_radix_init();
5660
5661         /*
5662          * POWER9 chips before version 2.02 can't have some threads in
5663          * HPT mode and some in radix mode on the same core.
5664          */
5665         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
5666                 unsigned int pvr = mfspr(SPRN_PVR);
5667                 if ((pvr >> 16) == PVR_POWER9 &&
5668                     (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) ||
5669                      ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101)))
5670                         no_mixing_hpt_and_radix = true;
5671         }
5672
5673         r = kvmppc_uvmem_init();
5674         if (r < 0)
5675                 pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r);
5676
5677         return r;
5678 }
5679
5680 static void kvmppc_book3s_exit_hv(void)
5681 {
5682         kvmppc_uvmem_free();
5683         kvmppc_free_host_rm_ops();
5684         if (kvmppc_radix_possible())
5685                 kvmppc_radix_exit();
5686         kvmppc_hv_ops = NULL;
5687         kvmhv_nested_exit();
5688 }
5689
5690 module_init(kvmppc_book3s_init_hv);
5691 module_exit(kvmppc_book3s_exit_hv);
5692 MODULE_LICENSE("GPL");
5693 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5694 MODULE_ALIAS("devname:kvm");