root/arch/powerpc/kvm/book3s_hv.c

/* [<][>][^][v][top][bottom][index][help] */

DEFINITIONS

This source file includes following definitions.
  1. nesting_enabled
  2. next_runnable_thread
  3. kvmppc_ipi_thread
  4. kvmppc_fast_vcpu_kick_hv
  5. kvmppc_core_start_stolen
  6. kvmppc_core_end_stolen
  7. kvmppc_core_vcpu_load_hv
  8. kvmppc_core_vcpu_put_hv
  9. kvmppc_set_msr_hv
  10. kvmppc_set_pvr_hv
  11. kvmppc_set_arch_compat
  12. kvmppc_dump_regs
  13. kvmppc_find_vcpu
  14. init_vpa
  15. set_vpa
  16. vpa_is_registered
  17. do_h_register_vpa
  18. kvmppc_update_vpa
  19. kvmppc_update_vpas
  20. vcore_stolen_time
  21. kvmppc_create_dtl_entry
  22. kvmppc_doorbell_pending
  23. kvmppc_power8_compatible
  24. kvmppc_h_set_mode
  25. kvmppc_copy_guest
  26. kvmppc_h_page_init
  27. kvm_arch_vcpu_yield_to
  28. kvmppc_get_yield_count
  29. kvmppc_pseries_do_hcall
  30. kvmppc_nested_cede
  31. kvmppc_hcall_impl_hv
  32. kvmppc_emulate_debug_inst
  33. do_nothing
  34. kvmppc_read_dpdes
  35. kvmppc_emulate_doorbell_instr
  36. kvmppc_handle_exit_hv
  37. kvmppc_handle_nested_exit
  38. kvm_arch_vcpu_ioctl_get_sregs_hv
  39. kvm_arch_vcpu_ioctl_set_sregs_hv
  40. kvmppc_set_lpcr
  41. kvmppc_get_one_reg_hv
  42. kvmppc_set_one_reg_hv
  43. threads_per_vcore
  44. kvmppc_vcore_create
  45. debugfs_timings_open
  46. debugfs_timings_release
  47. debugfs_timings_read
  48. debugfs_timings_write
  49. debugfs_vcpu_init
  50. debugfs_vcpu_init
  51. kvmppc_core_vcpu_create_hv
  52. kvmhv_set_smt_mode
  53. unpin_vpa
  54. kvmppc_core_vcpu_free_hv
  55. kvmppc_core_check_requests_hv
  56. kvmppc_set_timer
  57. kvmppc_end_cede
  58. kvmppc_remove_runnable
  59. kvmppc_grab_hwthread
  60. kvmppc_release_hwthread
  61. radix_flush_cpu
  62. kvmppc_prepare_radix_vcpu
  63. kvmppc_start_thread
  64. kvmppc_wait_for_nap
  65. on_primary_thread
  66. init_vcore_lists
  67. kvmppc_vcore_preempt
  68. kvmppc_vcore_end_preempt
  69. init_core_info
  70. subcore_config_ok
  71. init_vcore_to_run
  72. can_dynamic_split
  73. can_piggyback
  74. prepare_threads
  75. collect_piggybacks
  76. recheck_signals_and_mmu
  77. post_guest_process
  78. kvmppc_clear_host_core
  79. kvmppc_set_host_core
  80. set_irq_happened
  81. kvmppc_run_core
  82. kvmhv_load_hv_regs_and_go
  83. kvmhv_p9_guest_entry
  84. kvmppc_wait_for_exec
  85. grow_halt_poll_ns
  86. shrink_halt_poll_ns
  87. xive_interrupt_pending
  88. xive_interrupt_pending
  89. kvmppc_vcpu_woken
  90. kvmppc_vcore_check_block
  91. kvmppc_vcore_blocked
  92. kvmhv_setup_mmu
  93. kvmppc_run_vcpu
  94. kvmhv_run_single_vcpu
  95. kvmppc_vcpu_run_hv
  96. kvmppc_add_seg_page_size
  97. kvm_vm_ioctl_get_smmu_info_hv
  98. kvm_vm_ioctl_get_dirty_log_hv
  99. kvmppc_core_free_memslot_hv
  100. kvmppc_core_create_memslot_hv
  101. kvmppc_core_prepare_memory_region_hv
  102. kvmppc_core_commit_memory_region_hv
  103. kvmppc_update_lpcr
  104. kvmppc_mmu_destroy_hv
  105. kvmppc_setup_partition_table
  106. kvmppc_hv_setup_htab_rma
  107. kvmppc_switch_mmu_to_hpt
  108. kvmppc_switch_mmu_to_radix
  109. kvmppc_alloc_host_rm_ops
  110. kvmppc_free_host_rm_ops
  111. kvmppc_core_init_vm_hv
  112. kvmppc_free_vcores
  113. kvmppc_core_destroy_vm_hv
  114. kvmppc_core_emulate_op_hv
  115. kvmppc_core_emulate_mtspr_hv
  116. kvmppc_core_emulate_mfspr_hv
  117. kvmppc_core_check_processor_compat_hv
  118. kvmppc_free_pimap
  119. kvmppc_alloc_pimap
  120. kvmppc_set_passthru_irq
  121. kvmppc_clr_passthru_irq
  122. kvmppc_irq_bypass_add_producer_hv
  123. kvmppc_irq_bypass_del_producer_hv
  124. kvm_arch_vm_ioctl_hv
  125. init_default_hcalls
  126. kvmhv_configure_mmu
  127. kvmhv_enable_nested
  128. kvmhv_load_from_eaddr
  129. kvmhv_store_to_eaddr
  130. kvm_init_subcore_bitmap
  131. kvmppc_radix_possible
  132. kvmppc_book3s_init_hv
  133. kvmppc_book3s_exit_hv

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

/* [<][>][^][v][top][bottom][index][help] */