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