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