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/kvm_irqfd.h> 40 #include <linux/irqbypass.h> 41 #include <linux/module.h> 42 #include <linux/compiler.h> 43 #include <linux/of.h> 44 #include <linux/irqdomain.h> 45 #include <linux/smp.h> 46 47 #include <asm/ftrace.h> 48 #include <asm/reg.h> 49 #include <asm/ppc-opcode.h> 50 #include <asm/asm-prototypes.h> 51 #include <asm/archrandom.h> 52 #include <asm/debug.h> 53 #include <asm/disassemble.h> 54 #include <asm/cputable.h> 55 #include <asm/cacheflush.h> 56 #include <linux/uaccess.h> 57 #include <asm/interrupt.h> 58 #include <asm/io.h> 59 #include <asm/kvm_ppc.h> 60 #include <asm/kvm_book3s.h> 61 #include <asm/mmu_context.h> 62 #include <asm/lppaca.h> 63 #include <asm/pmc.h> 64 #include <asm/processor.h> 65 #include <asm/cputhreads.h> 66 #include <asm/page.h> 67 #include <asm/hvcall.h> 68 #include <asm/switch_to.h> 69 #include <asm/smp.h> 70 #include <asm/dbell.h> 71 #include <asm/hmi.h> 72 #include <asm/pnv-pci.h> 73 #include <asm/mmu.h> 74 #include <asm/opal.h> 75 #include <asm/xics.h> 76 #include <asm/xive.h> 77 #include <asm/hw_breakpoint.h> 78 #include <asm/kvm_book3s_uvmem.h> 79 #include <asm/ultravisor.h> 80 #include <asm/dtl.h> 81 #include <asm/plpar_wrappers.h> 82 83 #include <trace/events/ipi.h> 84 85 #include "book3s.h" 86 #include "book3s_hv.h" 87 88 #define CREATE_TRACE_POINTS 89 #include "trace_hv.h" 90 91 /* #define EXIT_DEBUG */ 92 /* #define EXIT_DEBUG_SIMPLE */ 93 /* #define EXIT_DEBUG_INT */ 94 95 /* Used to indicate that a guest page fault needs to be handled */ 96 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1) 97 /* Used to indicate that a guest passthrough interrupt needs to be handled */ 98 #define RESUME_PASSTHROUGH (RESUME_GUEST | RESUME_FLAG_ARCH2) 99 100 /* Used as a "null" value for timebase values */ 101 #define TB_NIL (~(u64)0) 102 103 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1); 104 105 static int dynamic_mt_modes = 6; 106 module_param(dynamic_mt_modes, int, 0644); 107 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)"); 108 static int target_smt_mode; 109 module_param(target_smt_mode, int, 0644); 110 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)"); 111 112 static bool one_vm_per_core; 113 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR); 114 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires POWER8 or older)"); 115 116 #ifdef CONFIG_KVM_XICS 117 static const struct kernel_param_ops module_param_ops = { 118 .set = param_set_int, 119 .get = param_get_int, 120 }; 121 122 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644); 123 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization"); 124 125 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644); 126 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core"); 127 #endif 128 129 /* If set, guests are allowed to create and control nested guests */ 130 static bool nested = true; 131 module_param(nested, bool, S_IRUGO | S_IWUSR); 132 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)"); 133 134 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu); 135 136 /* 137 * RWMR values for POWER8. These control the rate at which PURR 138 * and SPURR count and should be set according to the number of 139 * online threads in the vcore being run. 140 */ 141 #define RWMR_RPA_P8_1THREAD 0x164520C62609AECAUL 142 #define RWMR_RPA_P8_2THREAD 0x7FFF2908450D8DA9UL 143 #define RWMR_RPA_P8_3THREAD 0x164520C62609AECAUL 144 #define RWMR_RPA_P8_4THREAD 0x199A421245058DA9UL 145 #define RWMR_RPA_P8_5THREAD 0x164520C62609AECAUL 146 #define RWMR_RPA_P8_6THREAD 0x164520C62609AECAUL 147 #define RWMR_RPA_P8_7THREAD 0x164520C62609AECAUL 148 #define RWMR_RPA_P8_8THREAD 0x164520C62609AECAUL 149 150 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = { 151 RWMR_RPA_P8_1THREAD, 152 RWMR_RPA_P8_1THREAD, 153 RWMR_RPA_P8_2THREAD, 154 RWMR_RPA_P8_3THREAD, 155 RWMR_RPA_P8_4THREAD, 156 RWMR_RPA_P8_5THREAD, 157 RWMR_RPA_P8_6THREAD, 158 RWMR_RPA_P8_7THREAD, 159 RWMR_RPA_P8_8THREAD, 160 }; 161 162 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc, 163 int *ip) 164 { 165 int i = *ip; 166 struct kvm_vcpu *vcpu; 167 168 while (++i < MAX_SMT_THREADS) { 169 vcpu = READ_ONCE(vc->runnable_threads[i]); 170 if (vcpu) { 171 *ip = i; 172 return vcpu; 173 } 174 } 175 return NULL; 176 } 177 178 /* Used to traverse the list of runnable threads for a given vcore */ 179 #define for_each_runnable_thread(i, vcpu, vc) \ 180 for (i = -1; (vcpu = next_runnable_thread(vc, &i)); ) 181 182 static bool kvmppc_ipi_thread(int cpu) 183 { 184 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER); 185 186 /* If we're a nested hypervisor, fall back to ordinary IPIs for now */ 187 if (kvmhv_on_pseries()) 188 return false; 189 190 /* On POWER9 we can use msgsnd to IPI any cpu */ 191 if (cpu_has_feature(CPU_FTR_ARCH_300)) { 192 msg |= get_hard_smp_processor_id(cpu); 193 smp_mb(); 194 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg)); 195 return true; 196 } 197 198 /* On POWER8 for IPIs to threads in the same core, use msgsnd */ 199 if (cpu_has_feature(CPU_FTR_ARCH_207S)) { 200 preempt_disable(); 201 if (cpu_first_thread_sibling(cpu) == 202 cpu_first_thread_sibling(smp_processor_id())) { 203 msg |= cpu_thread_in_core(cpu); 204 smp_mb(); 205 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg)); 206 preempt_enable(); 207 return true; 208 } 209 preempt_enable(); 210 } 211 212 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP) 213 if (cpu >= 0 && cpu < nr_cpu_ids) { 214 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) { 215 xics_wake_cpu(cpu); 216 return true; 217 } 218 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY); 219 return true; 220 } 221 #endif 222 223 return false; 224 } 225 226 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu) 227 { 228 int cpu; 229 struct rcuwait *waitp; 230 231 /* 232 * rcuwait_wake_up contains smp_mb() which orders prior stores that 233 * create pending work vs below loads of cpu fields. The other side 234 * is the barrier in vcpu run that orders setting the cpu fields vs 235 * testing for pending work. 236 */ 237 238 waitp = kvm_arch_vcpu_get_wait(vcpu); 239 if (rcuwait_wake_up(waitp)) 240 ++vcpu->stat.generic.halt_wakeup; 241 242 cpu = READ_ONCE(vcpu->arch.thread_cpu); 243 if (cpu >= 0 && kvmppc_ipi_thread(cpu)) 244 return; 245 246 /* CPU points to the first thread of the core */ 247 cpu = vcpu->cpu; 248 if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu)) 249 smp_send_reschedule(cpu); 250 } 251 252 /* 253 * We use the vcpu_load/put functions to measure stolen time. 254 * 255 * Stolen time is counted as time when either the vcpu is able to 256 * run as part of a virtual core, but the task running the vcore 257 * is preempted or sleeping, or when the vcpu needs something done 258 * in the kernel by the task running the vcpu, but that task is 259 * preempted or sleeping. Those two things have to be counted 260 * separately, since one of the vcpu tasks will take on the job 261 * of running the core, and the other vcpu tasks in the vcore will 262 * sleep waiting for it to do that, but that sleep shouldn't count 263 * as stolen time. 264 * 265 * Hence we accumulate stolen time when the vcpu can run as part of 266 * a vcore using vc->stolen_tb, and the stolen time when the vcpu 267 * needs its task to do other things in the kernel (for example, 268 * service a page fault) in busy_stolen. We don't accumulate 269 * stolen time for a vcore when it is inactive, or for a vcpu 270 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of 271 * a misnomer; it means that the vcpu task is not executing in 272 * the KVM_RUN ioctl, i.e. it is in userspace or elsewhere in 273 * the kernel. We don't have any way of dividing up that time 274 * between time that the vcpu is genuinely stopped, time that 275 * the task is actively working on behalf of the vcpu, and time 276 * that the task is preempted, so we don't count any of it as 277 * stolen. 278 * 279 * Updates to busy_stolen are protected by arch.tbacct_lock; 280 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock 281 * lock. The stolen times are measured in units of timebase ticks. 282 * (Note that the != TB_NIL checks below are purely defensive; 283 * they should never fail.) 284 * 285 * The POWER9 path is simpler, one vcpu per virtual core so the 286 * former case does not exist. If a vcpu is preempted when it is 287 * BUSY_IN_HOST and not ceded or otherwise blocked, then accumulate 288 * the stolen cycles in busy_stolen. RUNNING is not a preemptible 289 * state in the P9 path. 290 */ 291 292 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc, u64 tb) 293 { 294 unsigned long flags; 295 296 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)); 297 298 spin_lock_irqsave(&vc->stoltb_lock, flags); 299 vc->preempt_tb = tb; 300 spin_unlock_irqrestore(&vc->stoltb_lock, flags); 301 } 302 303 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc, u64 tb) 304 { 305 unsigned long flags; 306 307 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)); 308 309 spin_lock_irqsave(&vc->stoltb_lock, flags); 310 if (vc->preempt_tb != TB_NIL) { 311 vc->stolen_tb += tb - vc->preempt_tb; 312 vc->preempt_tb = TB_NIL; 313 } 314 spin_unlock_irqrestore(&vc->stoltb_lock, flags); 315 } 316 317 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu) 318 { 319 struct kvmppc_vcore *vc = vcpu->arch.vcore; 320 unsigned long flags; 321 u64 now; 322 323 if (cpu_has_feature(CPU_FTR_ARCH_300)) { 324 if (vcpu->arch.busy_preempt != TB_NIL) { 325 WARN_ON_ONCE(vcpu->arch.state != KVMPPC_VCPU_BUSY_IN_HOST); 326 vc->stolen_tb += mftb() - vcpu->arch.busy_preempt; 327 vcpu->arch.busy_preempt = TB_NIL; 328 } 329 return; 330 } 331 332 now = mftb(); 333 334 /* 335 * We can test vc->runner without taking the vcore lock, 336 * because only this task ever sets vc->runner to this 337 * vcpu, and once it is set to this vcpu, only this task 338 * ever sets it to NULL. 339 */ 340 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING) 341 kvmppc_core_end_stolen(vc, now); 342 343 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags); 344 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST && 345 vcpu->arch.busy_preempt != TB_NIL) { 346 vcpu->arch.busy_stolen += now - vcpu->arch.busy_preempt; 347 vcpu->arch.busy_preempt = TB_NIL; 348 } 349 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags); 350 } 351 352 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu) 353 { 354 struct kvmppc_vcore *vc = vcpu->arch.vcore; 355 unsigned long flags; 356 u64 now; 357 358 if (cpu_has_feature(CPU_FTR_ARCH_300)) { 359 /* 360 * In the P9 path, RUNNABLE is not preemptible 361 * (nor takes host interrupts) 362 */ 363 WARN_ON_ONCE(vcpu->arch.state == KVMPPC_VCPU_RUNNABLE); 364 /* 365 * Account stolen time when preempted while the vcpu task is 366 * running in the kernel (but not in qemu, which is INACTIVE). 367 */ 368 if (task_is_running(current) && 369 vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST) 370 vcpu->arch.busy_preempt = mftb(); 371 return; 372 } 373 374 now = mftb(); 375 376 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING) 377 kvmppc_core_start_stolen(vc, now); 378 379 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags); 380 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST) 381 vcpu->arch.busy_preempt = now; 382 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags); 383 } 384 385 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr) 386 { 387 vcpu->arch.pvr = pvr; 388 } 389 390 /* Dummy value used in computing PCR value below */ 391 #define PCR_ARCH_32 (PCR_ARCH_31 << 1) 392 393 static inline unsigned long map_pcr_to_cap(unsigned long pcr) 394 { 395 unsigned long cap = 0; 396 397 switch (pcr) { 398 case PCR_ARCH_300: 399 cap = H_GUEST_CAP_POWER9; 400 break; 401 case PCR_ARCH_31: 402 if (cpu_has_feature(CPU_FTR_P11_PVR)) 403 cap = H_GUEST_CAP_POWER11; 404 else 405 cap = H_GUEST_CAP_POWER10; 406 break; 407 default: 408 break; 409 } 410 411 return cap; 412 } 413 414 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat) 415 { 416 unsigned long host_pcr_bit = 0, guest_pcr_bit = 0, cap = 0; 417 struct kvmppc_vcore *vc = vcpu->arch.vcore; 418 419 /* We can (emulate) our own architecture version and anything older */ 420 if (cpu_has_feature(CPU_FTR_ARCH_32)) 421 host_pcr_bit = PCR_ARCH_32; 422 else if (cpu_has_feature(CPU_FTR_P11_PVR) || cpu_has_feature(CPU_FTR_ARCH_31)) 423 host_pcr_bit = PCR_ARCH_31; 424 else if (cpu_has_feature(CPU_FTR_ARCH_300)) 425 host_pcr_bit = PCR_ARCH_300; 426 else if (cpu_has_feature(CPU_FTR_ARCH_207S)) 427 host_pcr_bit = PCR_ARCH_207; 428 else if (cpu_has_feature(CPU_FTR_ARCH_206)) 429 host_pcr_bit = PCR_ARCH_206; 430 else 431 host_pcr_bit = PCR_ARCH_205; 432 433 /* Determine lowest PCR bit needed to run guest in given PVR level */ 434 guest_pcr_bit = host_pcr_bit; 435 if (arch_compat) { 436 switch (arch_compat) { 437 case PVR_ARCH_205: 438 guest_pcr_bit = PCR_ARCH_205; 439 break; 440 case PVR_ARCH_206: 441 case PVR_ARCH_206p: 442 guest_pcr_bit = PCR_ARCH_206; 443 break; 444 case PVR_ARCH_207: 445 guest_pcr_bit = PCR_ARCH_207; 446 break; 447 case PVR_ARCH_300: 448 guest_pcr_bit = PCR_ARCH_300; 449 break; 450 case PVR_ARCH_31: 451 guest_pcr_bit = PCR_ARCH_31; 452 break; 453 case PVR_ARCH_31_P11: 454 /* 455 * Need to check this for ISA 3.1, as Power10 and 456 * Power11 share the same PCR. For any subsequent ISA 457 * versions, this will be taken care of by the guest vs 458 * host PCR comparison below. 459 */ 460 if (!cpu_has_feature(CPU_FTR_P11_PVR)) { 461 arch_compat = PVR_ARCH_INVALID; 462 goto out; 463 } 464 guest_pcr_bit = PCR_ARCH_31; 465 break; 466 case PVR_ARCH_32: 467 guest_pcr_bit = PCR_ARCH_32; 468 break; 469 default: 470 return -EINVAL; 471 } 472 } 473 474 /* Check requested PCR bits don't exceed our capabilities */ 475 if (guest_pcr_bit > host_pcr_bit) 476 return -EINVAL; 477 478 if (kvmhv_on_pseries() && kvmhv_is_nestedv2()) { 479 /* 480 * 'arch_compat == 0' would mean the guest should default to 481 * L1's compatibility. In this case, the guest would pick 482 * host's PCR and evaluate the corresponding capabilities. 483 */ 484 cap = map_pcr_to_cap(guest_pcr_bit); 485 if (!(cap & nested_capabilities)) 486 return -EINVAL; 487 } 488 489 out: 490 spin_lock(&vc->lock); 491 vc->arch_compat = arch_compat; 492 kvmhv_nestedv2_mark_dirty(vcpu, KVMPPC_GSID_LOGICAL_PVR); 493 /* 494 * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit 495 * Also set all reserved PCR bits 496 */ 497 vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK; 498 spin_unlock(&vc->lock); 499 500 return kvmppc_sanity_check(vcpu); 501 } 502 503 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu) 504 { 505 int r; 506 507 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id); 508 pr_err("pc = %.16lx msr = %.16llx trap = %x\n", 509 vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap); 510 for (r = 0; r < 16; ++r) 511 pr_err("r%2d = %.16lx r%d = %.16lx\n", 512 r, kvmppc_get_gpr(vcpu, r), 513 r+16, kvmppc_get_gpr(vcpu, r+16)); 514 pr_err("ctr = %.16lx lr = %.16lx\n", 515 vcpu->arch.regs.ctr, vcpu->arch.regs.link); 516 pr_err("srr0 = %.16llx srr1 = %.16llx\n", 517 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1); 518 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n", 519 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1); 520 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n", 521 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3); 522 pr_err("cr = %.8lx xer = %.16lx dsisr = %.8x\n", 523 vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr); 524 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar); 525 pr_err("fault dar = %.16lx dsisr = %.8x\n", 526 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr); 527 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max); 528 for (r = 0; r < vcpu->arch.slb_max; ++r) 529 pr_err(" ESID = %.16llx VSID = %.16llx\n", 530 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv); 531 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.16lx\n", 532 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1, 533 vcpu->arch.last_inst); 534 } 535 536 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id) 537 { 538 return kvm_get_vcpu_by_id(kvm, id); 539 } 540 541 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa) 542 { 543 vpa->__old_status |= LPPACA_OLD_SHARED_PROC; 544 vpa->yield_count = cpu_to_be32(1); 545 } 546 547 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v, 548 unsigned long addr, unsigned long len) 549 { 550 /* check address is cacheline aligned */ 551 if (addr & (L1_CACHE_BYTES - 1)) 552 return -EINVAL; 553 spin_lock(&vcpu->arch.vpa_update_lock); 554 if (v->next_gpa != addr || v->len != len) { 555 v->next_gpa = addr; 556 v->len = addr ? len : 0; 557 v->update_pending = 1; 558 } 559 spin_unlock(&vcpu->arch.vpa_update_lock); 560 return 0; 561 } 562 563 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */ 564 struct reg_vpa { 565 u32 dummy; 566 union { 567 __be16 hword; 568 __be32 word; 569 } length; 570 }; 571 572 static int vpa_is_registered(struct kvmppc_vpa *vpap) 573 { 574 if (vpap->update_pending) 575 return vpap->next_gpa != 0; 576 return vpap->pinned_addr != NULL; 577 } 578 579 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu, 580 unsigned long flags, 581 unsigned long vcpuid, unsigned long vpa) 582 { 583 struct kvm *kvm = vcpu->kvm; 584 unsigned long len, nb; 585 void *va; 586 struct kvm_vcpu *tvcpu; 587 int err; 588 int subfunc; 589 struct kvmppc_vpa *vpap; 590 591 tvcpu = kvmppc_find_vcpu(kvm, vcpuid); 592 if (!tvcpu) 593 return H_PARAMETER; 594 595 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK; 596 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL || 597 subfunc == H_VPA_REG_SLB) { 598 /* Registering new area - address must be cache-line aligned */ 599 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa) 600 return H_PARAMETER; 601 602 /* convert logical addr to kernel addr and read length */ 603 va = kvmppc_pin_guest_page(kvm, vpa, &nb); 604 if (va == NULL) 605 return H_PARAMETER; 606 if (subfunc == H_VPA_REG_VPA) 607 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword); 608 else 609 len = be32_to_cpu(((struct reg_vpa *)va)->length.word); 610 kvmppc_unpin_guest_page(kvm, va, vpa, false); 611 612 /* Check length */ 613 if (len > nb || len < sizeof(struct reg_vpa)) 614 return H_PARAMETER; 615 } else { 616 vpa = 0; 617 len = 0; 618 } 619 620 err = H_PARAMETER; 621 vpap = NULL; 622 spin_lock(&tvcpu->arch.vpa_update_lock); 623 624 switch (subfunc) { 625 case H_VPA_REG_VPA: /* register VPA */ 626 /* 627 * The size of our lppaca is 1kB because of the way we align 628 * it for the guest to avoid crossing a 4kB boundary. We only 629 * use 640 bytes of the structure though, so we should accept 630 * clients that set a size of 640. 631 */ 632 BUILD_BUG_ON(sizeof(struct lppaca) != 640); 633 if (len < sizeof(struct lppaca)) 634 break; 635 vpap = &tvcpu->arch.vpa; 636 err = 0; 637 break; 638 639 case H_VPA_REG_DTL: /* register DTL */ 640 if (len < sizeof(struct dtl_entry)) 641 break; 642 len -= len % sizeof(struct dtl_entry); 643 644 /* Check that they have previously registered a VPA */ 645 err = H_RESOURCE; 646 if (!vpa_is_registered(&tvcpu->arch.vpa)) 647 break; 648 649 vpap = &tvcpu->arch.dtl; 650 err = 0; 651 break; 652 653 case H_VPA_REG_SLB: /* register SLB shadow buffer */ 654 /* Check that they have previously registered a VPA */ 655 err = H_RESOURCE; 656 if (!vpa_is_registered(&tvcpu->arch.vpa)) 657 break; 658 659 vpap = &tvcpu->arch.slb_shadow; 660 err = 0; 661 break; 662 663 case H_VPA_DEREG_VPA: /* deregister VPA */ 664 /* Check they don't still have a DTL or SLB buf registered */ 665 err = H_RESOURCE; 666 if (vpa_is_registered(&tvcpu->arch.dtl) || 667 vpa_is_registered(&tvcpu->arch.slb_shadow)) 668 break; 669 670 vpap = &tvcpu->arch.vpa; 671 err = 0; 672 break; 673 674 case H_VPA_DEREG_DTL: /* deregister DTL */ 675 vpap = &tvcpu->arch.dtl; 676 err = 0; 677 break; 678 679 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */ 680 vpap = &tvcpu->arch.slb_shadow; 681 err = 0; 682 break; 683 } 684 685 if (vpap) { 686 vpap->next_gpa = vpa; 687 vpap->len = len; 688 vpap->update_pending = 1; 689 } 690 691 spin_unlock(&tvcpu->arch.vpa_update_lock); 692 693 return err; 694 } 695 696 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap, 697 struct kvmppc_vpa *old_vpap) 698 { 699 struct kvm *kvm = vcpu->kvm; 700 void *va; 701 unsigned long nb; 702 unsigned long gpa; 703 704 /* 705 * We need to pin the page pointed to by vpap->next_gpa, 706 * but we can't call kvmppc_pin_guest_page under the lock 707 * as it does get_user_pages() and down_read(). So we 708 * have to drop the lock, pin the page, then get the lock 709 * again and check that a new area didn't get registered 710 * in the meantime. 711 */ 712 for (;;) { 713 gpa = vpap->next_gpa; 714 spin_unlock(&vcpu->arch.vpa_update_lock); 715 va = NULL; 716 nb = 0; 717 if (gpa) 718 va = kvmppc_pin_guest_page(kvm, gpa, &nb); 719 spin_lock(&vcpu->arch.vpa_update_lock); 720 if (gpa == vpap->next_gpa) 721 break; 722 /* sigh... unpin that one and try again */ 723 if (va) 724 kvmppc_unpin_guest_page(kvm, va, gpa, false); 725 } 726 727 vpap->update_pending = 0; 728 if (va && nb < vpap->len) { 729 /* 730 * If it's now too short, it must be that userspace 731 * has changed the mappings underlying guest memory, 732 * so unregister the region. 733 */ 734 kvmppc_unpin_guest_page(kvm, va, gpa, false); 735 va = NULL; 736 } 737 *old_vpap = *vpap; 738 739 vpap->gpa = gpa; 740 vpap->pinned_addr = va; 741 vpap->dirty = false; 742 if (va) 743 vpap->pinned_end = va + vpap->len; 744 } 745 746 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu) 747 { 748 struct kvm *kvm = vcpu->kvm; 749 struct kvmppc_vpa old_vpa = { 0 }; 750 751 if (!(vcpu->arch.vpa.update_pending || 752 vcpu->arch.slb_shadow.update_pending || 753 vcpu->arch.dtl.update_pending)) 754 return; 755 756 spin_lock(&vcpu->arch.vpa_update_lock); 757 if (vcpu->arch.vpa.update_pending) { 758 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa, &old_vpa); 759 if (old_vpa.pinned_addr) { 760 if (kvmhv_is_nestedv2()) 761 kvmhv_nestedv2_set_vpa(vcpu, ~0ull); 762 kvmppc_unpin_guest_page(kvm, old_vpa.pinned_addr, old_vpa.gpa, 763 old_vpa.dirty); 764 } 765 if (vcpu->arch.vpa.pinned_addr) { 766 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr); 767 if (kvmhv_is_nestedv2()) 768 kvmhv_nestedv2_set_vpa(vcpu, __pa(vcpu->arch.vpa.pinned_addr)); 769 } 770 } 771 if (vcpu->arch.dtl.update_pending) { 772 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl, &old_vpa); 773 if (old_vpa.pinned_addr) 774 kvmppc_unpin_guest_page(kvm, old_vpa.pinned_addr, old_vpa.gpa, 775 old_vpa.dirty); 776 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr; 777 vcpu->arch.dtl_index = 0; 778 } 779 if (vcpu->arch.slb_shadow.update_pending) { 780 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow, &old_vpa); 781 if (old_vpa.pinned_addr) 782 kvmppc_unpin_guest_page(kvm, old_vpa.pinned_addr, old_vpa.gpa, 783 old_vpa.dirty); 784 } 785 786 spin_unlock(&vcpu->arch.vpa_update_lock); 787 } 788 789 /* 790 * Return the accumulated stolen time for the vcore up until `now'. 791 * The caller should hold the vcore lock. 792 */ 793 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now) 794 { 795 u64 p; 796 unsigned long flags; 797 798 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)); 799 800 spin_lock_irqsave(&vc->stoltb_lock, flags); 801 p = vc->stolen_tb; 802 if (vc->vcore_state != VCORE_INACTIVE && 803 vc->preempt_tb != TB_NIL) 804 p += now - vc->preempt_tb; 805 spin_unlock_irqrestore(&vc->stoltb_lock, flags); 806 return p; 807 } 808 809 static void __kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu, 810 struct lppaca *vpa, 811 unsigned int pcpu, u64 now, 812 unsigned long stolen) 813 { 814 struct dtl_entry *dt; 815 816 dt = vcpu->arch.dtl_ptr; 817 818 if (!dt) 819 return; 820 821 dt->dispatch_reason = 7; 822 dt->preempt_reason = 0; 823 dt->processor_id = cpu_to_be16(pcpu + vcpu->arch.ptid); 824 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen); 825 dt->ready_to_enqueue_time = 0; 826 dt->waiting_to_ready_time = 0; 827 dt->timebase = cpu_to_be64(now); 828 dt->fault_addr = 0; 829 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu)); 830 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr); 831 832 ++dt; 833 if (dt == vcpu->arch.dtl.pinned_end) 834 dt = vcpu->arch.dtl.pinned_addr; 835 vcpu->arch.dtl_ptr = dt; 836 /* order writing *dt vs. writing vpa->dtl_idx */ 837 smp_wmb(); 838 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index); 839 840 /* vcpu->arch.dtl.dirty is set by the caller */ 841 } 842 843 static void kvmppc_update_vpa_dispatch(struct kvm_vcpu *vcpu, 844 struct kvmppc_vcore *vc) 845 { 846 struct lppaca *vpa; 847 unsigned long stolen; 848 unsigned long core_stolen; 849 u64 now; 850 unsigned long flags; 851 852 vpa = vcpu->arch.vpa.pinned_addr; 853 if (!vpa) 854 return; 855 856 now = mftb(); 857 858 core_stolen = vcore_stolen_time(vc, now); 859 stolen = core_stolen - vcpu->arch.stolen_logged; 860 vcpu->arch.stolen_logged = core_stolen; 861 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags); 862 stolen += vcpu->arch.busy_stolen; 863 vcpu->arch.busy_stolen = 0; 864 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags); 865 866 vpa->enqueue_dispatch_tb = cpu_to_be64(be64_to_cpu(vpa->enqueue_dispatch_tb) + stolen); 867 868 __kvmppc_create_dtl_entry(vcpu, vpa, vc->pcpu, now + kvmppc_get_tb_offset(vcpu), stolen); 869 870 vcpu->arch.vpa.dirty = true; 871 } 872 873 static void kvmppc_update_vpa_dispatch_p9(struct kvm_vcpu *vcpu, 874 struct kvmppc_vcore *vc, 875 u64 now) 876 { 877 struct lppaca *vpa; 878 unsigned long stolen; 879 unsigned long stolen_delta; 880 881 vpa = vcpu->arch.vpa.pinned_addr; 882 if (!vpa) 883 return; 884 885 stolen = vc->stolen_tb; 886 stolen_delta = stolen - vcpu->arch.stolen_logged; 887 vcpu->arch.stolen_logged = stolen; 888 889 vpa->enqueue_dispatch_tb = cpu_to_be64(stolen); 890 891 __kvmppc_create_dtl_entry(vcpu, vpa, vc->pcpu, now, stolen_delta); 892 893 vcpu->arch.vpa.dirty = true; 894 } 895 896 /* See if there is a doorbell interrupt pending for a vcpu */ 897 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu) 898 { 899 int thr; 900 struct kvmppc_vcore *vc; 901 902 if (vcpu->arch.doorbell_request) 903 return true; 904 if (cpu_has_feature(CPU_FTR_ARCH_300)) 905 return false; 906 /* 907 * Ensure that the read of vcore->dpdes comes after the read 908 * of vcpu->doorbell_request. This barrier matches the 909 * smp_wmb() in kvmppc_guest_entry_inject(). 910 */ 911 smp_rmb(); 912 vc = vcpu->arch.vcore; 913 thr = vcpu->vcpu_id - vc->first_vcpuid; 914 return !!(vc->dpdes & (1 << thr)); 915 } 916 917 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu) 918 { 919 if (kvmppc_get_arch_compat(vcpu) >= PVR_ARCH_207) 920 return true; 921 if ((!kvmppc_get_arch_compat(vcpu)) && 922 cpu_has_feature(CPU_FTR_ARCH_207S)) 923 return true; 924 return false; 925 } 926 927 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags, 928 unsigned long resource, unsigned long value1, 929 unsigned long value2) 930 { 931 switch (resource) { 932 case H_SET_MODE_RESOURCE_SET_CIABR: 933 if (!kvmppc_power8_compatible(vcpu)) 934 return H_P2; 935 if (value2) 936 return H_P4; 937 if (mflags) 938 return H_UNSUPPORTED_FLAG_START; 939 /* Guests can't breakpoint the hypervisor */ 940 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER) 941 return H_P3; 942 kvmppc_set_ciabr_hv(vcpu, value1); 943 return H_SUCCESS; 944 case H_SET_MODE_RESOURCE_SET_DAWR0: 945 if (!kvmppc_power8_compatible(vcpu)) 946 return H_P2; 947 if (!ppc_breakpoint_available()) 948 return H_P2; 949 if (mflags) 950 return H_UNSUPPORTED_FLAG_START; 951 if (value2 & DABRX_HYP) 952 return H_P4; 953 kvmppc_set_dawr0_hv(vcpu, value1); 954 kvmppc_set_dawrx0_hv(vcpu, value2); 955 return H_SUCCESS; 956 case H_SET_MODE_RESOURCE_SET_DAWR1: 957 if (!kvmppc_power8_compatible(vcpu)) 958 return H_P2; 959 if (!ppc_breakpoint_available()) 960 return H_P2; 961 if (!cpu_has_feature(CPU_FTR_DAWR1)) 962 return H_P2; 963 if (!vcpu->kvm->arch.dawr1_enabled) 964 return H_FUNCTION; 965 if (mflags) 966 return H_UNSUPPORTED_FLAG_START; 967 if (value2 & DABRX_HYP) 968 return H_P4; 969 kvmppc_set_dawr1_hv(vcpu, value1); 970 kvmppc_set_dawrx1_hv(vcpu, value2); 971 return H_SUCCESS; 972 case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE: 973 /* 974 * KVM does not support mflags=2 (AIL=2) and AIL=1 is reserved. 975 * Keep this in synch with kvmppc_filter_guest_lpcr_hv. 976 */ 977 if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) && 978 kvmhv_vcpu_is_radix(vcpu) && mflags == 3) 979 return H_UNSUPPORTED_FLAG_START; 980 return H_TOO_HARD; 981 default: 982 return H_TOO_HARD; 983 } 984 } 985 986 /* Copy guest memory in place - must reside within a single memslot */ 987 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from, 988 unsigned long len) 989 { 990 struct kvm_memory_slot *to_memslot = NULL; 991 struct kvm_memory_slot *from_memslot = NULL; 992 unsigned long to_addr, from_addr; 993 int r; 994 995 /* Get HPA for from address */ 996 from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT); 997 if (!from_memslot) 998 return -EFAULT; 999 if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages) 1000 << PAGE_SHIFT)) 1001 return -EINVAL; 1002 from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT); 1003 if (kvm_is_error_hva(from_addr)) 1004 return -EFAULT; 1005 from_addr |= (from & (PAGE_SIZE - 1)); 1006 1007 /* Get HPA for to address */ 1008 to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT); 1009 if (!to_memslot) 1010 return -EFAULT; 1011 if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages) 1012 << PAGE_SHIFT)) 1013 return -EINVAL; 1014 to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT); 1015 if (kvm_is_error_hva(to_addr)) 1016 return -EFAULT; 1017 to_addr |= (to & (PAGE_SIZE - 1)); 1018 1019 /* Perform copy */ 1020 r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr, 1021 len); 1022 if (r) 1023 return -EFAULT; 1024 mark_page_dirty(kvm, to >> PAGE_SHIFT); 1025 return 0; 1026 } 1027 1028 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags, 1029 unsigned long dest, unsigned long src) 1030 { 1031 u64 pg_sz = SZ_4K; /* 4K page size */ 1032 u64 pg_mask = SZ_4K - 1; 1033 int ret; 1034 1035 /* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */ 1036 if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE | 1037 H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED)) 1038 return H_PARAMETER; 1039 1040 /* dest (and src if copy_page flag set) must be page aligned */ 1041 if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask))) 1042 return H_PARAMETER; 1043 1044 /* zero and/or copy the page as determined by the flags */ 1045 if (flags & H_COPY_PAGE) { 1046 ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz); 1047 if (ret < 0) 1048 return H_PARAMETER; 1049 } else if (flags & H_ZERO_PAGE) { 1050 ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz); 1051 if (ret < 0) 1052 return H_PARAMETER; 1053 } 1054 1055 /* We can ignore the remaining flags */ 1056 1057 return H_SUCCESS; 1058 } 1059 1060 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target) 1061 { 1062 struct kvmppc_vcore *vcore = target->arch.vcore; 1063 1064 /* 1065 * We expect to have been called by the real mode handler 1066 * (kvmppc_rm_h_confer()) which would have directly returned 1067 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may 1068 * have useful work to do and should not confer) so we don't 1069 * recheck that here. 1070 * 1071 * In the case of the P9 single vcpu per vcore case, the real 1072 * mode handler is not called but no other threads are in the 1073 * source vcore. 1074 */ 1075 if (!cpu_has_feature(CPU_FTR_ARCH_300)) { 1076 spin_lock(&vcore->lock); 1077 if (target->arch.state == KVMPPC_VCPU_RUNNABLE && 1078 vcore->vcore_state != VCORE_INACTIVE && 1079 vcore->runner) 1080 target = vcore->runner; 1081 spin_unlock(&vcore->lock); 1082 } 1083 1084 return kvm_vcpu_yield_to(target); 1085 } 1086 1087 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu) 1088 { 1089 int yield_count = 0; 1090 struct lppaca *lppaca; 1091 1092 spin_lock(&vcpu->arch.vpa_update_lock); 1093 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr; 1094 if (lppaca) 1095 yield_count = be32_to_cpu(lppaca->yield_count); 1096 spin_unlock(&vcpu->arch.vpa_update_lock); 1097 return yield_count; 1098 } 1099 1100 /* 1101 * H_RPT_INVALIDATE hcall handler for nested guests. 1102 * 1103 * Handles only nested process-scoped invalidation requests in L0. 1104 */ 1105 static int kvmppc_nested_h_rpt_invalidate(struct kvm_vcpu *vcpu) 1106 { 1107 unsigned long type = kvmppc_get_gpr(vcpu, 6); 1108 unsigned long pid, pg_sizes, start, end; 1109 1110 /* 1111 * The partition-scoped invalidations aren't handled here in L0. 1112 */ 1113 if (type & H_RPTI_TYPE_NESTED) 1114 return RESUME_HOST; 1115 1116 pid = kvmppc_get_gpr(vcpu, 4); 1117 pg_sizes = kvmppc_get_gpr(vcpu, 7); 1118 start = kvmppc_get_gpr(vcpu, 8); 1119 end = kvmppc_get_gpr(vcpu, 9); 1120 1121 do_h_rpt_invalidate_prt(pid, vcpu->arch.nested->shadow_lpid, 1122 type, pg_sizes, start, end); 1123 1124 kvmppc_set_gpr(vcpu, 3, H_SUCCESS); 1125 return RESUME_GUEST; 1126 } 1127 1128 static long kvmppc_h_rpt_invalidate(struct kvm_vcpu *vcpu, 1129 unsigned long id, unsigned long target, 1130 unsigned long type, unsigned long pg_sizes, 1131 unsigned long start, unsigned long end) 1132 { 1133 if (!kvm_is_radix(vcpu->kvm)) 1134 return H_UNSUPPORTED; 1135 1136 if (end < start) 1137 return H_P5; 1138 1139 /* 1140 * Partition-scoped invalidation for nested guests. 1141 */ 1142 if (type & H_RPTI_TYPE_NESTED) { 1143 if (!nesting_enabled(vcpu->kvm)) 1144 return H_FUNCTION; 1145 1146 /* Support only cores as target */ 1147 if (target != H_RPTI_TARGET_CMMU) 1148 return H_P2; 1149 1150 return do_h_rpt_invalidate_pat(vcpu, id, type, pg_sizes, 1151 start, end); 1152 } 1153 1154 /* 1155 * Process-scoped invalidation for L1 guests. 1156 */ 1157 do_h_rpt_invalidate_prt(id, vcpu->kvm->arch.lpid, 1158 type, pg_sizes, start, end); 1159 return H_SUCCESS; 1160 } 1161 1162 static long kvmppc_pseries_do_hpt_hcall(struct kvm_vcpu *vcpu, unsigned long req) 1163 { 1164 switch (req) { 1165 case H_REMOVE: 1166 return kvmppc_h_remove(vcpu, kvmppc_get_gpr(vcpu, 4), 1167 kvmppc_get_gpr(vcpu, 5), 1168 kvmppc_get_gpr(vcpu, 6)); 1169 case H_ENTER: 1170 return kvmppc_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4), 1171 kvmppc_get_gpr(vcpu, 5), 1172 kvmppc_get_gpr(vcpu, 6), 1173 kvmppc_get_gpr(vcpu, 7)); 1174 case H_READ: 1175 return kvmppc_h_read(vcpu, kvmppc_get_gpr(vcpu, 4), 1176 kvmppc_get_gpr(vcpu, 5)); 1177 case H_CLEAR_MOD: 1178 return kvmppc_h_clear_mod(vcpu, kvmppc_get_gpr(vcpu, 4), 1179 kvmppc_get_gpr(vcpu, 5)); 1180 case H_CLEAR_REF: 1181 return kvmppc_h_clear_ref(vcpu, kvmppc_get_gpr(vcpu, 4), 1182 kvmppc_get_gpr(vcpu, 5)); 1183 case H_PROTECT: 1184 return kvmppc_h_protect(vcpu, kvmppc_get_gpr(vcpu, 4), 1185 kvmppc_get_gpr(vcpu, 5), 1186 kvmppc_get_gpr(vcpu, 6)); 1187 case H_BULK_REMOVE: 1188 return kvmppc_h_bulk_remove(vcpu); 1189 } 1190 1191 return H_FUNCTION; 1192 } 1193 1194 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu) 1195 { 1196 struct kvm *kvm = vcpu->kvm; 1197 unsigned long req = kvmppc_get_gpr(vcpu, 3); 1198 unsigned long target, ret = H_SUCCESS; 1199 int yield_count; 1200 struct kvm_vcpu *tvcpu; 1201 int idx, rc; 1202 1203 if (req <= MAX_HCALL_OPCODE && 1204 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls)) 1205 return RESUME_HOST; 1206 1207 switch (req) { 1208 case H_REMOVE: 1209 case H_ENTER: 1210 case H_READ: 1211 case H_CLEAR_MOD: 1212 case H_CLEAR_REF: 1213 case H_PROTECT: 1214 case H_BULK_REMOVE: 1215 preempt_disable(); 1216 idx = srcu_read_lock(&kvm->srcu); 1217 ret = kvmppc_pseries_do_hpt_hcall(vcpu, req); 1218 srcu_read_unlock(&kvm->srcu, idx); 1219 preempt_enable(); 1220 if (ret == H_TOO_HARD) 1221 return RESUME_HOST; 1222 break; 1223 1224 case H_CEDE: 1225 break; 1226 case H_PROD: 1227 target = kvmppc_get_gpr(vcpu, 4); 1228 tvcpu = kvmppc_find_vcpu(kvm, target); 1229 if (!tvcpu) { 1230 ret = H_PARAMETER; 1231 break; 1232 } 1233 tvcpu->arch.prodded = 1; 1234 smp_mb(); /* This orders prodded store vs ceded load */ 1235 if (tvcpu->arch.ceded) 1236 kvmppc_fast_vcpu_kick_hv(tvcpu); 1237 break; 1238 case H_CONFER: 1239 target = kvmppc_get_gpr(vcpu, 4); 1240 if (target == -1) 1241 break; 1242 tvcpu = kvmppc_find_vcpu(kvm, target); 1243 if (!tvcpu) { 1244 ret = H_PARAMETER; 1245 break; 1246 } 1247 yield_count = kvmppc_get_gpr(vcpu, 5); 1248 if (kvmppc_get_yield_count(tvcpu) != yield_count) 1249 break; 1250 kvm_arch_vcpu_yield_to(tvcpu); 1251 break; 1252 case H_REGISTER_VPA: 1253 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4), 1254 kvmppc_get_gpr(vcpu, 5), 1255 kvmppc_get_gpr(vcpu, 6)); 1256 break; 1257 case H_RTAS: 1258 if (list_empty(&kvm->arch.rtas_tokens)) 1259 return RESUME_HOST; 1260 1261 idx = srcu_read_lock(&kvm->srcu); 1262 rc = kvmppc_rtas_hcall(vcpu); 1263 srcu_read_unlock(&kvm->srcu, idx); 1264 1265 if (rc == -ENOENT) 1266 return RESUME_HOST; 1267 else if (rc == 0) 1268 break; 1269 1270 /* Send the error out to userspace via KVM_RUN */ 1271 return rc; 1272 case H_LOGICAL_CI_LOAD: 1273 ret = kvmppc_h_logical_ci_load(vcpu); 1274 if (ret == H_TOO_HARD) 1275 return RESUME_HOST; 1276 break; 1277 case H_LOGICAL_CI_STORE: 1278 ret = kvmppc_h_logical_ci_store(vcpu); 1279 if (ret == H_TOO_HARD) 1280 return RESUME_HOST; 1281 break; 1282 case H_SET_MODE: 1283 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4), 1284 kvmppc_get_gpr(vcpu, 5), 1285 kvmppc_get_gpr(vcpu, 6), 1286 kvmppc_get_gpr(vcpu, 7)); 1287 if (ret == H_TOO_HARD) 1288 return RESUME_HOST; 1289 break; 1290 case H_XIRR: 1291 case H_CPPR: 1292 case H_EOI: 1293 case H_IPI: 1294 case H_IPOLL: 1295 case H_XIRR_X: 1296 if (kvmppc_xics_enabled(vcpu)) { 1297 if (xics_on_xive()) { 1298 ret = H_NOT_AVAILABLE; 1299 return RESUME_GUEST; 1300 } 1301 ret = kvmppc_xics_hcall(vcpu, req); 1302 break; 1303 } 1304 return RESUME_HOST; 1305 case H_SET_DABR: 1306 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4)); 1307 break; 1308 case H_SET_XDABR: 1309 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4), 1310 kvmppc_get_gpr(vcpu, 5)); 1311 break; 1312 #ifdef CONFIG_SPAPR_TCE_IOMMU 1313 case H_GET_TCE: 1314 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4), 1315 kvmppc_get_gpr(vcpu, 5)); 1316 if (ret == H_TOO_HARD) 1317 return RESUME_HOST; 1318 break; 1319 case H_PUT_TCE: 1320 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4), 1321 kvmppc_get_gpr(vcpu, 5), 1322 kvmppc_get_gpr(vcpu, 6)); 1323 if (ret == H_TOO_HARD) 1324 return RESUME_HOST; 1325 break; 1326 case H_PUT_TCE_INDIRECT: 1327 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4), 1328 kvmppc_get_gpr(vcpu, 5), 1329 kvmppc_get_gpr(vcpu, 6), 1330 kvmppc_get_gpr(vcpu, 7)); 1331 if (ret == H_TOO_HARD) 1332 return RESUME_HOST; 1333 break; 1334 case H_STUFF_TCE: 1335 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4), 1336 kvmppc_get_gpr(vcpu, 5), 1337 kvmppc_get_gpr(vcpu, 6), 1338 kvmppc_get_gpr(vcpu, 7)); 1339 if (ret == H_TOO_HARD) 1340 return RESUME_HOST; 1341 break; 1342 #endif 1343 case H_RANDOM: { 1344 unsigned long rand; 1345 1346 if (!arch_get_random_seed_longs(&rand, 1)) 1347 ret = H_HARDWARE; 1348 kvmppc_set_gpr(vcpu, 4, rand); 1349 break; 1350 } 1351 case H_RPT_INVALIDATE: 1352 ret = kvmppc_h_rpt_invalidate(vcpu, kvmppc_get_gpr(vcpu, 4), 1353 kvmppc_get_gpr(vcpu, 5), 1354 kvmppc_get_gpr(vcpu, 6), 1355 kvmppc_get_gpr(vcpu, 7), 1356 kvmppc_get_gpr(vcpu, 8), 1357 kvmppc_get_gpr(vcpu, 9)); 1358 break; 1359 1360 case H_SET_PARTITION_TABLE: 1361 ret = H_FUNCTION; 1362 if (nesting_enabled(kvm)) 1363 ret = kvmhv_set_partition_table(vcpu); 1364 break; 1365 case H_ENTER_NESTED: 1366 ret = H_FUNCTION; 1367 if (!nesting_enabled(kvm)) 1368 break; 1369 ret = kvmhv_enter_nested_guest(vcpu); 1370 if (ret == H_INTERRUPT) { 1371 kvmppc_set_gpr(vcpu, 3, 0); 1372 vcpu->arch.hcall_needed = 0; 1373 return -EINTR; 1374 } else if (ret == H_TOO_HARD) { 1375 kvmppc_set_gpr(vcpu, 3, 0); 1376 vcpu->arch.hcall_needed = 0; 1377 return RESUME_HOST; 1378 } 1379 break; 1380 case H_TLB_INVALIDATE: 1381 ret = H_FUNCTION; 1382 if (nesting_enabled(kvm)) 1383 ret = kvmhv_do_nested_tlbie(vcpu); 1384 break; 1385 case H_COPY_TOFROM_GUEST: 1386 ret = H_FUNCTION; 1387 if (nesting_enabled(kvm)) 1388 ret = kvmhv_copy_tofrom_guest_nested(vcpu); 1389 break; 1390 case H_PAGE_INIT: 1391 ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4), 1392 kvmppc_get_gpr(vcpu, 5), 1393 kvmppc_get_gpr(vcpu, 6)); 1394 break; 1395 case H_SVM_PAGE_IN: 1396 ret = H_UNSUPPORTED; 1397 if (kvmppc_get_srr1(vcpu) & MSR_S) 1398 ret = kvmppc_h_svm_page_in(kvm, 1399 kvmppc_get_gpr(vcpu, 4), 1400 kvmppc_get_gpr(vcpu, 5), 1401 kvmppc_get_gpr(vcpu, 6)); 1402 break; 1403 case H_SVM_PAGE_OUT: 1404 ret = H_UNSUPPORTED; 1405 if (kvmppc_get_srr1(vcpu) & MSR_S) 1406 ret = kvmppc_h_svm_page_out(kvm, 1407 kvmppc_get_gpr(vcpu, 4), 1408 kvmppc_get_gpr(vcpu, 5), 1409 kvmppc_get_gpr(vcpu, 6)); 1410 break; 1411 case H_SVM_INIT_START: 1412 ret = H_UNSUPPORTED; 1413 if (kvmppc_get_srr1(vcpu) & MSR_S) 1414 ret = kvmppc_h_svm_init_start(kvm); 1415 break; 1416 case H_SVM_INIT_DONE: 1417 ret = H_UNSUPPORTED; 1418 if (kvmppc_get_srr1(vcpu) & MSR_S) 1419 ret = kvmppc_h_svm_init_done(kvm); 1420 break; 1421 case H_SVM_INIT_ABORT: 1422 /* 1423 * Even if that call is made by the Ultravisor, the SSR1 value 1424 * is the guest context one, with the secure bit clear as it has 1425 * not yet been secured. So we can't check it here. 1426 * Instead the kvm->arch.secure_guest flag is checked inside 1427 * kvmppc_h_svm_init_abort(). 1428 */ 1429 ret = kvmppc_h_svm_init_abort(kvm); 1430 break; 1431 1432 default: 1433 return RESUME_HOST; 1434 } 1435 WARN_ON_ONCE(ret == H_TOO_HARD); 1436 kvmppc_set_gpr(vcpu, 3, ret); 1437 vcpu->arch.hcall_needed = 0; 1438 return RESUME_GUEST; 1439 } 1440 1441 /* 1442 * Handle H_CEDE in the P9 path where we don't call the real-mode hcall 1443 * handlers in book3s_hv_rmhandlers.S. 1444 * 1445 * This has to be done early, not in kvmppc_pseries_do_hcall(), so 1446 * that the cede logic in kvmppc_run_single_vcpu() works properly. 1447 */ 1448 static void kvmppc_cede(struct kvm_vcpu *vcpu) 1449 { 1450 __kvmppc_set_msr_hv(vcpu, __kvmppc_get_msr_hv(vcpu) | MSR_EE); 1451 vcpu->arch.ceded = 1; 1452 smp_mb(); 1453 if (vcpu->arch.prodded) { 1454 vcpu->arch.prodded = 0; 1455 smp_mb(); 1456 vcpu->arch.ceded = 0; 1457 } 1458 } 1459 1460 static int kvmppc_hcall_impl_hv(unsigned long cmd) 1461 { 1462 switch (cmd) { 1463 case H_CEDE: 1464 case H_PROD: 1465 case H_CONFER: 1466 case H_REGISTER_VPA: 1467 case H_SET_MODE: 1468 #ifdef CONFIG_SPAPR_TCE_IOMMU 1469 case H_GET_TCE: 1470 case H_PUT_TCE: 1471 case H_PUT_TCE_INDIRECT: 1472 case H_STUFF_TCE: 1473 #endif 1474 case H_LOGICAL_CI_LOAD: 1475 case H_LOGICAL_CI_STORE: 1476 #ifdef CONFIG_KVM_XICS 1477 case H_XIRR: 1478 case H_CPPR: 1479 case H_EOI: 1480 case H_IPI: 1481 case H_IPOLL: 1482 case H_XIRR_X: 1483 #endif 1484 case H_PAGE_INIT: 1485 case H_RPT_INVALIDATE: 1486 return 1; 1487 } 1488 1489 /* See if it's in the real-mode table */ 1490 return kvmppc_hcall_impl_hv_realmode(cmd); 1491 } 1492 1493 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu) 1494 { 1495 ppc_inst_t last_inst; 1496 1497 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) != 1498 EMULATE_DONE) { 1499 /* 1500 * Fetch failed, so return to guest and 1501 * try executing it again. 1502 */ 1503 return RESUME_GUEST; 1504 } 1505 1506 if (ppc_inst_val(last_inst) == KVMPPC_INST_SW_BREAKPOINT) { 1507 vcpu->run->exit_reason = KVM_EXIT_DEBUG; 1508 vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu); 1509 return RESUME_HOST; 1510 } else { 1511 kvmppc_core_queue_program(vcpu, SRR1_PROGILL | 1512 (kvmppc_get_msr(vcpu) & SRR1_PREFIXED)); 1513 return RESUME_GUEST; 1514 } 1515 } 1516 1517 static void do_nothing(void *x) 1518 { 1519 } 1520 1521 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu) 1522 { 1523 int thr, cpu, pcpu, nthreads; 1524 struct kvm_vcpu *v; 1525 unsigned long dpdes; 1526 1527 nthreads = vcpu->kvm->arch.emul_smt_mode; 1528 dpdes = 0; 1529 cpu = vcpu->vcpu_id & ~(nthreads - 1); 1530 for (thr = 0; thr < nthreads; ++thr, ++cpu) { 1531 v = kvmppc_find_vcpu(vcpu->kvm, cpu); 1532 if (!v) 1533 continue; 1534 /* 1535 * If the vcpu is currently running on a physical cpu thread, 1536 * interrupt it in order to pull it out of the guest briefly, 1537 * which will update its vcore->dpdes value. 1538 */ 1539 pcpu = READ_ONCE(v->cpu); 1540 if (pcpu >= 0) 1541 smp_call_function_single(pcpu, do_nothing, NULL, 1); 1542 if (kvmppc_doorbell_pending(v)) 1543 dpdes |= 1 << thr; 1544 } 1545 return dpdes; 1546 } 1547 1548 /* 1549 * On POWER9, emulate doorbell-related instructions in order to 1550 * give the guest the illusion of running on a multi-threaded core. 1551 * The instructions emulated are msgsndp, msgclrp, mfspr TIR, 1552 * and mfspr DPDES. 1553 */ 1554 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu) 1555 { 1556 u32 inst, rb, thr; 1557 unsigned long arg; 1558 struct kvm *kvm = vcpu->kvm; 1559 struct kvm_vcpu *tvcpu; 1560 ppc_inst_t pinst; 1561 1562 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &pinst) != EMULATE_DONE) 1563 return RESUME_GUEST; 1564 inst = ppc_inst_val(pinst); 1565 if (get_op(inst) != 31) 1566 return EMULATE_FAIL; 1567 rb = get_rb(inst); 1568 thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1); 1569 switch (get_xop(inst)) { 1570 case OP_31_XOP_MSGSNDP: 1571 arg = kvmppc_get_gpr(vcpu, rb); 1572 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER) 1573 break; 1574 arg &= 0x7f; 1575 if (arg >= kvm->arch.emul_smt_mode) 1576 break; 1577 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg); 1578 if (!tvcpu) 1579 break; 1580 if (!tvcpu->arch.doorbell_request) { 1581 tvcpu->arch.doorbell_request = 1; 1582 kvmppc_fast_vcpu_kick_hv(tvcpu); 1583 } 1584 break; 1585 case OP_31_XOP_MSGCLRP: 1586 arg = kvmppc_get_gpr(vcpu, rb); 1587 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER) 1588 break; 1589 vcpu->arch.vcore->dpdes = 0; 1590 vcpu->arch.doorbell_request = 0; 1591 break; 1592 case OP_31_XOP_MFSPR: 1593 switch (get_sprn(inst)) { 1594 case SPRN_TIR: 1595 arg = thr; 1596 break; 1597 case SPRN_DPDES: 1598 arg = kvmppc_read_dpdes(vcpu); 1599 break; 1600 default: 1601 return EMULATE_FAIL; 1602 } 1603 kvmppc_set_gpr(vcpu, get_rt(inst), arg); 1604 break; 1605 default: 1606 return EMULATE_FAIL; 1607 } 1608 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4); 1609 return RESUME_GUEST; 1610 } 1611 1612 /* 1613 * If the lppaca had pmcregs_in_use clear when we exited the guest, then 1614 * HFSCR_PM is cleared for next entry. If the guest then tries to access 1615 * the PMU SPRs, we get this facility unavailable interrupt. Putting HFSCR_PM 1616 * back in the guest HFSCR will cause the next entry to load the PMU SPRs and 1617 * allow the guest access to continue. 1618 */ 1619 static int kvmppc_pmu_unavailable(struct kvm_vcpu *vcpu) 1620 { 1621 if (!(vcpu->arch.hfscr_permitted & HFSCR_PM)) 1622 return EMULATE_FAIL; 1623 1624 kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_PM); 1625 1626 return RESUME_GUEST; 1627 } 1628 1629 static int kvmppc_ebb_unavailable(struct kvm_vcpu *vcpu) 1630 { 1631 if (!(vcpu->arch.hfscr_permitted & HFSCR_EBB)) 1632 return EMULATE_FAIL; 1633 1634 kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_EBB); 1635 1636 return RESUME_GUEST; 1637 } 1638 1639 static int kvmppc_tm_unavailable(struct kvm_vcpu *vcpu) 1640 { 1641 if (!(vcpu->arch.hfscr_permitted & HFSCR_TM)) 1642 return EMULATE_FAIL; 1643 1644 kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_TM); 1645 1646 return RESUME_GUEST; 1647 } 1648 1649 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu, 1650 struct task_struct *tsk) 1651 { 1652 struct kvm_run *run = vcpu->run; 1653 int r = RESUME_HOST; 1654 1655 vcpu->stat.sum_exits++; 1656 1657 /* 1658 * This can happen if an interrupt occurs in the last stages 1659 * of guest entry or the first stages of guest exit (i.e. after 1660 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV 1661 * and before setting it to KVM_GUEST_MODE_HOST_HV). 1662 * That can happen due to a bug, or due to a machine check 1663 * occurring at just the wrong time. 1664 */ 1665 if (!kvmhv_is_nestedv2() && (__kvmppc_get_msr_hv(vcpu) & MSR_HV)) { 1666 printk(KERN_EMERG "KVM trap in HV mode!\n"); 1667 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n", 1668 vcpu->arch.trap, kvmppc_get_pc(vcpu), 1669 vcpu->arch.shregs.msr); 1670 kvmppc_dump_regs(vcpu); 1671 run->exit_reason = KVM_EXIT_INTERNAL_ERROR; 1672 run->hw.hardware_exit_reason = vcpu->arch.trap; 1673 return RESUME_HOST; 1674 } 1675 run->exit_reason = KVM_EXIT_UNKNOWN; 1676 run->ready_for_interrupt_injection = 1; 1677 switch (vcpu->arch.trap) { 1678 /* We're good on these - the host merely wanted to get our attention */ 1679 case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER: 1680 WARN_ON_ONCE(1); /* Should never happen */ 1681 vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER; 1682 fallthrough; 1683 case BOOK3S_INTERRUPT_HV_DECREMENTER: 1684 vcpu->stat.dec_exits++; 1685 r = RESUME_GUEST; 1686 break; 1687 case BOOK3S_INTERRUPT_EXTERNAL: 1688 case BOOK3S_INTERRUPT_H_DOORBELL: 1689 case BOOK3S_INTERRUPT_H_VIRT: 1690 vcpu->stat.ext_intr_exits++; 1691 r = RESUME_GUEST; 1692 break; 1693 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/ 1694 case BOOK3S_INTERRUPT_HMI: 1695 case BOOK3S_INTERRUPT_PERFMON: 1696 case BOOK3S_INTERRUPT_SYSTEM_RESET: 1697 r = RESUME_GUEST; 1698 break; 1699 case BOOK3S_INTERRUPT_MACHINE_CHECK: { 1700 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL, 1701 DEFAULT_RATELIMIT_BURST); 1702 /* 1703 * Print the MCE event to host console. Ratelimit so the guest 1704 * can't flood the host log. 1705 */ 1706 if (__ratelimit(&rs)) 1707 machine_check_print_event_info(&vcpu->arch.mce_evt,false, true); 1708 1709 /* 1710 * If the guest can do FWNMI, exit to userspace so it can 1711 * deliver a FWNMI to the guest. 1712 * Otherwise we synthesize a machine check for the guest 1713 * so that it knows that the machine check occurred. 1714 */ 1715 if (!vcpu->kvm->arch.fwnmi_enabled) { 1716 ulong flags = (__kvmppc_get_msr_hv(vcpu) & 0x083c0000) | 1717 (kvmppc_get_msr(vcpu) & SRR1_PREFIXED); 1718 kvmppc_core_queue_machine_check(vcpu, flags); 1719 r = RESUME_GUEST; 1720 break; 1721 } 1722 1723 /* Exit to guest with KVM_EXIT_NMI as exit reason */ 1724 run->exit_reason = KVM_EXIT_NMI; 1725 run->hw.hardware_exit_reason = vcpu->arch.trap; 1726 /* Clear out the old NMI status from run->flags */ 1727 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK; 1728 /* Now set the NMI status */ 1729 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED) 1730 run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV; 1731 else 1732 run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV; 1733 1734 r = RESUME_HOST; 1735 break; 1736 } 1737 case BOOK3S_INTERRUPT_PROGRAM: 1738 { 1739 ulong flags; 1740 /* 1741 * Normally program interrupts are delivered directly 1742 * to the guest by the hardware, but we can get here 1743 * as a result of a hypervisor emulation interrupt 1744 * (e40) getting turned into a 700 by BML RTAS. 1745 */ 1746 flags = (__kvmppc_get_msr_hv(vcpu) & 0x1f0000ull) | 1747 (kvmppc_get_msr(vcpu) & SRR1_PREFIXED); 1748 kvmppc_core_queue_program(vcpu, flags); 1749 r = RESUME_GUEST; 1750 break; 1751 } 1752 case BOOK3S_INTERRUPT_SYSCALL: 1753 { 1754 int i; 1755 1756 if (!kvmhv_is_nestedv2() && unlikely(__kvmppc_get_msr_hv(vcpu) & MSR_PR)) { 1757 /* 1758 * Guest userspace executed sc 1. This can only be 1759 * reached by the P9 path because the old path 1760 * handles this case in realmode hcall handlers. 1761 */ 1762 if (!kvmhv_vcpu_is_radix(vcpu)) { 1763 /* 1764 * A guest could be running PR KVM, so this 1765 * may be a PR KVM hcall. It must be reflected 1766 * to the guest kernel as a sc interrupt. 1767 */ 1768 kvmppc_core_queue_syscall(vcpu); 1769 } else { 1770 /* 1771 * Radix guests can not run PR KVM or nested HV 1772 * hash guests which might run PR KVM, so this 1773 * is always a privilege fault. Send a program 1774 * check to guest kernel. 1775 */ 1776 kvmppc_core_queue_program(vcpu, SRR1_PROGPRIV); 1777 } 1778 r = RESUME_GUEST; 1779 break; 1780 } 1781 1782 /* 1783 * hcall - gather args and set exit_reason. This will next be 1784 * handled by kvmppc_pseries_do_hcall which may be able to deal 1785 * with it and resume guest, or may punt to userspace. 1786 */ 1787 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3); 1788 for (i = 0; i < 9; ++i) 1789 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i); 1790 run->exit_reason = KVM_EXIT_PAPR_HCALL; 1791 vcpu->arch.hcall_needed = 1; 1792 r = RESUME_HOST; 1793 break; 1794 } 1795 /* 1796 * We get these next two if the guest accesses a page which it thinks 1797 * it has mapped but which is not actually present, either because 1798 * it is for an emulated I/O device or because the corresonding 1799 * host page has been paged out. 1800 * 1801 * Any other HDSI/HISI interrupts have been handled already for P7/8 1802 * guests. For POWER9 hash guests not using rmhandlers, basic hash 1803 * fault handling is done here. 1804 */ 1805 case BOOK3S_INTERRUPT_H_DATA_STORAGE: { 1806 unsigned long vsid; 1807 long err; 1808 1809 if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) && 1810 unlikely(vcpu->arch.fault_dsisr == HDSISR_CANARY)) { 1811 r = RESUME_GUEST; /* Just retry if it's the canary */ 1812 break; 1813 } 1814 1815 if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) { 1816 /* 1817 * Radix doesn't require anything, and pre-ISAv3.0 hash 1818 * already attempted to handle this in rmhandlers. The 1819 * hash fault handling below is v3 only (it uses ASDR 1820 * via fault_gpa). 1821 */ 1822 r = RESUME_PAGE_FAULT; 1823 break; 1824 } 1825 1826 if (!(vcpu->arch.fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT))) { 1827 kvmppc_core_queue_data_storage(vcpu, 1828 kvmppc_get_msr(vcpu) & SRR1_PREFIXED, 1829 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr); 1830 r = RESUME_GUEST; 1831 break; 1832 } 1833 1834 if (!(__kvmppc_get_msr_hv(vcpu) & MSR_DR)) 1835 vsid = vcpu->kvm->arch.vrma_slb_v; 1836 else 1837 vsid = vcpu->arch.fault_gpa; 1838 1839 preempt_disable(); 1840 err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar, 1841 vsid, vcpu->arch.fault_dsisr, true); 1842 preempt_enable(); 1843 if (err == 0) { 1844 r = RESUME_GUEST; 1845 } else if (err == -1 || err == -2) { 1846 r = RESUME_PAGE_FAULT; 1847 } else { 1848 kvmppc_core_queue_data_storage(vcpu, 1849 kvmppc_get_msr(vcpu) & SRR1_PREFIXED, 1850 vcpu->arch.fault_dar, err); 1851 r = RESUME_GUEST; 1852 } 1853 break; 1854 } 1855 case BOOK3S_INTERRUPT_H_INST_STORAGE: { 1856 unsigned long vsid; 1857 long err; 1858 1859 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu); 1860 vcpu->arch.fault_dsisr = __kvmppc_get_msr_hv(vcpu) & 1861 DSISR_SRR1_MATCH_64S; 1862 if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) { 1863 /* 1864 * Radix doesn't require anything, and pre-ISAv3.0 hash 1865 * already attempted to handle this in rmhandlers. The 1866 * hash fault handling below is v3 only (it uses ASDR 1867 * via fault_gpa). 1868 */ 1869 if (__kvmppc_get_msr_hv(vcpu) & HSRR1_HISI_WRITE) 1870 vcpu->arch.fault_dsisr |= DSISR_ISSTORE; 1871 r = RESUME_PAGE_FAULT; 1872 break; 1873 } 1874 1875 if (!(vcpu->arch.fault_dsisr & SRR1_ISI_NOPT)) { 1876 kvmppc_core_queue_inst_storage(vcpu, 1877 vcpu->arch.fault_dsisr | 1878 (kvmppc_get_msr(vcpu) & SRR1_PREFIXED)); 1879 r = RESUME_GUEST; 1880 break; 1881 } 1882 1883 if (!(__kvmppc_get_msr_hv(vcpu) & MSR_IR)) 1884 vsid = vcpu->kvm->arch.vrma_slb_v; 1885 else 1886 vsid = vcpu->arch.fault_gpa; 1887 1888 preempt_disable(); 1889 err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar, 1890 vsid, vcpu->arch.fault_dsisr, false); 1891 preempt_enable(); 1892 if (err == 0) { 1893 r = RESUME_GUEST; 1894 } else if (err == -1) { 1895 r = RESUME_PAGE_FAULT; 1896 } else { 1897 kvmppc_core_queue_inst_storage(vcpu, 1898 err | (kvmppc_get_msr(vcpu) & SRR1_PREFIXED)); 1899 r = RESUME_GUEST; 1900 } 1901 break; 1902 } 1903 1904 /* 1905 * This occurs if the guest executes an illegal instruction. 1906 * If the guest debug is disabled, generate a program interrupt 1907 * to the guest. If guest debug is enabled, we need to check 1908 * whether the instruction is a software breakpoint instruction. 1909 * Accordingly return to Guest or Host. 1910 */ 1911 case BOOK3S_INTERRUPT_H_EMUL_ASSIST: 1912 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED) 1913 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ? 1914 swab32(vcpu->arch.emul_inst) : 1915 vcpu->arch.emul_inst; 1916 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) { 1917 r = kvmppc_emulate_debug_inst(vcpu); 1918 } else { 1919 kvmppc_core_queue_program(vcpu, SRR1_PROGILL | 1920 (kvmppc_get_msr(vcpu) & SRR1_PREFIXED)); 1921 r = RESUME_GUEST; 1922 } 1923 break; 1924 1925 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 1926 case BOOK3S_INTERRUPT_HV_SOFTPATCH: 1927 /* 1928 * This occurs for various TM-related instructions that 1929 * we need to emulate on POWER9 DD2.2. We have already 1930 * handled the cases where the guest was in real-suspend 1931 * mode and was transitioning to transactional state. 1932 */ 1933 r = kvmhv_p9_tm_emulation(vcpu); 1934 if (r != -1) 1935 break; 1936 fallthrough; /* go to facility unavailable handler */ 1937 #endif 1938 1939 /* 1940 * This occurs if the guest (kernel or userspace), does something that 1941 * is prohibited by HFSCR. 1942 * On POWER9, this could be a doorbell instruction that we need 1943 * to emulate. 1944 * Otherwise, we just generate a program interrupt to the guest. 1945 */ 1946 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: { 1947 u64 cause = kvmppc_get_hfscr_hv(vcpu) >> 56; 1948 1949 r = EMULATE_FAIL; 1950 if (cpu_has_feature(CPU_FTR_ARCH_300)) { 1951 switch (cause) { 1952 case FSCR_MSGP_LG: 1953 r = kvmppc_emulate_doorbell_instr(vcpu); 1954 break; 1955 case FSCR_PM_LG: 1956 r = kvmppc_pmu_unavailable(vcpu); 1957 break; 1958 case FSCR_EBB_LG: 1959 r = kvmppc_ebb_unavailable(vcpu); 1960 break; 1961 case FSCR_TM_LG: 1962 r = kvmppc_tm_unavailable(vcpu); 1963 break; 1964 default: 1965 break; 1966 } 1967 } 1968 if (r == EMULATE_FAIL) { 1969 kvmppc_core_queue_program(vcpu, SRR1_PROGILL | 1970 (kvmppc_get_msr(vcpu) & SRR1_PREFIXED)); 1971 r = RESUME_GUEST; 1972 } 1973 break; 1974 } 1975 1976 case BOOK3S_INTERRUPT_HV_RM_HARD: 1977 r = RESUME_PASSTHROUGH; 1978 break; 1979 default: 1980 kvmppc_dump_regs(vcpu); 1981 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n", 1982 vcpu->arch.trap, kvmppc_get_pc(vcpu), 1983 __kvmppc_get_msr_hv(vcpu)); 1984 run->hw.hardware_exit_reason = vcpu->arch.trap; 1985 r = RESUME_HOST; 1986 break; 1987 } 1988 1989 return r; 1990 } 1991 1992 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu) 1993 { 1994 int r; 1995 int srcu_idx; 1996 1997 vcpu->stat.sum_exits++; 1998 1999 /* 2000 * This can happen if an interrupt occurs in the last stages 2001 * of guest entry or the first stages of guest exit (i.e. after 2002 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV 2003 * and before setting it to KVM_GUEST_MODE_HOST_HV). 2004 * That can happen due to a bug, or due to a machine check 2005 * occurring at just the wrong time. 2006 */ 2007 if (__kvmppc_get_msr_hv(vcpu) & MSR_HV) { 2008 pr_emerg("KVM trap in HV mode while nested!\n"); 2009 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n", 2010 vcpu->arch.trap, kvmppc_get_pc(vcpu), 2011 __kvmppc_get_msr_hv(vcpu)); 2012 kvmppc_dump_regs(vcpu); 2013 return RESUME_HOST; 2014 } 2015 switch (vcpu->arch.trap) { 2016 /* We're good on these - the host merely wanted to get our attention */ 2017 case BOOK3S_INTERRUPT_HV_DECREMENTER: 2018 vcpu->stat.dec_exits++; 2019 r = RESUME_GUEST; 2020 break; 2021 case BOOK3S_INTERRUPT_EXTERNAL: 2022 vcpu->stat.ext_intr_exits++; 2023 r = RESUME_HOST; 2024 break; 2025 case BOOK3S_INTERRUPT_H_DOORBELL: 2026 case BOOK3S_INTERRUPT_H_VIRT: 2027 vcpu->stat.ext_intr_exits++; 2028 r = RESUME_GUEST; 2029 break; 2030 /* These need to go to the nested HV */ 2031 case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER: 2032 vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER; 2033 vcpu->stat.dec_exits++; 2034 r = RESUME_HOST; 2035 break; 2036 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/ 2037 case BOOK3S_INTERRUPT_HMI: 2038 case BOOK3S_INTERRUPT_PERFMON: 2039 case BOOK3S_INTERRUPT_SYSTEM_RESET: 2040 r = RESUME_GUEST; 2041 break; 2042 case BOOK3S_INTERRUPT_MACHINE_CHECK: 2043 { 2044 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL, 2045 DEFAULT_RATELIMIT_BURST); 2046 /* Pass the machine check to the L1 guest */ 2047 r = RESUME_HOST; 2048 /* Print the MCE event to host console. */ 2049 if (__ratelimit(&rs)) 2050 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true); 2051 break; 2052 } 2053 /* 2054 * We get these next two if the guest accesses a page which it thinks 2055 * it has mapped but which is not actually present, either because 2056 * it is for an emulated I/O device or because the corresonding 2057 * host page has been paged out. 2058 */ 2059 case BOOK3S_INTERRUPT_H_DATA_STORAGE: 2060 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); 2061 r = kvmhv_nested_page_fault(vcpu); 2062 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx); 2063 break; 2064 case BOOK3S_INTERRUPT_H_INST_STORAGE: 2065 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu); 2066 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) & 2067 DSISR_SRR1_MATCH_64S; 2068 if (__kvmppc_get_msr_hv(vcpu) & HSRR1_HISI_WRITE) 2069 vcpu->arch.fault_dsisr |= DSISR_ISSTORE; 2070 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); 2071 r = kvmhv_nested_page_fault(vcpu); 2072 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx); 2073 break; 2074 2075 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 2076 case BOOK3S_INTERRUPT_HV_SOFTPATCH: 2077 /* 2078 * This occurs for various TM-related instructions that 2079 * we need to emulate on POWER9 DD2.2. We have already 2080 * handled the cases where the guest was in real-suspend 2081 * mode and was transitioning to transactional state. 2082 */ 2083 r = kvmhv_p9_tm_emulation(vcpu); 2084 if (r != -1) 2085 break; 2086 fallthrough; /* go to facility unavailable handler */ 2087 #endif 2088 2089 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: 2090 r = RESUME_HOST; 2091 break; 2092 2093 case BOOK3S_INTERRUPT_HV_RM_HARD: 2094 vcpu->arch.trap = 0; 2095 r = RESUME_GUEST; 2096 if (!xics_on_xive()) 2097 kvmppc_xics_rm_complete(vcpu, 0); 2098 break; 2099 case BOOK3S_INTERRUPT_SYSCALL: 2100 { 2101 unsigned long req = kvmppc_get_gpr(vcpu, 3); 2102 2103 /* 2104 * The H_RPT_INVALIDATE hcalls issued by nested 2105 * guests for process-scoped invalidations when 2106 * GTSE=0, are handled here in L0. 2107 */ 2108 if (req == H_RPT_INVALIDATE) { 2109 r = kvmppc_nested_h_rpt_invalidate(vcpu); 2110 break; 2111 } 2112 2113 r = RESUME_HOST; 2114 break; 2115 } 2116 default: 2117 r = RESUME_HOST; 2118 break; 2119 } 2120 2121 return r; 2122 } 2123 2124 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu, 2125 struct kvm_sregs *sregs) 2126 { 2127 int i; 2128 2129 memset(sregs, 0, sizeof(struct kvm_sregs)); 2130 sregs->pvr = vcpu->arch.pvr; 2131 for (i = 0; i < vcpu->arch.slb_max; i++) { 2132 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige; 2133 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv; 2134 } 2135 2136 return 0; 2137 } 2138 2139 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu, 2140 struct kvm_sregs *sregs) 2141 { 2142 int i, j; 2143 2144 /* Only accept the same PVR as the host's, since we can't spoof it */ 2145 if (sregs->pvr != vcpu->arch.pvr) 2146 return -EINVAL; 2147 2148 j = 0; 2149 for (i = 0; i < vcpu->arch.slb_nr; i++) { 2150 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) { 2151 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe; 2152 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv; 2153 ++j; 2154 } 2155 } 2156 vcpu->arch.slb_max = j; 2157 2158 return 0; 2159 } 2160 2161 /* 2162 * Enforce limits on guest LPCR values based on hardware availability, 2163 * guest configuration, and possibly hypervisor support and security 2164 * concerns. 2165 */ 2166 unsigned long kvmppc_filter_lpcr_hv(struct kvm *kvm, unsigned long lpcr) 2167 { 2168 /* LPCR_TC only applies to HPT guests */ 2169 if (kvm_is_radix(kvm)) 2170 lpcr &= ~LPCR_TC; 2171 2172 /* On POWER8 and above, userspace can modify AIL */ 2173 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) 2174 lpcr &= ~LPCR_AIL; 2175 if ((lpcr & LPCR_AIL) != LPCR_AIL_3) 2176 lpcr &= ~LPCR_AIL; /* LPCR[AIL]=1/2 is disallowed */ 2177 /* 2178 * On some POWER9s we force AIL off for radix guests to prevent 2179 * executing in MSR[HV]=1 mode with the MMU enabled and PIDR set to 2180 * guest, which can result in Q0 translations with LPID=0 PID=PIDR to 2181 * be cached, which the host TLB management does not expect. 2182 */ 2183 if (kvm_is_radix(kvm) && cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG)) 2184 lpcr &= ~LPCR_AIL; 2185 2186 /* 2187 * On POWER9, allow userspace to enable large decrementer for the 2188 * guest, whether or not the host has it enabled. 2189 */ 2190 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 2191 lpcr &= ~LPCR_LD; 2192 2193 return lpcr; 2194 } 2195 2196 static void verify_lpcr(struct kvm *kvm, unsigned long lpcr) 2197 { 2198 if (lpcr != kvmppc_filter_lpcr_hv(kvm, lpcr)) { 2199 WARN_ONCE(1, "lpcr 0x%lx differs from filtered 0x%lx\n", 2200 lpcr, kvmppc_filter_lpcr_hv(kvm, lpcr)); 2201 } 2202 } 2203 2204 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr, 2205 bool preserve_top32) 2206 { 2207 struct kvm *kvm = vcpu->kvm; 2208 struct kvmppc_vcore *vc = vcpu->arch.vcore; 2209 u64 mask; 2210 2211 spin_lock(&vc->lock); 2212 2213 /* 2214 * Userspace can only modify 2215 * DPFD (default prefetch depth), ILE (interrupt little-endian), 2216 * TC (translation control), AIL (alternate interrupt location), 2217 * LD (large decrementer). 2218 * These are subject to restrictions from kvmppc_filter_lcpr_hv(). 2219 */ 2220 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD; 2221 2222 /* Broken 32-bit version of LPCR must not clear top bits */ 2223 if (preserve_top32) 2224 mask &= 0xFFFFFFFF; 2225 2226 new_lpcr = kvmppc_filter_lpcr_hv(kvm, 2227 (vc->lpcr & ~mask) | (new_lpcr & mask)); 2228 2229 /* 2230 * If ILE (interrupt little-endian) has changed, update the 2231 * MSR_LE bit in the intr_msr for each vcpu in this vcore. 2232 */ 2233 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) { 2234 struct kvm_vcpu *vcpu; 2235 unsigned long i; 2236 2237 kvm_for_each_vcpu(i, vcpu, kvm) { 2238 if (vcpu->arch.vcore != vc) 2239 continue; 2240 if (new_lpcr & LPCR_ILE) 2241 vcpu->arch.intr_msr |= MSR_LE; 2242 else 2243 vcpu->arch.intr_msr &= ~MSR_LE; 2244 } 2245 } 2246 2247 vc->lpcr = new_lpcr; 2248 kvmhv_nestedv2_mark_dirty(vcpu, KVMPPC_GSID_LPCR); 2249 2250 spin_unlock(&vc->lock); 2251 } 2252 2253 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id, 2254 union kvmppc_one_reg *val) 2255 { 2256 int r = 0; 2257 long int i; 2258 2259 switch (id) { 2260 case KVM_REG_PPC_DEBUG_INST: 2261 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT); 2262 break; 2263 case KVM_REG_PPC_HIOR: 2264 *val = get_reg_val(id, 0); 2265 break; 2266 case KVM_REG_PPC_DABR: 2267 *val = get_reg_val(id, vcpu->arch.dabr); 2268 break; 2269 case KVM_REG_PPC_DABRX: 2270 *val = get_reg_val(id, vcpu->arch.dabrx); 2271 break; 2272 case KVM_REG_PPC_DSCR: 2273 *val = get_reg_val(id, kvmppc_get_dscr_hv(vcpu)); 2274 break; 2275 case KVM_REG_PPC_PURR: 2276 *val = get_reg_val(id, kvmppc_get_purr_hv(vcpu)); 2277 break; 2278 case KVM_REG_PPC_SPURR: 2279 *val = get_reg_val(id, kvmppc_get_spurr_hv(vcpu)); 2280 break; 2281 case KVM_REG_PPC_AMR: 2282 *val = get_reg_val(id, kvmppc_get_amr_hv(vcpu)); 2283 break; 2284 case KVM_REG_PPC_UAMOR: 2285 *val = get_reg_val(id, kvmppc_get_uamor_hv(vcpu)); 2286 break; 2287 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1: 2288 i = id - KVM_REG_PPC_MMCR0; 2289 *val = get_reg_val(id, kvmppc_get_mmcr_hv(vcpu, i)); 2290 break; 2291 case KVM_REG_PPC_MMCR2: 2292 *val = get_reg_val(id, kvmppc_get_mmcr_hv(vcpu, 2)); 2293 break; 2294 case KVM_REG_PPC_MMCRA: 2295 *val = get_reg_val(id, kvmppc_get_mmcra_hv(vcpu)); 2296 break; 2297 case KVM_REG_PPC_MMCRS: 2298 *val = get_reg_val(id, vcpu->arch.mmcrs); 2299 break; 2300 case KVM_REG_PPC_MMCR3: 2301 *val = get_reg_val(id, kvmppc_get_mmcr_hv(vcpu, 3)); 2302 break; 2303 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8: 2304 i = id - KVM_REG_PPC_PMC1; 2305 *val = get_reg_val(id, kvmppc_get_pmc_hv(vcpu, i)); 2306 break; 2307 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2: 2308 i = id - KVM_REG_PPC_SPMC1; 2309 *val = get_reg_val(id, vcpu->arch.spmc[i]); 2310 break; 2311 case KVM_REG_PPC_SIAR: 2312 *val = get_reg_val(id, kvmppc_get_siar_hv(vcpu)); 2313 break; 2314 case KVM_REG_PPC_SDAR: 2315 *val = get_reg_val(id, kvmppc_get_sdar_hv(vcpu)); 2316 break; 2317 case KVM_REG_PPC_SIER: 2318 *val = get_reg_val(id, kvmppc_get_sier_hv(vcpu, 0)); 2319 break; 2320 case KVM_REG_PPC_SIER2: 2321 *val = get_reg_val(id, kvmppc_get_sier_hv(vcpu, 1)); 2322 break; 2323 case KVM_REG_PPC_SIER3: 2324 *val = get_reg_val(id, kvmppc_get_sier_hv(vcpu, 2)); 2325 break; 2326 case KVM_REG_PPC_IAMR: 2327 *val = get_reg_val(id, kvmppc_get_iamr_hv(vcpu)); 2328 break; 2329 case KVM_REG_PPC_PSPB: 2330 *val = get_reg_val(id, kvmppc_get_pspb_hv(vcpu)); 2331 break; 2332 case KVM_REG_PPC_DPDES: 2333 /* 2334 * On POWER9, where we are emulating msgsndp etc., 2335 * we return 1 bit for each vcpu, which can come from 2336 * either vcore->dpdes or doorbell_request. 2337 * On POWER8, doorbell_request is 0. 2338 */ 2339 if (cpu_has_feature(CPU_FTR_ARCH_300)) 2340 *val = get_reg_val(id, vcpu->arch.doorbell_request); 2341 else 2342 *val = get_reg_val(id, vcpu->arch.vcore->dpdes); 2343 break; 2344 case KVM_REG_PPC_VTB: 2345 *val = get_reg_val(id, kvmppc_get_vtb(vcpu)); 2346 break; 2347 case KVM_REG_PPC_DAWR: 2348 *val = get_reg_val(id, kvmppc_get_dawr0_hv(vcpu)); 2349 break; 2350 case KVM_REG_PPC_DAWRX: 2351 *val = get_reg_val(id, kvmppc_get_dawrx0_hv(vcpu)); 2352 break; 2353 case KVM_REG_PPC_DAWR1: 2354 *val = get_reg_val(id, kvmppc_get_dawr1_hv(vcpu)); 2355 break; 2356 case KVM_REG_PPC_DAWRX1: 2357 *val = get_reg_val(id, kvmppc_get_dawrx1_hv(vcpu)); 2358 break; 2359 case KVM_REG_PPC_DEXCR: 2360 *val = get_reg_val(id, kvmppc_get_dexcr_hv(vcpu)); 2361 break; 2362 case KVM_REG_PPC_HASHKEYR: 2363 *val = get_reg_val(id, kvmppc_get_hashkeyr_hv(vcpu)); 2364 break; 2365 case KVM_REG_PPC_HASHPKEYR: 2366 *val = get_reg_val(id, kvmppc_get_hashpkeyr_hv(vcpu)); 2367 break; 2368 case KVM_REG_PPC_CIABR: 2369 *val = get_reg_val(id, kvmppc_get_ciabr_hv(vcpu)); 2370 break; 2371 case KVM_REG_PPC_CSIGR: 2372 *val = get_reg_val(id, vcpu->arch.csigr); 2373 break; 2374 case KVM_REG_PPC_TACR: 2375 *val = get_reg_val(id, vcpu->arch.tacr); 2376 break; 2377 case KVM_REG_PPC_TCSCR: 2378 *val = get_reg_val(id, vcpu->arch.tcscr); 2379 break; 2380 case KVM_REG_PPC_PID: 2381 *val = get_reg_val(id, kvmppc_get_pid(vcpu)); 2382 break; 2383 case KVM_REG_PPC_ACOP: 2384 *val = get_reg_val(id, vcpu->arch.acop); 2385 break; 2386 case KVM_REG_PPC_WORT: 2387 *val = get_reg_val(id, kvmppc_get_wort_hv(vcpu)); 2388 break; 2389 case KVM_REG_PPC_TIDR: 2390 *val = get_reg_val(id, vcpu->arch.tid); 2391 break; 2392 case KVM_REG_PPC_PSSCR: 2393 *val = get_reg_val(id, vcpu->arch.psscr); 2394 break; 2395 case KVM_REG_PPC_VPA_ADDR: 2396 spin_lock(&vcpu->arch.vpa_update_lock); 2397 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa); 2398 spin_unlock(&vcpu->arch.vpa_update_lock); 2399 break; 2400 case KVM_REG_PPC_VPA_SLB: 2401 spin_lock(&vcpu->arch.vpa_update_lock); 2402 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa; 2403 val->vpaval.length = vcpu->arch.slb_shadow.len; 2404 spin_unlock(&vcpu->arch.vpa_update_lock); 2405 break; 2406 case KVM_REG_PPC_VPA_DTL: 2407 spin_lock(&vcpu->arch.vpa_update_lock); 2408 val->vpaval.addr = vcpu->arch.dtl.next_gpa; 2409 val->vpaval.length = vcpu->arch.dtl.len; 2410 spin_unlock(&vcpu->arch.vpa_update_lock); 2411 break; 2412 case KVM_REG_PPC_TB_OFFSET: 2413 *val = get_reg_val(id, kvmppc_get_tb_offset(vcpu)); 2414 break; 2415 case KVM_REG_PPC_LPCR: 2416 case KVM_REG_PPC_LPCR_64: 2417 *val = get_reg_val(id, kvmppc_get_lpcr(vcpu)); 2418 break; 2419 case KVM_REG_PPC_PPR: 2420 *val = get_reg_val(id, kvmppc_get_ppr_hv(vcpu)); 2421 break; 2422 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 2423 case KVM_REG_PPC_TFHAR: 2424 *val = get_reg_val(id, vcpu->arch.tfhar); 2425 break; 2426 case KVM_REG_PPC_TFIAR: 2427 *val = get_reg_val(id, vcpu->arch.tfiar); 2428 break; 2429 case KVM_REG_PPC_TEXASR: 2430 *val = get_reg_val(id, vcpu->arch.texasr); 2431 break; 2432 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31: 2433 i = id - KVM_REG_PPC_TM_GPR0; 2434 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]); 2435 break; 2436 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63: 2437 { 2438 int j; 2439 i = id - KVM_REG_PPC_TM_VSR0; 2440 if (i < 32) 2441 for (j = 0; j < TS_FPRWIDTH; j++) 2442 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j]; 2443 else { 2444 if (cpu_has_feature(CPU_FTR_ALTIVEC)) 2445 val->vval = vcpu->arch.vr_tm.vr[i-32]; 2446 else 2447 r = -ENXIO; 2448 } 2449 break; 2450 } 2451 case KVM_REG_PPC_TM_CR: 2452 *val = get_reg_val(id, vcpu->arch.cr_tm); 2453 break; 2454 case KVM_REG_PPC_TM_XER: 2455 *val = get_reg_val(id, vcpu->arch.xer_tm); 2456 break; 2457 case KVM_REG_PPC_TM_LR: 2458 *val = get_reg_val(id, vcpu->arch.lr_tm); 2459 break; 2460 case KVM_REG_PPC_TM_CTR: 2461 *val = get_reg_val(id, vcpu->arch.ctr_tm); 2462 break; 2463 case KVM_REG_PPC_TM_FPSCR: 2464 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr); 2465 break; 2466 case KVM_REG_PPC_TM_AMR: 2467 *val = get_reg_val(id, vcpu->arch.amr_tm); 2468 break; 2469 case KVM_REG_PPC_TM_PPR: 2470 *val = get_reg_val(id, vcpu->arch.ppr_tm); 2471 break; 2472 case KVM_REG_PPC_TM_VRSAVE: 2473 *val = get_reg_val(id, vcpu->arch.vrsave_tm); 2474 break; 2475 case KVM_REG_PPC_TM_VSCR: 2476 if (cpu_has_feature(CPU_FTR_ALTIVEC)) 2477 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]); 2478 else 2479 r = -ENXIO; 2480 break; 2481 case KVM_REG_PPC_TM_DSCR: 2482 *val = get_reg_val(id, vcpu->arch.dscr_tm); 2483 break; 2484 case KVM_REG_PPC_TM_TAR: 2485 *val = get_reg_val(id, vcpu->arch.tar_tm); 2486 break; 2487 #endif 2488 case KVM_REG_PPC_ARCH_COMPAT: 2489 *val = get_reg_val(id, kvmppc_get_arch_compat(vcpu)); 2490 break; 2491 case KVM_REG_PPC_DEC_EXPIRY: 2492 *val = get_reg_val(id, kvmppc_get_dec_expires(vcpu)); 2493 break; 2494 case KVM_REG_PPC_ONLINE: 2495 *val = get_reg_val(id, vcpu->arch.online); 2496 break; 2497 case KVM_REG_PPC_PTCR: 2498 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr); 2499 break; 2500 case KVM_REG_PPC_FSCR: 2501 *val = get_reg_val(id, kvmppc_get_fscr_hv(vcpu)); 2502 break; 2503 default: 2504 r = -EINVAL; 2505 break; 2506 } 2507 2508 return r; 2509 } 2510 2511 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id, 2512 union kvmppc_one_reg *val) 2513 { 2514 int r = 0; 2515 long int i; 2516 unsigned long addr, len; 2517 2518 switch (id) { 2519 case KVM_REG_PPC_HIOR: 2520 /* Only allow this to be set to zero */ 2521 if (set_reg_val(id, *val)) 2522 r = -EINVAL; 2523 break; 2524 case KVM_REG_PPC_DABR: 2525 vcpu->arch.dabr = set_reg_val(id, *val); 2526 break; 2527 case KVM_REG_PPC_DABRX: 2528 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP; 2529 break; 2530 case KVM_REG_PPC_DSCR: 2531 kvmppc_set_dscr_hv(vcpu, set_reg_val(id, *val)); 2532 break; 2533 case KVM_REG_PPC_PURR: 2534 kvmppc_set_purr_hv(vcpu, set_reg_val(id, *val)); 2535 break; 2536 case KVM_REG_PPC_SPURR: 2537 kvmppc_set_spurr_hv(vcpu, set_reg_val(id, *val)); 2538 break; 2539 case KVM_REG_PPC_AMR: 2540 kvmppc_set_amr_hv(vcpu, set_reg_val(id, *val)); 2541 break; 2542 case KVM_REG_PPC_UAMOR: 2543 kvmppc_set_uamor_hv(vcpu, set_reg_val(id, *val)); 2544 break; 2545 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1: 2546 i = id - KVM_REG_PPC_MMCR0; 2547 kvmppc_set_mmcr_hv(vcpu, i, set_reg_val(id, *val)); 2548 break; 2549 case KVM_REG_PPC_MMCR2: 2550 kvmppc_set_mmcr_hv(vcpu, 2, set_reg_val(id, *val)); 2551 break; 2552 case KVM_REG_PPC_MMCRA: 2553 kvmppc_set_mmcra_hv(vcpu, set_reg_val(id, *val)); 2554 break; 2555 case KVM_REG_PPC_MMCRS: 2556 vcpu->arch.mmcrs = set_reg_val(id, *val); 2557 break; 2558 case KVM_REG_PPC_MMCR3: 2559 kvmppc_set_mmcr_hv(vcpu, 3, set_reg_val(id, *val)); 2560 break; 2561 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8: 2562 i = id - KVM_REG_PPC_PMC1; 2563 kvmppc_set_pmc_hv(vcpu, i, set_reg_val(id, *val)); 2564 break; 2565 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2: 2566 i = id - KVM_REG_PPC_SPMC1; 2567 vcpu->arch.spmc[i] = set_reg_val(id, *val); 2568 break; 2569 case KVM_REG_PPC_SIAR: 2570 kvmppc_set_siar_hv(vcpu, set_reg_val(id, *val)); 2571 break; 2572 case KVM_REG_PPC_SDAR: 2573 kvmppc_set_sdar_hv(vcpu, set_reg_val(id, *val)); 2574 break; 2575 case KVM_REG_PPC_SIER: 2576 kvmppc_set_sier_hv(vcpu, 0, set_reg_val(id, *val)); 2577 break; 2578 case KVM_REG_PPC_SIER2: 2579 kvmppc_set_sier_hv(vcpu, 1, set_reg_val(id, *val)); 2580 break; 2581 case KVM_REG_PPC_SIER3: 2582 kvmppc_set_sier_hv(vcpu, 2, set_reg_val(id, *val)); 2583 break; 2584 case KVM_REG_PPC_IAMR: 2585 kvmppc_set_iamr_hv(vcpu, set_reg_val(id, *val)); 2586 break; 2587 case KVM_REG_PPC_PSPB: 2588 kvmppc_set_pspb_hv(vcpu, set_reg_val(id, *val)); 2589 break; 2590 case KVM_REG_PPC_DPDES: 2591 if (cpu_has_feature(CPU_FTR_ARCH_300)) 2592 vcpu->arch.doorbell_request = set_reg_val(id, *val) & 1; 2593 else 2594 vcpu->arch.vcore->dpdes = set_reg_val(id, *val); 2595 break; 2596 case KVM_REG_PPC_VTB: 2597 kvmppc_set_vtb(vcpu, set_reg_val(id, *val)); 2598 break; 2599 case KVM_REG_PPC_DAWR: 2600 kvmppc_set_dawr0_hv(vcpu, set_reg_val(id, *val)); 2601 break; 2602 case KVM_REG_PPC_DAWRX: 2603 kvmppc_set_dawrx0_hv(vcpu, set_reg_val(id, *val) & ~DAWRX_HYP); 2604 break; 2605 case KVM_REG_PPC_DAWR1: 2606 kvmppc_set_dawr1_hv(vcpu, set_reg_val(id, *val)); 2607 break; 2608 case KVM_REG_PPC_DAWRX1: 2609 kvmppc_set_dawrx1_hv(vcpu, set_reg_val(id, *val) & ~DAWRX_HYP); 2610 break; 2611 case KVM_REG_PPC_DEXCR: 2612 kvmppc_set_dexcr_hv(vcpu, set_reg_val(id, *val)); 2613 break; 2614 case KVM_REG_PPC_HASHKEYR: 2615 kvmppc_set_hashkeyr_hv(vcpu, set_reg_val(id, *val)); 2616 break; 2617 case KVM_REG_PPC_HASHPKEYR: 2618 kvmppc_set_hashpkeyr_hv(vcpu, set_reg_val(id, *val)); 2619 break; 2620 case KVM_REG_PPC_CIABR: 2621 kvmppc_set_ciabr_hv(vcpu, set_reg_val(id, *val)); 2622 /* Don't allow setting breakpoints in hypervisor code */ 2623 if ((kvmppc_get_ciabr_hv(vcpu) & CIABR_PRIV) == CIABR_PRIV_HYPER) 2624 kvmppc_set_ciabr_hv(vcpu, kvmppc_get_ciabr_hv(vcpu) & ~CIABR_PRIV); 2625 break; 2626 case KVM_REG_PPC_CSIGR: 2627 vcpu->arch.csigr = set_reg_val(id, *val); 2628 break; 2629 case KVM_REG_PPC_TACR: 2630 vcpu->arch.tacr = set_reg_val(id, *val); 2631 break; 2632 case KVM_REG_PPC_TCSCR: 2633 vcpu->arch.tcscr = set_reg_val(id, *val); 2634 break; 2635 case KVM_REG_PPC_PID: 2636 kvmppc_set_pid(vcpu, set_reg_val(id, *val)); 2637 break; 2638 case KVM_REG_PPC_ACOP: 2639 vcpu->arch.acop = set_reg_val(id, *val); 2640 break; 2641 case KVM_REG_PPC_WORT: 2642 kvmppc_set_wort_hv(vcpu, set_reg_val(id, *val)); 2643 break; 2644 case KVM_REG_PPC_TIDR: 2645 vcpu->arch.tid = set_reg_val(id, *val); 2646 break; 2647 case KVM_REG_PPC_PSSCR: 2648 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS; 2649 break; 2650 case KVM_REG_PPC_VPA_ADDR: 2651 addr = set_reg_val(id, *val); 2652 r = -EINVAL; 2653 if (!addr && (vcpu->arch.slb_shadow.next_gpa || 2654 vcpu->arch.dtl.next_gpa)) 2655 break; 2656 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca)); 2657 break; 2658 case KVM_REG_PPC_VPA_SLB: 2659 addr = val->vpaval.addr; 2660 len = val->vpaval.length; 2661 r = -EINVAL; 2662 if (addr && !vcpu->arch.vpa.next_gpa) 2663 break; 2664 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len); 2665 break; 2666 case KVM_REG_PPC_VPA_DTL: 2667 addr = val->vpaval.addr; 2668 len = val->vpaval.length; 2669 r = -EINVAL; 2670 if (addr && (len < sizeof(struct dtl_entry) || 2671 !vcpu->arch.vpa.next_gpa)) 2672 break; 2673 len -= len % sizeof(struct dtl_entry); 2674 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len); 2675 break; 2676 case KVM_REG_PPC_TB_OFFSET: 2677 { 2678 /* round up to multiple of 2^24 */ 2679 u64 tb_offset = ALIGN(set_reg_val(id, *val), 1UL << 24); 2680 2681 /* 2682 * Now that we know the timebase offset, update the 2683 * decrementer expiry with a guest timebase value. If 2684 * the userspace does not set DEC_EXPIRY, this ensures 2685 * a migrated vcpu at least starts with an expired 2686 * decrementer, which is better than a large one that 2687 * causes a hang. 2688 */ 2689 kvmppc_set_tb_offset(vcpu, tb_offset); 2690 if (!kvmppc_get_dec_expires(vcpu) && tb_offset) 2691 kvmppc_set_dec_expires(vcpu, get_tb() + tb_offset); 2692 2693 kvmppc_set_tb_offset(vcpu, tb_offset); 2694 break; 2695 } 2696 case KVM_REG_PPC_LPCR: 2697 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true); 2698 break; 2699 case KVM_REG_PPC_LPCR_64: 2700 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false); 2701 break; 2702 case KVM_REG_PPC_PPR: 2703 kvmppc_set_ppr_hv(vcpu, set_reg_val(id, *val)); 2704 break; 2705 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 2706 case KVM_REG_PPC_TFHAR: 2707 vcpu->arch.tfhar = set_reg_val(id, *val); 2708 break; 2709 case KVM_REG_PPC_TFIAR: 2710 vcpu->arch.tfiar = set_reg_val(id, *val); 2711 break; 2712 case KVM_REG_PPC_TEXASR: 2713 vcpu->arch.texasr = set_reg_val(id, *val); 2714 break; 2715 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31: 2716 i = id - KVM_REG_PPC_TM_GPR0; 2717 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val); 2718 break; 2719 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63: 2720 { 2721 int j; 2722 i = id - KVM_REG_PPC_TM_VSR0; 2723 if (i < 32) 2724 for (j = 0; j < TS_FPRWIDTH; j++) 2725 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j]; 2726 else 2727 if (cpu_has_feature(CPU_FTR_ALTIVEC)) 2728 vcpu->arch.vr_tm.vr[i-32] = val->vval; 2729 else 2730 r = -ENXIO; 2731 break; 2732 } 2733 case KVM_REG_PPC_TM_CR: 2734 vcpu->arch.cr_tm = set_reg_val(id, *val); 2735 break; 2736 case KVM_REG_PPC_TM_XER: 2737 vcpu->arch.xer_tm = set_reg_val(id, *val); 2738 break; 2739 case KVM_REG_PPC_TM_LR: 2740 vcpu->arch.lr_tm = set_reg_val(id, *val); 2741 break; 2742 case KVM_REG_PPC_TM_CTR: 2743 vcpu->arch.ctr_tm = set_reg_val(id, *val); 2744 break; 2745 case KVM_REG_PPC_TM_FPSCR: 2746 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val); 2747 break; 2748 case KVM_REG_PPC_TM_AMR: 2749 vcpu->arch.amr_tm = set_reg_val(id, *val); 2750 break; 2751 case KVM_REG_PPC_TM_PPR: 2752 vcpu->arch.ppr_tm = set_reg_val(id, *val); 2753 break; 2754 case KVM_REG_PPC_TM_VRSAVE: 2755 vcpu->arch.vrsave_tm = set_reg_val(id, *val); 2756 break; 2757 case KVM_REG_PPC_TM_VSCR: 2758 if (cpu_has_feature(CPU_FTR_ALTIVEC)) 2759 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val); 2760 else 2761 r = - ENXIO; 2762 break; 2763 case KVM_REG_PPC_TM_DSCR: 2764 vcpu->arch.dscr_tm = set_reg_val(id, *val); 2765 break; 2766 case KVM_REG_PPC_TM_TAR: 2767 vcpu->arch.tar_tm = set_reg_val(id, *val); 2768 break; 2769 #endif 2770 case KVM_REG_PPC_ARCH_COMPAT: 2771 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val)); 2772 break; 2773 case KVM_REG_PPC_DEC_EXPIRY: 2774 kvmppc_set_dec_expires(vcpu, set_reg_val(id, *val)); 2775 break; 2776 case KVM_REG_PPC_ONLINE: 2777 i = set_reg_val(id, *val); 2778 if (i && !vcpu->arch.online) 2779 atomic_inc(&vcpu->arch.vcore->online_count); 2780 else if (!i && vcpu->arch.online) 2781 atomic_dec(&vcpu->arch.vcore->online_count); 2782 vcpu->arch.online = i; 2783 break; 2784 case KVM_REG_PPC_PTCR: 2785 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val); 2786 break; 2787 case KVM_REG_PPC_FSCR: 2788 kvmppc_set_fscr_hv(vcpu, set_reg_val(id, *val)); 2789 break; 2790 default: 2791 r = -EINVAL; 2792 break; 2793 } 2794 2795 return r; 2796 } 2797 2798 /* 2799 * On POWER9, threads are independent and can be in different partitions. 2800 * Therefore we consider each thread to be a subcore. 2801 * There is a restriction that all threads have to be in the same 2802 * MMU mode (radix or HPT), unfortunately, but since we only support 2803 * HPT guests on a HPT host so far, that isn't an impediment yet. 2804 */ 2805 static int threads_per_vcore(struct kvm *kvm) 2806 { 2807 if (cpu_has_feature(CPU_FTR_ARCH_300)) 2808 return 1; 2809 return threads_per_subcore; 2810 } 2811 2812 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id) 2813 { 2814 struct kvmppc_vcore *vcore; 2815 2816 vcore = kzalloc_obj(struct kvmppc_vcore); 2817 2818 if (vcore == NULL) 2819 return NULL; 2820 2821 spin_lock_init(&vcore->lock); 2822 spin_lock_init(&vcore->stoltb_lock); 2823 rcuwait_init(&vcore->wait); 2824 vcore->preempt_tb = TB_NIL; 2825 vcore->lpcr = kvm->arch.lpcr; 2826 vcore->first_vcpuid = id; 2827 vcore->kvm = kvm; 2828 INIT_LIST_HEAD(&vcore->preempt_list); 2829 2830 return vcore; 2831 } 2832 2833 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING 2834 static struct debugfs_timings_element { 2835 const char *name; 2836 size_t offset; 2837 } timings[] = { 2838 #ifdef CONFIG_KVM_BOOK3S_HV_P9_TIMING 2839 {"vcpu_entry", offsetof(struct kvm_vcpu, arch.vcpu_entry)}, 2840 {"guest_entry", offsetof(struct kvm_vcpu, arch.guest_entry)}, 2841 {"in_guest", offsetof(struct kvm_vcpu, arch.in_guest)}, 2842 {"guest_exit", offsetof(struct kvm_vcpu, arch.guest_exit)}, 2843 {"vcpu_exit", offsetof(struct kvm_vcpu, arch.vcpu_exit)}, 2844 {"hypercall", offsetof(struct kvm_vcpu, arch.hcall)}, 2845 {"page_fault", offsetof(struct kvm_vcpu, arch.pg_fault)}, 2846 #else 2847 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)}, 2848 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)}, 2849 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)}, 2850 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)}, 2851 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)}, 2852 #endif 2853 }; 2854 2855 #define N_TIMINGS (ARRAY_SIZE(timings)) 2856 2857 struct debugfs_timings_state { 2858 struct kvm_vcpu *vcpu; 2859 unsigned int buflen; 2860 char buf[N_TIMINGS * 100]; 2861 }; 2862 2863 static int debugfs_timings_open(struct inode *inode, struct file *file) 2864 { 2865 struct kvm_vcpu *vcpu = inode->i_private; 2866 struct debugfs_timings_state *p; 2867 2868 p = kzalloc_obj(*p); 2869 if (!p) 2870 return -ENOMEM; 2871 2872 kvm_get_kvm(vcpu->kvm); 2873 p->vcpu = vcpu; 2874 file->private_data = p; 2875 2876 return nonseekable_open(inode, file); 2877 } 2878 2879 static int debugfs_timings_release(struct inode *inode, struct file *file) 2880 { 2881 struct debugfs_timings_state *p = file->private_data; 2882 2883 kvm_put_kvm(p->vcpu->kvm); 2884 kfree(p); 2885 return 0; 2886 } 2887 2888 static ssize_t debugfs_timings_read(struct file *file, char __user *buf, 2889 size_t len, loff_t *ppos) 2890 { 2891 struct debugfs_timings_state *p = file->private_data; 2892 struct kvm_vcpu *vcpu = p->vcpu; 2893 char *s, *buf_end; 2894 struct kvmhv_tb_accumulator tb; 2895 u64 count; 2896 loff_t pos; 2897 ssize_t n; 2898 int i, loops; 2899 bool ok; 2900 2901 if (!p->buflen) { 2902 s = p->buf; 2903 buf_end = s + sizeof(p->buf); 2904 for (i = 0; i < N_TIMINGS; ++i) { 2905 struct kvmhv_tb_accumulator *acc; 2906 2907 acc = (struct kvmhv_tb_accumulator *) 2908 ((unsigned long)vcpu + timings[i].offset); 2909 ok = false; 2910 for (loops = 0; loops < 1000; ++loops) { 2911 count = acc->seqcount; 2912 if (!(count & 1)) { 2913 smp_rmb(); 2914 tb = *acc; 2915 smp_rmb(); 2916 if (count == acc->seqcount) { 2917 ok = true; 2918 break; 2919 } 2920 } 2921 udelay(1); 2922 } 2923 if (!ok) 2924 snprintf(s, buf_end - s, "%s: stuck\n", 2925 timings[i].name); 2926 else 2927 snprintf(s, buf_end - s, 2928 "%s: %llu %llu %llu %llu\n", 2929 timings[i].name, count / 2, 2930 tb_to_ns(tb.tb_total), 2931 tb_to_ns(tb.tb_min), 2932 tb_to_ns(tb.tb_max)); 2933 s += strlen(s); 2934 } 2935 p->buflen = s - p->buf; 2936 } 2937 2938 pos = *ppos; 2939 if (pos >= p->buflen) 2940 return 0; 2941 if (len > p->buflen - pos) 2942 len = p->buflen - pos; 2943 n = copy_to_user(buf, p->buf + pos, len); 2944 if (n) { 2945 if (n == len) 2946 return -EFAULT; 2947 len -= n; 2948 } 2949 *ppos = pos + len; 2950 return len; 2951 } 2952 2953 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf, 2954 size_t len, loff_t *ppos) 2955 { 2956 return -EACCES; 2957 } 2958 2959 static const struct file_operations debugfs_timings_ops = { 2960 .owner = THIS_MODULE, 2961 .open = debugfs_timings_open, 2962 .release = debugfs_timings_release, 2963 .read = debugfs_timings_read, 2964 .write = debugfs_timings_write, 2965 .llseek = generic_file_llseek, 2966 }; 2967 2968 /* Create a debugfs directory for the vcpu */ 2969 static int kvmppc_arch_create_vcpu_debugfs_hv(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry) 2970 { 2971 if (cpu_has_feature(CPU_FTR_ARCH_300) == IS_ENABLED(CONFIG_KVM_BOOK3S_HV_P9_TIMING)) 2972 debugfs_create_file("timings", 0444, debugfs_dentry, vcpu, 2973 &debugfs_timings_ops); 2974 return 0; 2975 } 2976 2977 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */ 2978 static int kvmppc_arch_create_vcpu_debugfs_hv(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry) 2979 { 2980 return 0; 2981 } 2982 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */ 2983 2984 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu) 2985 { 2986 int err; 2987 int core; 2988 struct kvmppc_vcore *vcore; 2989 struct kvm *kvm; 2990 unsigned int id; 2991 2992 kvm = vcpu->kvm; 2993 id = vcpu->vcpu_id; 2994 2995 vcpu->arch.shared = &vcpu->arch.shregs; 2996 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE 2997 /* 2998 * The shared struct is never shared on HV, 2999 * so we can always use host endianness 3000 */ 3001 #ifdef __BIG_ENDIAN__ 3002 vcpu->arch.shared_big_endian = true; 3003 #else 3004 vcpu->arch.shared_big_endian = false; 3005 #endif 3006 #endif 3007 3008 if (kvmhv_is_nestedv2()) { 3009 err = kvmhv_nestedv2_vcpu_create(vcpu, &vcpu->arch.nestedv2_io); 3010 if (err < 0) 3011 return err; 3012 } 3013 3014 kvmppc_set_mmcr_hv(vcpu, 0, MMCR0_FC); 3015 if (cpu_has_feature(CPU_FTR_ARCH_31)) { 3016 kvmppc_set_mmcr_hv(vcpu, 0, kvmppc_get_mmcr_hv(vcpu, 0) | MMCR0_PMCCEXT); 3017 kvmppc_set_mmcra_hv(vcpu, MMCRA_BHRB_DISABLE); 3018 } 3019 3020 kvmppc_set_ctrl_hv(vcpu, CTRL_RUNLATCH); 3021 /* default to host PVR, since we can't spoof it */ 3022 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR)); 3023 spin_lock_init(&vcpu->arch.vpa_update_lock); 3024 spin_lock_init(&vcpu->arch.tbacct_lock); 3025 vcpu->arch.busy_preempt = TB_NIL; 3026 __kvmppc_set_msr_hv(vcpu, MSR_ME); 3027 vcpu->arch.intr_msr = MSR_SF | MSR_ME; 3028 3029 /* 3030 * Set the default HFSCR for the guest from the host value. 3031 * This value is only used on POWER9 and later. 3032 * On >= POWER9, we want to virtualize the doorbell facility, so we 3033 * don't set the HFSCR_MSGP bit, and that causes those instructions 3034 * to trap and then we emulate them. 3035 */ 3036 kvmppc_set_hfscr_hv(vcpu, HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB | 3037 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP); 3038 3039 /* On POWER10 and later, allow prefixed instructions */ 3040 if (cpu_has_feature(CPU_FTR_ARCH_31)) 3041 kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_PREFIX); 3042 3043 if (cpu_has_feature(CPU_FTR_HVMODE)) { 3044 kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) & mfspr(SPRN_HFSCR)); 3045 3046 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 3047 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) 3048 kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) | HFSCR_TM); 3049 #endif 3050 } 3051 if (cpu_has_feature(CPU_FTR_TM_COMP)) 3052 vcpu->arch.hfscr |= HFSCR_TM; 3053 3054 vcpu->arch.hfscr_permitted = kvmppc_get_hfscr_hv(vcpu); 3055 3056 /* 3057 * PM, EBB, TM are demand-faulted so start with it clear. 3058 */ 3059 kvmppc_set_hfscr_hv(vcpu, kvmppc_get_hfscr_hv(vcpu) & ~(HFSCR_PM | HFSCR_EBB | HFSCR_TM)); 3060 3061 kvmppc_mmu_book3s_hv_init(vcpu); 3062 3063 vcpu->arch.state = KVMPPC_VCPU_NOTREADY; 3064 3065 init_waitqueue_head(&vcpu->arch.cpu_run); 3066 3067 vcore = NULL; 3068 err = -EINVAL; 3069 if (cpu_has_feature(CPU_FTR_ARCH_300)) { 3070 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) { 3071 pr_devel("KVM: VCPU ID too high\n"); 3072 core = KVM_MAX_VCORES; 3073 } else { 3074 BUG_ON(kvm->arch.smt_mode != 1); 3075 core = kvmppc_pack_vcpu_id(kvm, id); 3076 } 3077 } else { 3078 core = id / kvm->arch.smt_mode; 3079 } 3080 if (core < KVM_MAX_VCORES) { 3081 vcore = kvm->arch.vcores[core]; 3082 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) { 3083 pr_devel("KVM: collision on id %u", id); 3084 vcore = NULL; 3085 } else if (!vcore) { 3086 /* 3087 * Take mmu_setup_lock for mutual exclusion 3088 * with kvmppc_update_lpcr(). 3089 */ 3090 err = -ENOMEM; 3091 vcore = kvmppc_vcore_create(kvm, 3092 id & ~(kvm->arch.smt_mode - 1)); 3093 mutex_lock(&kvm->arch.mmu_setup_lock); 3094 kvm->arch.vcores[core] = vcore; 3095 kvm->arch.online_vcores++; 3096 mutex_unlock(&kvm->arch.mmu_setup_lock); 3097 } 3098 } 3099 3100 if (!vcore) 3101 return err; 3102 3103 spin_lock(&vcore->lock); 3104 ++vcore->num_threads; 3105 spin_unlock(&vcore->lock); 3106 vcpu->arch.vcore = vcore; 3107 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid; 3108 vcpu->arch.thread_cpu = -1; 3109 vcpu->arch.prev_cpu = -1; 3110 3111 vcpu->arch.cpu_type = KVM_CPU_3S_64; 3112 kvmppc_sanity_check(vcpu); 3113 3114 return 0; 3115 } 3116 3117 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode, 3118 unsigned long flags) 3119 { 3120 int err; 3121 int esmt = 0; 3122 3123 if (flags) 3124 return -EINVAL; 3125 if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode)) 3126 return -EINVAL; 3127 if (!cpu_has_feature(CPU_FTR_ARCH_300)) { 3128 /* 3129 * On POWER8 (or POWER7), the threading mode is "strict", 3130 * so we pack smt_mode vcpus per vcore. 3131 */ 3132 if (smt_mode > threads_per_subcore) 3133 return -EINVAL; 3134 } else { 3135 /* 3136 * On POWER9, the threading mode is "loose", 3137 * so each vcpu gets its own vcore. 3138 */ 3139 esmt = smt_mode; 3140 smt_mode = 1; 3141 } 3142 mutex_lock(&kvm->lock); 3143 err = -EBUSY; 3144 if (!kvm->arch.online_vcores) { 3145 kvm->arch.smt_mode = smt_mode; 3146 kvm->arch.emul_smt_mode = esmt; 3147 err = 0; 3148 } 3149 mutex_unlock(&kvm->lock); 3150 3151 return err; 3152 } 3153 3154 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa) 3155 { 3156 if (vpa->pinned_addr) 3157 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa, 3158 vpa->dirty); 3159 } 3160 3161 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu) 3162 { 3163 spin_lock(&vcpu->arch.vpa_update_lock); 3164 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl); 3165 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow); 3166 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa); 3167 spin_unlock(&vcpu->arch.vpa_update_lock); 3168 if (kvmhv_is_nestedv2()) 3169 kvmhv_nestedv2_vcpu_free(vcpu, &vcpu->arch.nestedv2_io); 3170 } 3171 3172 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu) 3173 { 3174 /* Indicate we want to get back into the guest */ 3175 return 1; 3176 } 3177 3178 static void kvmppc_set_timer(struct kvm_vcpu *vcpu) 3179 { 3180 unsigned long dec_nsec, now; 3181 3182 now = get_tb(); 3183 if (now > kvmppc_dec_expires_host_tb(vcpu)) { 3184 /* decrementer has already gone negative */ 3185 kvmppc_core_queue_dec(vcpu); 3186 kvmppc_core_prepare_to_enter(vcpu); 3187 return; 3188 } 3189 dec_nsec = tb_to_ns(kvmppc_dec_expires_host_tb(vcpu) - now); 3190 hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL); 3191 vcpu->arch.timer_running = 1; 3192 } 3193 3194 extern int __kvmppc_vcore_entry(void); 3195 3196 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc, 3197 struct kvm_vcpu *vcpu, u64 tb) 3198 { 3199 u64 now; 3200 3201 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE) 3202 return; 3203 spin_lock_irq(&vcpu->arch.tbacct_lock); 3204 now = tb; 3205 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) - 3206 vcpu->arch.stolen_logged; 3207 vcpu->arch.busy_preempt = now; 3208 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST; 3209 spin_unlock_irq(&vcpu->arch.tbacct_lock); 3210 --vc->n_runnable; 3211 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL); 3212 } 3213 3214 static int kvmppc_grab_hwthread(int cpu) 3215 { 3216 struct paca_struct *tpaca; 3217 long timeout = 10000; 3218 3219 tpaca = paca_ptrs[cpu]; 3220 3221 /* Ensure the thread won't go into the kernel if it wakes */ 3222 tpaca->kvm_hstate.kvm_vcpu = NULL; 3223 tpaca->kvm_hstate.kvm_vcore = NULL; 3224 tpaca->kvm_hstate.napping = 0; 3225 smp_wmb(); 3226 tpaca->kvm_hstate.hwthread_req = 1; 3227 3228 /* 3229 * If the thread is already executing in the kernel (e.g. handling 3230 * a stray interrupt), wait for it to get back to nap mode. 3231 * The smp_mb() is to ensure that our setting of hwthread_req 3232 * is visible before we look at hwthread_state, so if this 3233 * races with the code at system_reset_pSeries and the thread 3234 * misses our setting of hwthread_req, we are sure to see its 3235 * setting of hwthread_state, and vice versa. 3236 */ 3237 smp_mb(); 3238 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) { 3239 if (--timeout <= 0) { 3240 pr_err("KVM: couldn't grab cpu %d\n", cpu); 3241 return -EBUSY; 3242 } 3243 udelay(1); 3244 } 3245 return 0; 3246 } 3247 3248 static void kvmppc_release_hwthread(int cpu) 3249 { 3250 struct paca_struct *tpaca; 3251 3252 tpaca = paca_ptrs[cpu]; 3253 tpaca->kvm_hstate.hwthread_req = 0; 3254 tpaca->kvm_hstate.kvm_vcpu = NULL; 3255 tpaca->kvm_hstate.kvm_vcore = NULL; 3256 tpaca->kvm_hstate.kvm_split_mode = NULL; 3257 } 3258 3259 static DEFINE_PER_CPU(struct kvm *, cpu_in_guest); 3260 3261 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu) 3262 { 3263 struct kvm_nested_guest *nested = vcpu->arch.nested; 3264 cpumask_t *need_tlb_flush; 3265 int i; 3266 3267 if (nested) 3268 need_tlb_flush = &nested->need_tlb_flush; 3269 else 3270 need_tlb_flush = &kvm->arch.need_tlb_flush; 3271 3272 cpu = cpu_first_tlb_thread_sibling(cpu); 3273 for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu); 3274 i += cpu_tlb_thread_sibling_step()) 3275 cpumask_set_cpu(i, need_tlb_flush); 3276 3277 /* 3278 * Make sure setting of bit in need_tlb_flush precedes testing of 3279 * cpu_in_guest. The matching barrier on the other side is hwsync 3280 * when switching to guest MMU mode, which happens between 3281 * cpu_in_guest being set to the guest kvm, and need_tlb_flush bit 3282 * being tested. 3283 */ 3284 smp_mb(); 3285 3286 for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu); 3287 i += cpu_tlb_thread_sibling_step()) { 3288 struct kvm *running = *per_cpu_ptr(&cpu_in_guest, i); 3289 3290 if (running == kvm) 3291 smp_call_function_single(i, do_nothing, NULL, 1); 3292 } 3293 } 3294 3295 static void do_migrate_away_vcpu(void *arg) 3296 { 3297 struct kvm_vcpu *vcpu = arg; 3298 struct kvm *kvm = vcpu->kvm; 3299 3300 /* 3301 * If the guest has GTSE, it may execute tlbie, so do a eieio; tlbsync; 3302 * ptesync sequence on the old CPU before migrating to a new one, in 3303 * case we interrupted the guest between a tlbie ; eieio ; 3304 * tlbsync; ptesync sequence. 3305 * 3306 * Otherwise, ptesync is sufficient for ordering tlbiel sequences. 3307 */ 3308 if (kvm->arch.lpcr & LPCR_GTSE) 3309 asm volatile("eieio; tlbsync; ptesync"); 3310 else 3311 asm volatile("ptesync"); 3312 } 3313 3314 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu) 3315 { 3316 struct kvm_nested_guest *nested = vcpu->arch.nested; 3317 struct kvm *kvm = vcpu->kvm; 3318 int prev_cpu; 3319 3320 if (!cpu_has_feature(CPU_FTR_HVMODE)) 3321 return; 3322 3323 if (nested) 3324 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id]; 3325 else 3326 prev_cpu = vcpu->arch.prev_cpu; 3327 3328 /* 3329 * With radix, the guest can do TLB invalidations itself, 3330 * and it could choose to use the local form (tlbiel) if 3331 * it is invalidating a translation that has only ever been 3332 * used on one vcpu. However, that doesn't mean it has 3333 * only ever been used on one physical cpu, since vcpus 3334 * can move around between pcpus. To cope with this, when 3335 * a vcpu moves from one pcpu to another, we need to tell 3336 * any vcpus running on the same core as this vcpu previously 3337 * ran to flush the TLB. 3338 */ 3339 if (prev_cpu != pcpu) { 3340 if (prev_cpu >= 0) { 3341 if (cpu_first_tlb_thread_sibling(prev_cpu) != 3342 cpu_first_tlb_thread_sibling(pcpu)) 3343 radix_flush_cpu(kvm, prev_cpu, vcpu); 3344 3345 smp_call_function_single(prev_cpu, 3346 do_migrate_away_vcpu, vcpu, 1); 3347 } 3348 if (nested) 3349 nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu; 3350 else 3351 vcpu->arch.prev_cpu = pcpu; 3352 } 3353 } 3354 3355 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc) 3356 { 3357 int cpu; 3358 struct paca_struct *tpaca; 3359 3360 cpu = vc->pcpu; 3361 if (vcpu) { 3362 if (vcpu->arch.timer_running) { 3363 hrtimer_try_to_cancel(&vcpu->arch.dec_timer); 3364 vcpu->arch.timer_running = 0; 3365 } 3366 cpu += vcpu->arch.ptid; 3367 vcpu->cpu = vc->pcpu; 3368 vcpu->arch.thread_cpu = cpu; 3369 } 3370 tpaca = paca_ptrs[cpu]; 3371 tpaca->kvm_hstate.kvm_vcpu = vcpu; 3372 tpaca->kvm_hstate.ptid = cpu - vc->pcpu; 3373 tpaca->kvm_hstate.fake_suspend = 0; 3374 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */ 3375 smp_wmb(); 3376 tpaca->kvm_hstate.kvm_vcore = vc; 3377 if (cpu != smp_processor_id()) 3378 kvmppc_ipi_thread(cpu); 3379 } 3380 3381 static void kvmppc_wait_for_nap(int n_threads) 3382 { 3383 int cpu = smp_processor_id(); 3384 int i, loops; 3385 3386 if (n_threads <= 1) 3387 return; 3388 for (loops = 0; loops < 1000000; ++loops) { 3389 /* 3390 * Check if all threads are finished. 3391 * We set the vcore pointer when starting a thread 3392 * and the thread clears it when finished, so we look 3393 * for any threads that still have a non-NULL vcore ptr. 3394 */ 3395 for (i = 1; i < n_threads; ++i) 3396 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore) 3397 break; 3398 if (i == n_threads) { 3399 HMT_medium(); 3400 return; 3401 } 3402 HMT_low(); 3403 } 3404 HMT_medium(); 3405 for (i = 1; i < n_threads; ++i) 3406 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore) 3407 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i); 3408 } 3409 3410 /* 3411 * Check that we are on thread 0 and that any other threads in 3412 * this core are off-line. Then grab the threads so they can't 3413 * enter the kernel. 3414 */ 3415 static int on_primary_thread(void) 3416 { 3417 int cpu = smp_processor_id(); 3418 int thr; 3419 3420 /* Are we on a primary subcore? */ 3421 if (cpu_thread_in_subcore(cpu)) 3422 return 0; 3423 3424 thr = 0; 3425 while (++thr < threads_per_subcore) 3426 if (cpu_online(cpu + thr)) 3427 return 0; 3428 3429 /* Grab all hw threads so they can't go into the kernel */ 3430 for (thr = 1; thr < threads_per_subcore; ++thr) { 3431 if (kvmppc_grab_hwthread(cpu + thr)) { 3432 /* Couldn't grab one; let the others go */ 3433 do { 3434 kvmppc_release_hwthread(cpu + thr); 3435 } while (--thr > 0); 3436 return 0; 3437 } 3438 } 3439 return 1; 3440 } 3441 3442 /* 3443 * A list of virtual cores for each physical CPU. 3444 * These are vcores that could run but their runner VCPU tasks are 3445 * (or may be) preempted. 3446 */ 3447 struct preempted_vcore_list { 3448 struct list_head list; 3449 spinlock_t lock; 3450 }; 3451 3452 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores); 3453 3454 static void init_vcore_lists(void) 3455 { 3456 int cpu; 3457 3458 for_each_possible_cpu(cpu) { 3459 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu); 3460 spin_lock_init(&lp->lock); 3461 INIT_LIST_HEAD(&lp->list); 3462 } 3463 } 3464 3465 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc) 3466 { 3467 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores); 3468 3469 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)); 3470 3471 vc->vcore_state = VCORE_PREEMPT; 3472 vc->pcpu = smp_processor_id(); 3473 if (vc->num_threads < threads_per_vcore(vc->kvm)) { 3474 spin_lock(&lp->lock); 3475 list_add_tail(&vc->preempt_list, &lp->list); 3476 spin_unlock(&lp->lock); 3477 } 3478 3479 /* Start accumulating stolen time */ 3480 kvmppc_core_start_stolen(vc, mftb()); 3481 } 3482 3483 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc) 3484 { 3485 struct preempted_vcore_list *lp; 3486 3487 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)); 3488 3489 kvmppc_core_end_stolen(vc, mftb()); 3490 if (!list_empty(&vc->preempt_list)) { 3491 lp = &per_cpu(preempted_vcores, vc->pcpu); 3492 spin_lock(&lp->lock); 3493 list_del_init(&vc->preempt_list); 3494 spin_unlock(&lp->lock); 3495 } 3496 vc->vcore_state = VCORE_INACTIVE; 3497 } 3498 3499 /* 3500 * This stores information about the virtual cores currently 3501 * assigned to a physical core. 3502 */ 3503 struct core_info { 3504 int n_subcores; 3505 int max_subcore_threads; 3506 int total_threads; 3507 int subcore_threads[MAX_SUBCORES]; 3508 struct kvmppc_vcore *vc[MAX_SUBCORES]; 3509 }; 3510 3511 /* 3512 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7 3513 * respectively in 2-way micro-threading (split-core) mode on POWER8. 3514 */ 3515 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 }; 3516 3517 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc) 3518 { 3519 memset(cip, 0, sizeof(*cip)); 3520 cip->n_subcores = 1; 3521 cip->max_subcore_threads = vc->num_threads; 3522 cip->total_threads = vc->num_threads; 3523 cip->subcore_threads[0] = vc->num_threads; 3524 cip->vc[0] = vc; 3525 } 3526 3527 static bool subcore_config_ok(int n_subcores, int n_threads) 3528 { 3529 /* 3530 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way 3531 * split-core mode, with one thread per subcore. 3532 */ 3533 if (cpu_has_feature(CPU_FTR_ARCH_300)) 3534 return n_subcores <= 4 && n_threads == 1; 3535 3536 /* On POWER8, can only dynamically split if unsplit to begin with */ 3537 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS) 3538 return false; 3539 if (n_subcores > MAX_SUBCORES) 3540 return false; 3541 if (n_subcores > 1) { 3542 if (!(dynamic_mt_modes & 2)) 3543 n_subcores = 4; 3544 if (n_subcores > 2 && !(dynamic_mt_modes & 4)) 3545 return false; 3546 } 3547 3548 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS; 3549 } 3550 3551 static void init_vcore_to_run(struct kvmppc_vcore *vc) 3552 { 3553 vc->entry_exit_map = 0; 3554 vc->in_guest = 0; 3555 vc->napping_threads = 0; 3556 vc->conferring_threads = 0; 3557 vc->tb_offset_applied = 0; 3558 } 3559 3560 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip) 3561 { 3562 int n_threads = vc->num_threads; 3563 int sub; 3564 3565 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) 3566 return false; 3567 3568 /* In one_vm_per_core mode, require all vcores to be from the same vm */ 3569 if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm) 3570 return false; 3571 3572 if (n_threads < cip->max_subcore_threads) 3573 n_threads = cip->max_subcore_threads; 3574 if (!subcore_config_ok(cip->n_subcores + 1, n_threads)) 3575 return false; 3576 cip->max_subcore_threads = n_threads; 3577 3578 sub = cip->n_subcores; 3579 ++cip->n_subcores; 3580 cip->total_threads += vc->num_threads; 3581 cip->subcore_threads[sub] = vc->num_threads; 3582 cip->vc[sub] = vc; 3583 init_vcore_to_run(vc); 3584 list_del_init(&vc->preempt_list); 3585 3586 return true; 3587 } 3588 3589 /* 3590 * Work out whether it is possible to piggyback the execution of 3591 * vcore *pvc onto the execution of the other vcores described in *cip. 3592 */ 3593 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip, 3594 int target_threads) 3595 { 3596 if (cip->total_threads + pvc->num_threads > target_threads) 3597 return false; 3598 3599 return can_dynamic_split(pvc, cip); 3600 } 3601 3602 static void prepare_threads(struct kvmppc_vcore *vc) 3603 { 3604 int i; 3605 struct kvm_vcpu *vcpu; 3606 3607 for_each_runnable_thread(i, vcpu, vc) { 3608 if (signal_pending(vcpu->arch.run_task)) 3609 vcpu->arch.ret = -EINTR; 3610 else if (vcpu->arch.vpa.update_pending || 3611 vcpu->arch.slb_shadow.update_pending || 3612 vcpu->arch.dtl.update_pending) 3613 vcpu->arch.ret = RESUME_GUEST; 3614 else 3615 continue; 3616 kvmppc_remove_runnable(vc, vcpu, mftb()); 3617 wake_up(&vcpu->arch.cpu_run); 3618 } 3619 } 3620 3621 static void collect_piggybacks(struct core_info *cip, int target_threads) 3622 { 3623 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores); 3624 struct kvmppc_vcore *pvc, *vcnext; 3625 3626 spin_lock(&lp->lock); 3627 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) { 3628 if (!spin_trylock(&pvc->lock)) 3629 continue; 3630 prepare_threads(pvc); 3631 if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) { 3632 list_del_init(&pvc->preempt_list); 3633 if (pvc->runner == NULL) { 3634 pvc->vcore_state = VCORE_INACTIVE; 3635 kvmppc_core_end_stolen(pvc, mftb()); 3636 } 3637 spin_unlock(&pvc->lock); 3638 continue; 3639 } 3640 if (!can_piggyback(pvc, cip, target_threads)) { 3641 spin_unlock(&pvc->lock); 3642 continue; 3643 } 3644 kvmppc_core_end_stolen(pvc, mftb()); 3645 pvc->vcore_state = VCORE_PIGGYBACK; 3646 if (cip->total_threads >= target_threads) 3647 break; 3648 } 3649 spin_unlock(&lp->lock); 3650 } 3651 3652 static bool recheck_signals_and_mmu(struct core_info *cip) 3653 { 3654 int sub, i; 3655 struct kvm_vcpu *vcpu; 3656 struct kvmppc_vcore *vc; 3657 3658 for (sub = 0; sub < cip->n_subcores; ++sub) { 3659 vc = cip->vc[sub]; 3660 if (!vc->kvm->arch.mmu_ready) 3661 return true; 3662 for_each_runnable_thread(i, vcpu, vc) 3663 if (signal_pending(vcpu->arch.run_task)) 3664 return true; 3665 } 3666 return false; 3667 } 3668 3669 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master) 3670 { 3671 int still_running = 0, i; 3672 u64 now; 3673 long ret; 3674 struct kvm_vcpu *vcpu; 3675 3676 spin_lock(&vc->lock); 3677 now = get_tb(); 3678 for_each_runnable_thread(i, vcpu, vc) { 3679 /* 3680 * It's safe to unlock the vcore in the loop here, because 3681 * for_each_runnable_thread() is safe against removal of 3682 * the vcpu, and the vcore state is VCORE_EXITING here, 3683 * so any vcpus becoming runnable will have their arch.trap 3684 * set to zero and can't actually run in the guest. 3685 */ 3686 spin_unlock(&vc->lock); 3687 /* cancel pending dec exception if dec is positive */ 3688 if (now < kvmppc_dec_expires_host_tb(vcpu) && 3689 kvmppc_core_pending_dec(vcpu)) 3690 kvmppc_core_dequeue_dec(vcpu); 3691 3692 trace_kvm_guest_exit(vcpu); 3693 3694 ret = RESUME_GUEST; 3695 if (vcpu->arch.trap) 3696 ret = kvmppc_handle_exit_hv(vcpu, 3697 vcpu->arch.run_task); 3698 3699 vcpu->arch.ret = ret; 3700 vcpu->arch.trap = 0; 3701 3702 spin_lock(&vc->lock); 3703 if (is_kvmppc_resume_guest(vcpu->arch.ret)) { 3704 if (vcpu->arch.pending_exceptions) 3705 kvmppc_core_prepare_to_enter(vcpu); 3706 if (vcpu->arch.ceded) 3707 kvmppc_set_timer(vcpu); 3708 else 3709 ++still_running; 3710 } else { 3711 kvmppc_remove_runnable(vc, vcpu, mftb()); 3712 wake_up(&vcpu->arch.cpu_run); 3713 } 3714 } 3715 if (!is_master) { 3716 if (still_running > 0) { 3717 kvmppc_vcore_preempt(vc); 3718 } else if (vc->runner) { 3719 vc->vcore_state = VCORE_PREEMPT; 3720 kvmppc_core_start_stolen(vc, mftb()); 3721 } else { 3722 vc->vcore_state = VCORE_INACTIVE; 3723 } 3724 if (vc->n_runnable > 0 && vc->runner == NULL) { 3725 /* make sure there's a candidate runner awake */ 3726 i = -1; 3727 vcpu = next_runnable_thread(vc, &i); 3728 wake_up(&vcpu->arch.cpu_run); 3729 } 3730 } 3731 spin_unlock(&vc->lock); 3732 } 3733 3734 /* 3735 * Clear core from the list of active host cores as we are about to 3736 * enter the guest. Only do this if it is the primary thread of the 3737 * core (not if a subcore) that is entering the guest. 3738 */ 3739 static inline int kvmppc_clear_host_core(unsigned int cpu) 3740 { 3741 int core; 3742 3743 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu)) 3744 return 0; 3745 /* 3746 * Memory barrier can be omitted here as we will do a smp_wmb() 3747 * later in kvmppc_start_thread and we need ensure that state is 3748 * visible to other CPUs only after we enter guest. 3749 */ 3750 core = cpu >> threads_shift; 3751 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0; 3752 return 0; 3753 } 3754 3755 /* 3756 * Advertise this core as an active host core since we exited the guest 3757 * Only need to do this if it is the primary thread of the core that is 3758 * exiting. 3759 */ 3760 static inline int kvmppc_set_host_core(unsigned int cpu) 3761 { 3762 int core; 3763 3764 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu)) 3765 return 0; 3766 3767 /* 3768 * Memory barrier can be omitted here because we do a spin_unlock 3769 * immediately after this which provides the memory barrier. 3770 */ 3771 core = cpu >> threads_shift; 3772 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1; 3773 return 0; 3774 } 3775 3776 static void set_irq_happened(int trap) 3777 { 3778 switch (trap) { 3779 case BOOK3S_INTERRUPT_EXTERNAL: 3780 local_paca->irq_happened |= PACA_IRQ_EE; 3781 break; 3782 case BOOK3S_INTERRUPT_H_DOORBELL: 3783 local_paca->irq_happened |= PACA_IRQ_DBELL; 3784 break; 3785 case BOOK3S_INTERRUPT_HMI: 3786 local_paca->irq_happened |= PACA_IRQ_HMI; 3787 break; 3788 case BOOK3S_INTERRUPT_SYSTEM_RESET: 3789 replay_system_reset(); 3790 break; 3791 } 3792 } 3793 3794 /* 3795 * Run a set of guest threads on a physical core. 3796 * Called with vc->lock held. 3797 */ 3798 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc) 3799 { 3800 struct kvm_vcpu *vcpu; 3801 int i; 3802 int srcu_idx; 3803 struct core_info core_info; 3804 struct kvmppc_vcore *pvc; 3805 struct kvm_split_mode split_info, *sip; 3806 int split, subcore_size, active; 3807 int sub; 3808 bool thr0_done; 3809 unsigned long cmd_bit, stat_bit; 3810 int pcpu, thr; 3811 int target_threads; 3812 int controlled_threads; 3813 int trap; 3814 bool is_power8; 3815 3816 if (WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300))) 3817 return; 3818 3819 /* 3820 * Remove from the list any threads that have a signal pending 3821 * or need a VPA update done 3822 */ 3823 prepare_threads(vc); 3824 3825 /* if the runner is no longer runnable, let the caller pick a new one */ 3826 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE) 3827 return; 3828 3829 /* 3830 * Initialize *vc. 3831 */ 3832 init_vcore_to_run(vc); 3833 vc->preempt_tb = TB_NIL; 3834 3835 /* 3836 * Number of threads that we will be controlling: the same as 3837 * the number of threads per subcore, except on POWER9, 3838 * where it's 1 because the threads are (mostly) independent. 3839 */ 3840 controlled_threads = threads_per_vcore(vc->kvm); 3841 3842 /* 3843 * Make sure we are running on primary threads, and that secondary 3844 * threads are offline. Also check if the number of threads in this 3845 * guest are greater than the current system threads per guest. 3846 */ 3847 if ((controlled_threads > 1) && 3848 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) { 3849 for_each_runnable_thread(i, vcpu, vc) { 3850 vcpu->arch.ret = -EBUSY; 3851 kvmppc_remove_runnable(vc, vcpu, mftb()); 3852 wake_up(&vcpu->arch.cpu_run); 3853 } 3854 goto out; 3855 } 3856 3857 /* 3858 * See if we could run any other vcores on the physical core 3859 * along with this one. 3860 */ 3861 init_core_info(&core_info, vc); 3862 pcpu = smp_processor_id(); 3863 target_threads = controlled_threads; 3864 if (target_smt_mode && target_smt_mode < target_threads) 3865 target_threads = target_smt_mode; 3866 if (vc->num_threads < target_threads) 3867 collect_piggybacks(&core_info, target_threads); 3868 3869 /* 3870 * Hard-disable interrupts, and check resched flag and signals. 3871 * If we need to reschedule or deliver a signal, clean up 3872 * and return without going into the guest(s). 3873 * If the mmu_ready flag has been cleared, don't go into the 3874 * guest because that means a HPT resize operation is in progress. 3875 */ 3876 local_irq_disable(); 3877 hard_irq_disable(); 3878 xfer_to_guest_mode_prepare(); 3879 if (lazy_irq_pending() || xfer_to_guest_mode_work_pending() || 3880 recheck_signals_and_mmu(&core_info)) { 3881 local_irq_enable(); 3882 vc->vcore_state = VCORE_INACTIVE; 3883 /* Unlock all except the primary vcore */ 3884 for (sub = 1; sub < core_info.n_subcores; ++sub) { 3885 pvc = core_info.vc[sub]; 3886 /* Put back on to the preempted vcores list */ 3887 kvmppc_vcore_preempt(pvc); 3888 spin_unlock(&pvc->lock); 3889 } 3890 for (i = 0; i < controlled_threads; ++i) 3891 kvmppc_release_hwthread(pcpu + i); 3892 return; 3893 } 3894 3895 kvmppc_clear_host_core(pcpu); 3896 3897 /* Decide on micro-threading (split-core) mode */ 3898 subcore_size = threads_per_subcore; 3899 cmd_bit = stat_bit = 0; 3900 split = core_info.n_subcores; 3901 sip = NULL; 3902 is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S); 3903 3904 if (split > 1) { 3905 sip = &split_info; 3906 memset(&split_info, 0, sizeof(split_info)); 3907 for (sub = 0; sub < core_info.n_subcores; ++sub) 3908 split_info.vc[sub] = core_info.vc[sub]; 3909 3910 if (is_power8) { 3911 if (split == 2 && (dynamic_mt_modes & 2)) { 3912 cmd_bit = HID0_POWER8_1TO2LPAR; 3913 stat_bit = HID0_POWER8_2LPARMODE; 3914 } else { 3915 split = 4; 3916 cmd_bit = HID0_POWER8_1TO4LPAR; 3917 stat_bit = HID0_POWER8_4LPARMODE; 3918 } 3919 subcore_size = MAX_SMT_THREADS / split; 3920 split_info.rpr = mfspr(SPRN_RPR); 3921 split_info.pmmar = mfspr(SPRN_PMMAR); 3922 split_info.ldbar = mfspr(SPRN_LDBAR); 3923 split_info.subcore_size = subcore_size; 3924 } else { 3925 split_info.subcore_size = 1; 3926 } 3927 3928 /* order writes to split_info before kvm_split_mode pointer */ 3929 smp_wmb(); 3930 } 3931 3932 for (thr = 0; thr < controlled_threads; ++thr) { 3933 struct paca_struct *paca = paca_ptrs[pcpu + thr]; 3934 3935 paca->kvm_hstate.napping = 0; 3936 paca->kvm_hstate.kvm_split_mode = sip; 3937 } 3938 3939 /* Initiate micro-threading (split-core) on POWER8 if required */ 3940 if (cmd_bit) { 3941 unsigned long hid0 = mfspr(SPRN_HID0); 3942 3943 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS; 3944 mb(); 3945 mtspr(SPRN_HID0, hid0); 3946 isync(); 3947 for (;;) { 3948 hid0 = mfspr(SPRN_HID0); 3949 if (hid0 & stat_bit) 3950 break; 3951 cpu_relax(); 3952 } 3953 } 3954 3955 /* 3956 * On POWER8, set RWMR register. 3957 * Since it only affects PURR and SPURR, it doesn't affect 3958 * the host, so we don't save/restore the host value. 3959 */ 3960 if (is_power8) { 3961 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD; 3962 int n_online = atomic_read(&vc->online_count); 3963 3964 /* 3965 * Use the 8-thread value if we're doing split-core 3966 * or if the vcore's online count looks bogus. 3967 */ 3968 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS && 3969 n_online >= 1 && n_online <= MAX_SMT_THREADS) 3970 rwmr_val = p8_rwmr_values[n_online]; 3971 mtspr(SPRN_RWMR, rwmr_val); 3972 } 3973 3974 /* Start all the threads */ 3975 active = 0; 3976 for (sub = 0; sub < core_info.n_subcores; ++sub) { 3977 thr = is_power8 ? subcore_thread_map[sub] : sub; 3978 thr0_done = false; 3979 active |= 1 << thr; 3980 pvc = core_info.vc[sub]; 3981 pvc->pcpu = pcpu + thr; 3982 for_each_runnable_thread(i, vcpu, pvc) { 3983 /* 3984 * XXX: is kvmppc_start_thread called too late here? 3985 * It updates vcpu->cpu and vcpu->arch.thread_cpu 3986 * which are used by kvmppc_fast_vcpu_kick_hv(), but 3987 * kick is called after new exceptions become available 3988 * and exceptions are checked earlier than here, by 3989 * kvmppc_core_prepare_to_enter. 3990 */ 3991 kvmppc_start_thread(vcpu, pvc); 3992 kvmppc_update_vpa_dispatch(vcpu, pvc); 3993 trace_kvm_guest_enter(vcpu); 3994 if (!vcpu->arch.ptid) 3995 thr0_done = true; 3996 active |= 1 << (thr + vcpu->arch.ptid); 3997 } 3998 /* 3999 * We need to start the first thread of each subcore 4000 * even if it doesn't have a vcpu. 4001 */ 4002 if (!thr0_done) 4003 kvmppc_start_thread(NULL, pvc); 4004 } 4005 4006 /* 4007 * Ensure that split_info.do_nap is set after setting 4008 * the vcore pointer in the PACA of the secondaries. 4009 */ 4010 smp_mb(); 4011 4012 /* 4013 * When doing micro-threading, poke the inactive threads as well. 4014 * This gets them to the nap instruction after kvm_do_nap, 4015 * which reduces the time taken to unsplit later. 4016 */ 4017 if (cmd_bit) { 4018 split_info.do_nap = 1; /* ask secondaries to nap when done */ 4019 for (thr = 1; thr < threads_per_subcore; ++thr) 4020 if (!(active & (1 << thr))) 4021 kvmppc_ipi_thread(pcpu + thr); 4022 } 4023 4024 vc->vcore_state = VCORE_RUNNING; 4025 preempt_disable(); 4026 4027 trace_kvmppc_run_core(vc, 0); 4028 4029 for (sub = 0; sub < core_info.n_subcores; ++sub) 4030 spin_unlock(&core_info.vc[sub]->lock); 4031 4032 guest_timing_enter_irqoff(); 4033 4034 srcu_idx = srcu_read_lock(&vc->kvm->srcu); 4035 4036 guest_state_enter_irqoff(); 4037 this_cpu_disable_ftrace(); 4038 4039 trap = __kvmppc_vcore_entry(); 4040 4041 this_cpu_enable_ftrace(); 4042 guest_state_exit_irqoff(); 4043 4044 srcu_read_unlock(&vc->kvm->srcu, srcu_idx); 4045 4046 set_irq_happened(trap); 4047 4048 spin_lock(&vc->lock); 4049 /* prevent other vcpu threads from doing kvmppc_start_thread() now */ 4050 vc->vcore_state = VCORE_EXITING; 4051 4052 /* wait for secondary threads to finish writing their state to memory */ 4053 kvmppc_wait_for_nap(controlled_threads); 4054 4055 /* Return to whole-core mode if we split the core earlier */ 4056 if (cmd_bit) { 4057 unsigned long hid0 = mfspr(SPRN_HID0); 4058 4059 hid0 &= ~HID0_POWER8_DYNLPARDIS; 4060 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE; 4061 mb(); 4062 mtspr(SPRN_HID0, hid0); 4063 isync(); 4064 for (;;) { 4065 hid0 = mfspr(SPRN_HID0); 4066 if (!(hid0 & stat_bit)) 4067 break; 4068 cpu_relax(); 4069 } 4070 split_info.do_nap = 0; 4071 } 4072 4073 kvmppc_set_host_core(pcpu); 4074 4075 if (!vtime_accounting_enabled_this_cpu()) { 4076 local_irq_enable(); 4077 /* 4078 * Service IRQs here before guest_timing_exit_irqoff() so any 4079 * ticks that occurred while running the guest are accounted to 4080 * the guest. If vtime accounting is enabled, accounting uses 4081 * TB rather than ticks, so it can be done without enabling 4082 * interrupts here, which has the problem that it accounts 4083 * interrupt processing overhead to the host. 4084 */ 4085 local_irq_disable(); 4086 } 4087 guest_timing_exit_irqoff(); 4088 4089 local_irq_enable(); 4090 4091 /* Let secondaries go back to the offline loop */ 4092 for (i = 0; i < controlled_threads; ++i) { 4093 kvmppc_release_hwthread(pcpu + i); 4094 if (sip && sip->napped[i]) 4095 kvmppc_ipi_thread(pcpu + i); 4096 } 4097 4098 spin_unlock(&vc->lock); 4099 4100 /* make sure updates to secondary vcpu structs are visible now */ 4101 smp_mb(); 4102 4103 preempt_enable(); 4104 4105 for (sub = 0; sub < core_info.n_subcores; ++sub) { 4106 pvc = core_info.vc[sub]; 4107 post_guest_process(pvc, pvc == vc); 4108 } 4109 4110 spin_lock(&vc->lock); 4111 4112 out: 4113 vc->vcore_state = VCORE_INACTIVE; 4114 trace_kvmppc_run_core(vc, 1); 4115 } 4116 4117 static inline bool hcall_is_xics(unsigned long req) 4118 { 4119 return req == H_EOI || req == H_CPPR || req == H_IPI || 4120 req == H_IPOLL || req == H_XIRR || req == H_XIRR_X; 4121 } 4122 4123 static void vcpu_vpa_increment_dispatch(struct kvm_vcpu *vcpu) 4124 { 4125 struct lppaca *lp = vcpu->arch.vpa.pinned_addr; 4126 if (lp) { 4127 u32 yield_count = be32_to_cpu(lp->yield_count) + 1; 4128 lp->yield_count = cpu_to_be32(yield_count); 4129 vcpu->arch.vpa.dirty = 1; 4130 } 4131 } 4132 4133 /* Helper functions for reading L2's stats from L1's VPA */ 4134 #ifdef CONFIG_PPC_PSERIES 4135 static DEFINE_PER_CPU(u64, l1_to_l2_cs); 4136 static DEFINE_PER_CPU(u64, l2_to_l1_cs); 4137 static DEFINE_PER_CPU(u64, l2_runtime_agg); 4138 4139 int kvmhv_get_l2_counters_status(void) 4140 { 4141 return firmware_has_feature(FW_FEATURE_LPAR) && 4142 get_lppaca()->l2_counters_enable; 4143 } 4144 4145 void kvmhv_set_l2_counters_status(int cpu, bool status) 4146 { 4147 if (!firmware_has_feature(FW_FEATURE_LPAR)) 4148 return; 4149 if (status) 4150 lppaca_of(cpu).l2_counters_enable = 1; 4151 else 4152 lppaca_of(cpu).l2_counters_enable = 0; 4153 } 4154 EXPORT_SYMBOL(kvmhv_set_l2_counters_status); 4155 4156 int kvmhv_counters_tracepoint_regfunc(void) 4157 { 4158 int cpu; 4159 4160 for_each_present_cpu(cpu) { 4161 kvmhv_set_l2_counters_status(cpu, true); 4162 } 4163 return 0; 4164 } 4165 4166 void kvmhv_counters_tracepoint_unregfunc(void) 4167 { 4168 int cpu; 4169 4170 for_each_present_cpu(cpu) { 4171 kvmhv_set_l2_counters_status(cpu, false); 4172 } 4173 } 4174 4175 static void do_trace_nested_cs_time(struct kvm_vcpu *vcpu) 4176 { 4177 struct lppaca *lp = get_lppaca(); 4178 u64 l1_to_l2_ns, l2_to_l1_ns, l2_runtime_ns; 4179 u64 *l1_to_l2_cs_ptr = this_cpu_ptr(&l1_to_l2_cs); 4180 u64 *l2_to_l1_cs_ptr = this_cpu_ptr(&l2_to_l1_cs); 4181 u64 *l2_runtime_agg_ptr = this_cpu_ptr(&l2_runtime_agg); 4182 4183 l1_to_l2_ns = tb_to_ns(be64_to_cpu(lp->l1_to_l2_cs_tb)); 4184 l2_to_l1_ns = tb_to_ns(be64_to_cpu(lp->l2_to_l1_cs_tb)); 4185 l2_runtime_ns = tb_to_ns(be64_to_cpu(lp->l2_runtime_tb)); 4186 trace_kvmppc_vcpu_stats(vcpu, l1_to_l2_ns - *l1_to_l2_cs_ptr, 4187 l2_to_l1_ns - *l2_to_l1_cs_ptr, 4188 l2_runtime_ns - *l2_runtime_agg_ptr); 4189 *l1_to_l2_cs_ptr = l1_to_l2_ns; 4190 *l2_to_l1_cs_ptr = l2_to_l1_ns; 4191 *l2_runtime_agg_ptr = l2_runtime_ns; 4192 vcpu->arch.l1_to_l2_cs = l1_to_l2_ns; 4193 vcpu->arch.l2_to_l1_cs = l2_to_l1_ns; 4194 vcpu->arch.l2_runtime_agg = l2_runtime_ns; 4195 } 4196 4197 u64 kvmhv_get_l1_to_l2_cs_time(void) 4198 { 4199 return tb_to_ns(be64_to_cpu(get_lppaca()->l1_to_l2_cs_tb)); 4200 } 4201 EXPORT_SYMBOL(kvmhv_get_l1_to_l2_cs_time); 4202 4203 u64 kvmhv_get_l2_to_l1_cs_time(void) 4204 { 4205 return tb_to_ns(be64_to_cpu(get_lppaca()->l2_to_l1_cs_tb)); 4206 } 4207 EXPORT_SYMBOL(kvmhv_get_l2_to_l1_cs_time); 4208 4209 u64 kvmhv_get_l2_runtime_agg(void) 4210 { 4211 return tb_to_ns(be64_to_cpu(get_lppaca()->l2_runtime_tb)); 4212 } 4213 EXPORT_SYMBOL(kvmhv_get_l2_runtime_agg); 4214 4215 u64 kvmhv_get_l1_to_l2_cs_time_vcpu(void) 4216 { 4217 struct kvm_vcpu *vcpu; 4218 struct kvm_vcpu_arch *arch; 4219 4220 vcpu = local_paca->kvm_hstate.kvm_vcpu; 4221 if (vcpu) { 4222 arch = &vcpu->arch; 4223 return arch->l1_to_l2_cs; 4224 } else { 4225 return 0; 4226 } 4227 } 4228 EXPORT_SYMBOL(kvmhv_get_l1_to_l2_cs_time_vcpu); 4229 4230 u64 kvmhv_get_l2_to_l1_cs_time_vcpu(void) 4231 { 4232 struct kvm_vcpu *vcpu; 4233 struct kvm_vcpu_arch *arch; 4234 4235 vcpu = local_paca->kvm_hstate.kvm_vcpu; 4236 if (vcpu) { 4237 arch = &vcpu->arch; 4238 return arch->l2_to_l1_cs; 4239 } else { 4240 return 0; 4241 } 4242 } 4243 EXPORT_SYMBOL(kvmhv_get_l2_to_l1_cs_time_vcpu); 4244 4245 u64 kvmhv_get_l2_runtime_agg_vcpu(void) 4246 { 4247 struct kvm_vcpu *vcpu; 4248 struct kvm_vcpu_arch *arch; 4249 4250 vcpu = local_paca->kvm_hstate.kvm_vcpu; 4251 if (vcpu) { 4252 arch = &vcpu->arch; 4253 return arch->l2_runtime_agg; 4254 } else { 4255 return 0; 4256 } 4257 } 4258 EXPORT_SYMBOL(kvmhv_get_l2_runtime_agg_vcpu); 4259 4260 #else 4261 int kvmhv_get_l2_counters_status(void) 4262 { 4263 return 0; 4264 } 4265 4266 static void do_trace_nested_cs_time(struct kvm_vcpu *vcpu) 4267 { 4268 } 4269 #endif 4270 4271 static int kvmhv_vcpu_entry_nestedv2(struct kvm_vcpu *vcpu, u64 time_limit, 4272 unsigned long lpcr, u64 *tb) 4273 { 4274 struct kvmhv_nestedv2_io *io; 4275 unsigned long msr, i; 4276 int trap; 4277 long rc; 4278 4279 if (vcpu->arch.doorbell_request) { 4280 vcpu->arch.doorbell_request = 0; 4281 kvmppc_set_dpdes(vcpu, 1); 4282 } 4283 4284 io = &vcpu->arch.nestedv2_io; 4285 4286 msr = mfmsr(); 4287 kvmppc_msr_hard_disable_set_facilities(vcpu, msr); 4288 if (lazy_irq_pending()) 4289 return 0; 4290 4291 rc = kvmhv_nestedv2_flush_vcpu(vcpu, time_limit); 4292 if (rc < 0) 4293 return -EINVAL; 4294 4295 kvmppc_gse_put_u64(io->vcpu_run_input, KVMPPC_GSID_LPCR, lpcr); 4296 4297 accumulate_time(vcpu, &vcpu->arch.in_guest); 4298 rc = plpar_guest_run_vcpu(0, vcpu->kvm->arch.lpid, vcpu->vcpu_id, 4299 &trap, &i); 4300 4301 if (rc != H_SUCCESS) { 4302 pr_err("KVM Guest Run VCPU hcall failed\n"); 4303 if (rc == H_INVALID_ELEMENT_ID) 4304 pr_err("KVM: Guest Run VCPU invalid element id at %ld\n", i); 4305 else if (rc == H_INVALID_ELEMENT_SIZE) 4306 pr_err("KVM: Guest Run VCPU invalid element size at %ld\n", i); 4307 else if (rc == H_INVALID_ELEMENT_VALUE) 4308 pr_err("KVM: Guest Run VCPU invalid element value at %ld\n", i); 4309 return -EINVAL; 4310 } 4311 accumulate_time(vcpu, &vcpu->arch.guest_exit); 4312 4313 *tb = mftb(); 4314 kvmppc_gsm_reset(io->vcpu_message); 4315 kvmppc_gsm_reset(io->vcore_message); 4316 kvmppc_gsbm_zero(&io->valids); 4317 4318 rc = kvmhv_nestedv2_parse_output(vcpu); 4319 if (rc < 0) 4320 return -EINVAL; 4321 4322 timer_rearm_host_dec(*tb); 4323 4324 /* Record context switch and guest_run_time data */ 4325 if (kvmhv_get_l2_counters_status()) 4326 do_trace_nested_cs_time(vcpu); 4327 4328 return trap; 4329 } 4330 4331 /* call our hypervisor to load up HV regs and go */ 4332 static int kvmhv_vcpu_entry_p9_nested(struct kvm_vcpu *vcpu, u64 time_limit, unsigned long lpcr, u64 *tb) 4333 { 4334 unsigned long host_psscr; 4335 unsigned long msr; 4336 struct hv_guest_state hvregs; 4337 struct p9_host_os_sprs host_os_sprs; 4338 s64 dec; 4339 int trap; 4340 4341 msr = mfmsr(); 4342 4343 save_p9_host_os_sprs(&host_os_sprs); 4344 4345 /* 4346 * We need to save and restore the guest visible part of the 4347 * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor 4348 * doesn't do this for us. Note only required if pseries since 4349 * this is done in kvmhv_vcpu_entry_p9() below otherwise. 4350 */ 4351 host_psscr = mfspr(SPRN_PSSCR_PR); 4352 4353 kvmppc_msr_hard_disable_set_facilities(vcpu, msr); 4354 if (lazy_irq_pending()) 4355 return 0; 4356 4357 if (unlikely(load_vcpu_state(vcpu, &host_os_sprs))) 4358 msr = mfmsr(); /* TM restore can update msr */ 4359 4360 if (vcpu->arch.psscr != host_psscr) 4361 mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr); 4362 4363 kvmhv_save_hv_regs(vcpu, &hvregs); 4364 hvregs.lpcr = lpcr; 4365 hvregs.amor = ~0; 4366 vcpu->arch.regs.msr = vcpu->arch.shregs.msr; 4367 hvregs.version = HV_GUEST_STATE_VERSION; 4368 if (vcpu->arch.nested) { 4369 hvregs.lpid = vcpu->arch.nested->shadow_lpid; 4370 hvregs.vcpu_token = vcpu->arch.nested_vcpu_id; 4371 } else { 4372 hvregs.lpid = vcpu->kvm->arch.lpid; 4373 hvregs.vcpu_token = vcpu->vcpu_id; 4374 } 4375 hvregs.hdec_expiry = time_limit; 4376 4377 /* 4378 * hvregs has the doorbell status, so zero it here which 4379 * enables us to receive doorbells when H_ENTER_NESTED is 4380 * in progress for this vCPU 4381 */ 4382 4383 if (vcpu->arch.doorbell_request) 4384 vcpu->arch.doorbell_request = 0; 4385 4386 /* 4387 * When setting DEC, we must always deal with irq_work_raise 4388 * via NMI vs setting DEC. The problem occurs right as we 4389 * switch into guest mode if a NMI hits and sets pending work 4390 * and sets DEC, then that will apply to the guest and not 4391 * bring us back to the host. 4392 * 4393 * irq_work_raise could check a flag (or possibly LPCR[HDICE] 4394 * for example) and set HDEC to 1? That wouldn't solve the 4395 * nested hv case which needs to abort the hcall or zero the 4396 * time limit. 4397 * 4398 * XXX: Another day's problem. 4399 */ 4400 mtspr(SPRN_DEC, kvmppc_dec_expires_host_tb(vcpu) - *tb); 4401 4402 mtspr(SPRN_DAR, vcpu->arch.shregs.dar); 4403 mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr); 4404 switch_pmu_to_guest(vcpu, &host_os_sprs); 4405 accumulate_time(vcpu, &vcpu->arch.in_guest); 4406 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs), 4407 __pa(&vcpu->arch.regs)); 4408 accumulate_time(vcpu, &vcpu->arch.guest_exit); 4409 kvmhv_restore_hv_return_state(vcpu, &hvregs); 4410 switch_pmu_to_host(vcpu, &host_os_sprs); 4411 vcpu->arch.shregs.msr = vcpu->arch.regs.msr; 4412 vcpu->arch.shregs.dar = mfspr(SPRN_DAR); 4413 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR); 4414 vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR); 4415 4416 store_vcpu_state(vcpu); 4417 4418 dec = mfspr(SPRN_DEC); 4419 if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */ 4420 dec = (s32) dec; 4421 *tb = mftb(); 4422 vcpu->arch.dec_expires = dec + (*tb + kvmppc_get_tb_offset(vcpu)); 4423 4424 timer_rearm_host_dec(*tb); 4425 4426 restore_p9_host_os_sprs(vcpu, &host_os_sprs); 4427 if (vcpu->arch.psscr != host_psscr) 4428 mtspr(SPRN_PSSCR_PR, host_psscr); 4429 4430 return trap; 4431 } 4432 4433 /* 4434 * Guest entry for POWER9 and later CPUs. 4435 */ 4436 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit, 4437 unsigned long lpcr, u64 *tb) 4438 { 4439 struct kvm *kvm = vcpu->kvm; 4440 struct kvm_nested_guest *nested = vcpu->arch.nested; 4441 u64 next_timer; 4442 int trap; 4443 4444 next_timer = timer_get_next_tb(); 4445 if (*tb >= next_timer) 4446 return BOOK3S_INTERRUPT_HV_DECREMENTER; 4447 if (next_timer < time_limit) 4448 time_limit = next_timer; 4449 else if (*tb >= time_limit) /* nested time limit */ 4450 return BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER; 4451 4452 vcpu->arch.ceded = 0; 4453 4454 vcpu_vpa_increment_dispatch(vcpu); 4455 4456 if (kvmhv_on_pseries()) { 4457 if (kvmhv_is_nestedv1()) 4458 trap = kvmhv_vcpu_entry_p9_nested(vcpu, time_limit, lpcr, tb); 4459 else 4460 trap = kvmhv_vcpu_entry_nestedv2(vcpu, time_limit, lpcr, tb); 4461 4462 /* H_CEDE has to be handled now, not later */ 4463 if (trap == BOOK3S_INTERRUPT_SYSCALL && !nested && 4464 kvmppc_get_gpr(vcpu, 3) == H_CEDE) { 4465 kvmppc_cede(vcpu); 4466 kvmppc_set_gpr(vcpu, 3, 0); 4467 trap = 0; 4468 } 4469 4470 } else if (nested) { 4471 __this_cpu_write(cpu_in_guest, kvm); 4472 trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr, tb); 4473 __this_cpu_write(cpu_in_guest, NULL); 4474 4475 } else { 4476 kvmppc_xive_push_vcpu(vcpu); 4477 4478 __this_cpu_write(cpu_in_guest, kvm); 4479 trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr, tb); 4480 __this_cpu_write(cpu_in_guest, NULL); 4481 4482 if (trap == BOOK3S_INTERRUPT_SYSCALL && 4483 !(__kvmppc_get_msr_hv(vcpu) & MSR_PR)) { 4484 unsigned long req = kvmppc_get_gpr(vcpu, 3); 4485 4486 /* 4487 * XIVE rearm and XICS hcalls must be handled 4488 * before xive context is pulled (is this 4489 * true?) 4490 */ 4491 if (req == H_CEDE) { 4492 /* H_CEDE has to be handled now */ 4493 kvmppc_cede(vcpu); 4494 if (!kvmppc_xive_rearm_escalation(vcpu)) { 4495 /* 4496 * Pending escalation so abort 4497 * the cede. 4498 */ 4499 vcpu->arch.ceded = 0; 4500 } 4501 kvmppc_set_gpr(vcpu, 3, 0); 4502 trap = 0; 4503 4504 } else if (req == H_ENTER_NESTED) { 4505 /* 4506 * L2 should not run with the L1 4507 * context so rearm and pull it. 4508 */ 4509 if (!kvmppc_xive_rearm_escalation(vcpu)) { 4510 /* 4511 * Pending escalation so abort 4512 * H_ENTER_NESTED. 4513 */ 4514 kvmppc_set_gpr(vcpu, 3, 0); 4515 trap = 0; 4516 } 4517 4518 } else if (hcall_is_xics(req)) { 4519 int ret; 4520 4521 ret = kvmppc_xive_xics_hcall(vcpu, req); 4522 if (ret != H_TOO_HARD) { 4523 kvmppc_set_gpr(vcpu, 3, ret); 4524 trap = 0; 4525 } 4526 } 4527 } 4528 kvmppc_xive_pull_vcpu(vcpu); 4529 4530 if (kvm_is_radix(kvm)) 4531 vcpu->arch.slb_max = 0; 4532 } 4533 4534 vcpu_vpa_increment_dispatch(vcpu); 4535 4536 return trap; 4537 } 4538 4539 /* 4540 * Wait for some other vcpu thread to execute us, and 4541 * wake us up when we need to handle something in the host. 4542 */ 4543 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc, 4544 struct kvm_vcpu *vcpu, int wait_state) 4545 { 4546 DEFINE_WAIT(wait); 4547 4548 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state); 4549 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) { 4550 spin_unlock(&vc->lock); 4551 schedule(); 4552 spin_lock(&vc->lock); 4553 } 4554 finish_wait(&vcpu->arch.cpu_run, &wait); 4555 } 4556 4557 static void grow_halt_poll_ns(struct kvmppc_vcore *vc) 4558 { 4559 if (!halt_poll_ns_grow) 4560 return; 4561 4562 vc->halt_poll_ns *= halt_poll_ns_grow; 4563 if (vc->halt_poll_ns < halt_poll_ns_grow_start) 4564 vc->halt_poll_ns = halt_poll_ns_grow_start; 4565 } 4566 4567 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc) 4568 { 4569 if (halt_poll_ns_shrink == 0) 4570 vc->halt_poll_ns = 0; 4571 else 4572 vc->halt_poll_ns /= halt_poll_ns_shrink; 4573 } 4574 4575 #ifdef CONFIG_KVM_XICS 4576 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu) 4577 { 4578 if (!xics_on_xive()) 4579 return false; 4580 return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr < 4581 vcpu->arch.xive_saved_state.cppr; 4582 } 4583 #else 4584 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu) 4585 { 4586 return false; 4587 } 4588 #endif /* CONFIG_KVM_XICS */ 4589 4590 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu) 4591 { 4592 if (vcpu->arch.pending_exceptions || vcpu->arch.prodded || 4593 kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu)) 4594 return true; 4595 4596 return false; 4597 } 4598 4599 static bool kvmppc_vcpu_check_block(struct kvm_vcpu *vcpu) 4600 { 4601 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu)) 4602 return true; 4603 return false; 4604 } 4605 4606 /* 4607 * Check to see if any of the runnable vcpus on the vcore have pending 4608 * exceptions or are no longer ceded 4609 */ 4610 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc) 4611 { 4612 struct kvm_vcpu *vcpu; 4613 int i; 4614 4615 for_each_runnable_thread(i, vcpu, vc) { 4616 if (kvmppc_vcpu_check_block(vcpu)) 4617 return 1; 4618 } 4619 4620 return 0; 4621 } 4622 4623 /* 4624 * All the vcpus in this vcore are idle, so wait for a decrementer 4625 * or external interrupt to one of the vcpus. vc->lock is held. 4626 */ 4627 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc) 4628 { 4629 ktime_t cur, start_poll, start_wait; 4630 int do_sleep = 1; 4631 u64 block_ns; 4632 4633 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)); 4634 4635 /* Poll for pending exceptions and ceded state */ 4636 cur = start_poll = ktime_get(); 4637 if (vc->halt_poll_ns) { 4638 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns); 4639 ++vc->runner->stat.generic.halt_attempted_poll; 4640 4641 vc->vcore_state = VCORE_POLLING; 4642 spin_unlock(&vc->lock); 4643 4644 do { 4645 if (kvmppc_vcore_check_block(vc)) { 4646 do_sleep = 0; 4647 break; 4648 } 4649 cur = ktime_get(); 4650 } while (kvm_vcpu_can_poll(cur, stop)); 4651 4652 spin_lock(&vc->lock); 4653 vc->vcore_state = VCORE_INACTIVE; 4654 4655 if (!do_sleep) { 4656 ++vc->runner->stat.generic.halt_successful_poll; 4657 goto out; 4658 } 4659 } 4660 4661 prepare_to_rcuwait(&vc->wait); 4662 set_current_state(TASK_INTERRUPTIBLE); 4663 if (kvmppc_vcore_check_block(vc)) { 4664 finish_rcuwait(&vc->wait); 4665 do_sleep = 0; 4666 /* If we polled, count this as a successful poll */ 4667 if (vc->halt_poll_ns) 4668 ++vc->runner->stat.generic.halt_successful_poll; 4669 goto out; 4670 } 4671 4672 start_wait = ktime_get(); 4673 4674 vc->vcore_state = VCORE_SLEEPING; 4675 trace_kvmppc_vcore_blocked(vc->runner, 0); 4676 spin_unlock(&vc->lock); 4677 schedule(); 4678 finish_rcuwait(&vc->wait); 4679 spin_lock(&vc->lock); 4680 vc->vcore_state = VCORE_INACTIVE; 4681 trace_kvmppc_vcore_blocked(vc->runner, 1); 4682 ++vc->runner->stat.halt_successful_wait; 4683 4684 cur = ktime_get(); 4685 4686 out: 4687 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll); 4688 4689 /* Attribute wait time */ 4690 if (do_sleep) { 4691 vc->runner->stat.generic.halt_wait_ns += 4692 ktime_to_ns(cur) - ktime_to_ns(start_wait); 4693 KVM_STATS_LOG_HIST_UPDATE( 4694 vc->runner->stat.generic.halt_wait_hist, 4695 ktime_to_ns(cur) - ktime_to_ns(start_wait)); 4696 /* Attribute failed poll time */ 4697 if (vc->halt_poll_ns) { 4698 vc->runner->stat.generic.halt_poll_fail_ns += 4699 ktime_to_ns(start_wait) - 4700 ktime_to_ns(start_poll); 4701 KVM_STATS_LOG_HIST_UPDATE( 4702 vc->runner->stat.generic.halt_poll_fail_hist, 4703 ktime_to_ns(start_wait) - 4704 ktime_to_ns(start_poll)); 4705 } 4706 } else { 4707 /* Attribute successful poll time */ 4708 if (vc->halt_poll_ns) { 4709 vc->runner->stat.generic.halt_poll_success_ns += 4710 ktime_to_ns(cur) - 4711 ktime_to_ns(start_poll); 4712 KVM_STATS_LOG_HIST_UPDATE( 4713 vc->runner->stat.generic.halt_poll_success_hist, 4714 ktime_to_ns(cur) - ktime_to_ns(start_poll)); 4715 } 4716 } 4717 4718 /* Adjust poll time */ 4719 if (halt_poll_ns) { 4720 if (block_ns <= vc->halt_poll_ns) 4721 ; 4722 /* We slept and blocked for longer than the max halt time */ 4723 else if (vc->halt_poll_ns && block_ns > halt_poll_ns) 4724 shrink_halt_poll_ns(vc); 4725 /* We slept and our poll time is too small */ 4726 else if (vc->halt_poll_ns < halt_poll_ns && 4727 block_ns < halt_poll_ns) 4728 grow_halt_poll_ns(vc); 4729 if (vc->halt_poll_ns > halt_poll_ns) 4730 vc->halt_poll_ns = halt_poll_ns; 4731 } else 4732 vc->halt_poll_ns = 0; 4733 4734 trace_kvmppc_vcore_wakeup(do_sleep, block_ns); 4735 } 4736 4737 /* 4738 * This never fails for a radix guest, as none of the operations it does 4739 * for a radix guest can fail or have a way to report failure. 4740 */ 4741 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu) 4742 { 4743 int r = 0; 4744 struct kvm *kvm = vcpu->kvm; 4745 4746 mutex_lock(&kvm->arch.mmu_setup_lock); 4747 if (!kvm->arch.mmu_ready) { 4748 if (!kvm_is_radix(kvm)) 4749 r = kvmppc_hv_setup_htab_rma(vcpu); 4750 if (!r) { 4751 if (cpu_has_feature(CPU_FTR_ARCH_300)) 4752 kvmppc_setup_partition_table(kvm); 4753 kvm->arch.mmu_ready = 1; 4754 } 4755 } 4756 mutex_unlock(&kvm->arch.mmu_setup_lock); 4757 return r; 4758 } 4759 4760 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu) 4761 { 4762 struct kvm_run *run = vcpu->run; 4763 int n_ceded, i, r; 4764 struct kvmppc_vcore *vc; 4765 struct kvm_vcpu *v; 4766 4767 trace_kvmppc_run_vcpu_enter(vcpu); 4768 4769 run->exit_reason = 0; 4770 vcpu->arch.ret = RESUME_GUEST; 4771 vcpu->arch.trap = 0; 4772 kvmppc_update_vpas(vcpu); 4773 4774 /* 4775 * Synchronize with other threads in this virtual core 4776 */ 4777 vc = vcpu->arch.vcore; 4778 spin_lock(&vc->lock); 4779 vcpu->arch.ceded = 0; 4780 vcpu->arch.run_task = current; 4781 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb()); 4782 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE; 4783 vcpu->arch.busy_preempt = TB_NIL; 4784 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu); 4785 ++vc->n_runnable; 4786 4787 /* 4788 * This happens the first time this is called for a vcpu. 4789 * If the vcore is already running, we may be able to start 4790 * this thread straight away and have it join in. 4791 */ 4792 if (!signal_pending(current)) { 4793 if ((vc->vcore_state == VCORE_PIGGYBACK || 4794 vc->vcore_state == VCORE_RUNNING) && 4795 !VCORE_IS_EXITING(vc)) { 4796 kvmppc_update_vpa_dispatch(vcpu, vc); 4797 kvmppc_start_thread(vcpu, vc); 4798 trace_kvm_guest_enter(vcpu); 4799 } else if (vc->vcore_state == VCORE_SLEEPING) { 4800 rcuwait_wake_up(&vc->wait); 4801 } 4802 4803 } 4804 4805 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE && 4806 !signal_pending(current)) { 4807 /* See if the MMU is ready to go */ 4808 if (!vcpu->kvm->arch.mmu_ready) { 4809 spin_unlock(&vc->lock); 4810 r = kvmhv_setup_mmu(vcpu); 4811 spin_lock(&vc->lock); 4812 if (r) { 4813 run->exit_reason = KVM_EXIT_FAIL_ENTRY; 4814 run->fail_entry. 4815 hardware_entry_failure_reason = 0; 4816 vcpu->arch.ret = r; 4817 break; 4818 } 4819 } 4820 4821 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL) 4822 kvmppc_vcore_end_preempt(vc); 4823 4824 if (vc->vcore_state != VCORE_INACTIVE) { 4825 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE); 4826 continue; 4827 } 4828 for_each_runnable_thread(i, v, vc) { 4829 kvmppc_core_prepare_to_enter(v); 4830 if (signal_pending(v->arch.run_task)) { 4831 kvmppc_remove_runnable(vc, v, mftb()); 4832 v->stat.signal_exits++; 4833 v->run->exit_reason = KVM_EXIT_INTR; 4834 v->arch.ret = -EINTR; 4835 wake_up(&v->arch.cpu_run); 4836 } 4837 } 4838 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE) 4839 break; 4840 n_ceded = 0; 4841 for_each_runnable_thread(i, v, vc) { 4842 if (!kvmppc_vcpu_woken(v)) 4843 n_ceded += v->arch.ceded; 4844 else 4845 v->arch.ceded = 0; 4846 } 4847 vc->runner = vcpu; 4848 if (n_ceded == vc->n_runnable) { 4849 kvmppc_vcore_blocked(vc); 4850 } else if (__xfer_to_guest_mode_work_pending()) { 4851 kvmppc_vcore_preempt(vc); 4852 /* 4853 * Let something else run. The raw helper is used as 4854 * signal exits are accounted by this path already; 4855 * it may schedule(), so drop the vcore lock. 4856 */ 4857 spin_unlock(&vc->lock); 4858 xfer_to_guest_mode_handle_work(); 4859 spin_lock(&vc->lock); 4860 if (vc->vcore_state == VCORE_PREEMPT) 4861 kvmppc_vcore_end_preempt(vc); 4862 } else { 4863 kvmppc_run_core(vc); 4864 } 4865 vc->runner = NULL; 4866 } 4867 4868 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE && 4869 (vc->vcore_state == VCORE_RUNNING || 4870 vc->vcore_state == VCORE_EXITING || 4871 vc->vcore_state == VCORE_PIGGYBACK)) 4872 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE); 4873 4874 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL) 4875 kvmppc_vcore_end_preempt(vc); 4876 4877 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) { 4878 kvmppc_remove_runnable(vc, vcpu, mftb()); 4879 vcpu->stat.signal_exits++; 4880 run->exit_reason = KVM_EXIT_INTR; 4881 vcpu->arch.ret = -EINTR; 4882 } 4883 4884 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) { 4885 /* Wake up some vcpu to run the core */ 4886 i = -1; 4887 v = next_runnable_thread(vc, &i); 4888 wake_up(&v->arch.cpu_run); 4889 } 4890 4891 trace_kvmppc_run_vcpu_exit(vcpu); 4892 spin_unlock(&vc->lock); 4893 return vcpu->arch.ret; 4894 } 4895 4896 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit, 4897 unsigned long lpcr) 4898 { 4899 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu); 4900 struct kvm_run *run = vcpu->run; 4901 int trap, r, pcpu; 4902 int srcu_idx; 4903 struct kvmppc_vcore *vc; 4904 struct kvm *kvm = vcpu->kvm; 4905 struct kvm_nested_guest *nested = vcpu->arch.nested; 4906 unsigned long flags; 4907 u64 tb; 4908 4909 trace_kvmppc_run_vcpu_enter(vcpu); 4910 4911 run->exit_reason = 0; 4912 vcpu->arch.ret = RESUME_GUEST; 4913 vcpu->arch.trap = 0; 4914 4915 vc = vcpu->arch.vcore; 4916 vcpu->arch.ceded = 0; 4917 vcpu->arch.run_task = current; 4918 vcpu->arch.last_inst = KVM_INST_FETCH_FAILED; 4919 4920 /* See if the MMU is ready to go */ 4921 if (unlikely(!kvm->arch.mmu_ready)) { 4922 r = kvmhv_setup_mmu(vcpu); 4923 if (r) { 4924 run->exit_reason = KVM_EXIT_FAIL_ENTRY; 4925 run->fail_entry.hardware_entry_failure_reason = 0; 4926 vcpu->arch.ret = r; 4927 goto done; 4928 } 4929 } 4930 4931 r = kvm_xfer_to_guest_mode_handle_work(vcpu); 4932 if (r) { 4933 /* -EINTR: signal pending, exit to userspace (KVM_EXIT_INTR) */ 4934 vcpu->arch.ret = r; 4935 goto done; 4936 } 4937 4938 kvmppc_update_vpas(vcpu); 4939 4940 preempt_disable(); 4941 pcpu = smp_processor_id(); 4942 if (kvm_is_radix(kvm)) 4943 kvmppc_prepare_radix_vcpu(vcpu, pcpu); 4944 4945 /* flags save not required, but irq_pmu has no disable/enable API */ 4946 powerpc_local_irq_pmu_save(flags); 4947 4948 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE; 4949 4950 xfer_to_guest_mode_prepare(); 4951 4952 /* 4953 * IRQs are disabled here, so on pending work bail to the outer loop, 4954 * which handles it via kvm_xfer_to_guest_mode_handle_work() above. 4955 */ 4956 if (xfer_to_guest_mode_work_pending() || !kvm->arch.mmu_ready) 4957 goto out; 4958 4959 vcpu->cpu = pcpu; 4960 vcpu->arch.thread_cpu = pcpu; 4961 vc->pcpu = pcpu; 4962 local_paca->kvm_hstate.kvm_vcpu = vcpu; 4963 local_paca->kvm_hstate.ptid = 0; 4964 local_paca->kvm_hstate.fake_suspend = 0; 4965 4966 /* 4967 * Orders set cpu/thread_cpu vs testing for pending interrupts and 4968 * doorbells below. The other side is when these fields are set vs 4969 * kvmppc_fast_vcpu_kick_hv reading the cpu/thread_cpu fields to 4970 * kick a vCPU to notice the pending interrupt. 4971 */ 4972 smp_mb(); 4973 4974 if (!nested) { 4975 kvmppc_core_prepare_to_enter(vcpu); 4976 if (test_bit(BOOK3S_IRQPRIO_EXTERNAL, 4977 &vcpu->arch.pending_exceptions) || 4978 xive_interrupt_pending(vcpu)) { 4979 /* 4980 * For nested HV, don't synthesize but always pass MER, 4981 * the L0 will be able to optimise that more 4982 * effectively than manipulating registers directly. 4983 */ 4984 if (!kvmhv_on_pseries() && (__kvmppc_get_msr_hv(vcpu) & MSR_EE)) 4985 kvmppc_inject_interrupt_hv(vcpu, 4986 BOOK3S_INTERRUPT_EXTERNAL, 0); 4987 else 4988 lpcr |= LPCR_MER; 4989 } else { 4990 /* 4991 * L1's copy of L2's LPCR (vcpu->arch.vcore->lpcr) can get its MER bit 4992 * unexpectedly set - for e.g. during NMI handling when all register 4993 * states are synchronized from L0 to L1. L1 needs to inform L0 about 4994 * MER=1 only when there are pending external interrupts. 4995 * In the above if check, MER bit is set if there are pending 4996 * external interrupts. Hence, explicitly mask off MER bit 4997 * here as otherwise it may generate spurious interrupts in L2 KVM 4998 * causing an endless loop, which results in L2 guest getting hung. 4999 */ 5000 lpcr &= ~LPCR_MER; 5001 } 5002 } else if (vcpu->arch.pending_exceptions || 5003 xive_interrupt_pending(vcpu)) { 5004 vcpu->arch.ret = RESUME_HOST; 5005 goto out; 5006 } 5007 5008 if (vcpu->arch.timer_running) { 5009 hrtimer_try_to_cancel(&vcpu->arch.dec_timer); 5010 vcpu->arch.timer_running = 0; 5011 } 5012 5013 tb = mftb(); 5014 5015 kvmppc_update_vpa_dispatch_p9(vcpu, vc, tb + kvmppc_get_tb_offset(vcpu)); 5016 5017 trace_kvm_guest_enter(vcpu); 5018 5019 guest_timing_enter_irqoff(); 5020 5021 srcu_idx = srcu_read_lock(&kvm->srcu); 5022 5023 guest_state_enter_irqoff(); 5024 this_cpu_disable_ftrace(); 5025 5026 trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr, &tb); 5027 vcpu->arch.trap = trap; 5028 5029 this_cpu_enable_ftrace(); 5030 guest_state_exit_irqoff(); 5031 5032 srcu_read_unlock(&kvm->srcu, srcu_idx); 5033 5034 set_irq_happened(trap); 5035 5036 vcpu->cpu = -1; 5037 vcpu->arch.thread_cpu = -1; 5038 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST; 5039 5040 if (!vtime_accounting_enabled_this_cpu()) { 5041 powerpc_local_irq_pmu_restore(flags); 5042 /* 5043 * Service IRQs here before guest_timing_exit_irqoff() so any 5044 * ticks that occurred while running the guest are accounted to 5045 * the guest. If vtime accounting is enabled, accounting uses 5046 * TB rather than ticks, so it can be done without enabling 5047 * interrupts here, which has the problem that it accounts 5048 * interrupt processing overhead to the host. 5049 */ 5050 powerpc_local_irq_pmu_save(flags); 5051 } 5052 guest_timing_exit_irqoff(); 5053 5054 powerpc_local_irq_pmu_restore(flags); 5055 5056 preempt_enable(); 5057 5058 /* 5059 * cancel pending decrementer exception if DEC is now positive, or if 5060 * entering a nested guest in which case the decrementer is now owned 5061 * by L2 and the L1 decrementer is provided in hdec_expires 5062 */ 5063 if (kvmppc_core_pending_dec(vcpu) && 5064 ((tb < kvmppc_dec_expires_host_tb(vcpu)) || 5065 (trap == BOOK3S_INTERRUPT_SYSCALL && 5066 kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED))) 5067 kvmppc_core_dequeue_dec(vcpu); 5068 5069 trace_kvm_guest_exit(vcpu); 5070 r = RESUME_GUEST; 5071 if (trap) { 5072 if (!nested) 5073 r = kvmppc_handle_exit_hv(vcpu, current); 5074 else 5075 r = kvmppc_handle_nested_exit(vcpu); 5076 } 5077 vcpu->arch.ret = r; 5078 5079 if (is_kvmppc_resume_guest(r) && !kvmppc_vcpu_check_block(vcpu)) { 5080 kvmppc_set_timer(vcpu); 5081 5082 prepare_to_rcuwait(wait); 5083 for (;;) { 5084 set_current_state(TASK_INTERRUPTIBLE); 5085 if (signal_pending(current)) { 5086 vcpu->stat.signal_exits++; 5087 run->exit_reason = KVM_EXIT_INTR; 5088 vcpu->arch.ret = -EINTR; 5089 break; 5090 } 5091 5092 if (kvmppc_vcpu_check_block(vcpu)) 5093 break; 5094 5095 trace_kvmppc_vcore_blocked(vcpu, 0); 5096 schedule(); 5097 trace_kvmppc_vcore_blocked(vcpu, 1); 5098 } 5099 finish_rcuwait(wait); 5100 } 5101 vcpu->arch.ceded = 0; 5102 5103 done: 5104 trace_kvmppc_run_vcpu_exit(vcpu); 5105 5106 return vcpu->arch.ret; 5107 5108 out: 5109 vcpu->cpu = -1; 5110 vcpu->arch.thread_cpu = -1; 5111 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST; 5112 powerpc_local_irq_pmu_restore(flags); 5113 preempt_enable(); 5114 goto done; 5115 } 5116 5117 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu) 5118 { 5119 struct kvm_run *run = vcpu->run; 5120 int r; 5121 int srcu_idx; 5122 struct kvm *kvm; 5123 unsigned long msr; 5124 5125 start_timing(vcpu, &vcpu->arch.vcpu_entry); 5126 5127 if (!vcpu->arch.sane) { 5128 run->exit_reason = KVM_EXIT_INTERNAL_ERROR; 5129 return -EINVAL; 5130 } 5131 5132 /* No need to go into the guest when all we'll do is come back out */ 5133 if (signal_pending(current)) { 5134 run->exit_reason = KVM_EXIT_INTR; 5135 return -EINTR; 5136 } 5137 5138 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 5139 /* 5140 * Don't allow entry with a suspended transaction, because 5141 * the guest entry/exit code will lose it. 5142 */ 5143 if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs && 5144 (current->thread.regs->msr & MSR_TM)) { 5145 if (MSR_TM_ACTIVE(current->thread.regs->msr)) { 5146 run->exit_reason = KVM_EXIT_FAIL_ENTRY; 5147 run->fail_entry.hardware_entry_failure_reason = 0; 5148 return -EINVAL; 5149 } 5150 } 5151 #endif 5152 5153 /* 5154 * Force online to 1 for the sake of old userspace which doesn't 5155 * set it. 5156 */ 5157 if (!vcpu->arch.online) { 5158 atomic_inc(&vcpu->arch.vcore->online_count); 5159 vcpu->arch.online = 1; 5160 } 5161 5162 kvmppc_core_prepare_to_enter(vcpu); 5163 5164 kvm = vcpu->kvm; 5165 atomic_inc(&kvm->arch.vcpus_running); 5166 /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */ 5167 smp_mb(); 5168 5169 msr = 0; 5170 if (IS_ENABLED(CONFIG_PPC_FPU)) 5171 msr |= MSR_FP; 5172 if (cpu_has_feature(CPU_FTR_ALTIVEC)) 5173 msr |= MSR_VEC; 5174 if (cpu_has_feature(CPU_FTR_VSX)) 5175 msr |= MSR_VSX; 5176 if ((cpu_has_feature(CPU_FTR_TM) || 5177 cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) && 5178 (kvmppc_get_hfscr_hv(vcpu) & HFSCR_TM)) 5179 msr |= MSR_TM; 5180 msr = msr_check_and_set(msr); 5181 5182 kvmppc_save_user_regs(); 5183 5184 kvmppc_save_current_sprs(); 5185 5186 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 5187 vcpu->arch.waitp = &vcpu->arch.vcore->wait; 5188 vcpu->arch.pgdir = kvm->mm->pgd; 5189 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST; 5190 5191 do { 5192 accumulate_time(vcpu, &vcpu->arch.guest_entry); 5193 if (cpu_has_feature(CPU_FTR_ARCH_300)) 5194 r = kvmhv_run_single_vcpu(vcpu, ~(u64)0, 5195 vcpu->arch.vcore->lpcr); 5196 else 5197 r = kvmppc_run_vcpu(vcpu); 5198 5199 if (run->exit_reason == KVM_EXIT_PAPR_HCALL) { 5200 accumulate_time(vcpu, &vcpu->arch.hcall); 5201 5202 if (!kvmhv_is_nestedv2() && WARN_ON_ONCE(__kvmppc_get_msr_hv(vcpu) & MSR_PR)) { 5203 /* 5204 * These should have been caught reflected 5205 * into the guest by now. Final sanity check: 5206 * don't allow userspace to execute hcalls in 5207 * the hypervisor. 5208 */ 5209 r = RESUME_GUEST; 5210 continue; 5211 } 5212 trace_kvm_hcall_enter(vcpu); 5213 r = kvmppc_pseries_do_hcall(vcpu); 5214 trace_kvm_hcall_exit(vcpu, r); 5215 kvmppc_core_prepare_to_enter(vcpu); 5216 } else if (r == RESUME_PAGE_FAULT) { 5217 accumulate_time(vcpu, &vcpu->arch.pg_fault); 5218 srcu_idx = srcu_read_lock(&kvm->srcu); 5219 r = kvmppc_book3s_hv_page_fault(vcpu, 5220 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr); 5221 srcu_read_unlock(&kvm->srcu, srcu_idx); 5222 } else if (r == RESUME_PASSTHROUGH) { 5223 if (WARN_ON(xics_on_xive())) 5224 r = H_SUCCESS; 5225 else 5226 r = kvmppc_xics_rm_complete(vcpu, 0); 5227 } 5228 } while (is_kvmppc_resume_guest(r)); 5229 accumulate_time(vcpu, &vcpu->arch.vcpu_exit); 5230 5231 vcpu->arch.state = KVMPPC_VCPU_NOTREADY; 5232 atomic_dec(&kvm->arch.vcpus_running); 5233 5234 srr_regs_clobbered(); 5235 5236 end_timing(vcpu); 5237 5238 return r; 5239 } 5240 5241 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps, 5242 int shift, int sllp) 5243 { 5244 (*sps)->page_shift = shift; 5245 (*sps)->slb_enc = sllp; 5246 (*sps)->enc[0].page_shift = shift; 5247 (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift); 5248 /* 5249 * Add 16MB MPSS support (may get filtered out by userspace) 5250 */ 5251 if (shift != 24) { 5252 int penc = kvmppc_pgsize_lp_encoding(shift, 24); 5253 if (penc != -1) { 5254 (*sps)->enc[1].page_shift = 24; 5255 (*sps)->enc[1].pte_enc = penc; 5256 } 5257 } 5258 (*sps)++; 5259 } 5260 5261 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm, 5262 struct kvm_ppc_smmu_info *info) 5263 { 5264 struct kvm_ppc_one_seg_page_size *sps; 5265 5266 /* 5267 * POWER7, POWER8 and POWER9 all support 32 storage keys for data. 5268 * POWER7 doesn't support keys for instruction accesses, 5269 * POWER8 and POWER9 do. 5270 */ 5271 info->data_keys = 32; 5272 info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0; 5273 5274 /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */ 5275 info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS; 5276 info->slb_size = 32; 5277 5278 /* We only support these sizes for now, and no muti-size segments */ 5279 sps = &info->sps[0]; 5280 kvmppc_add_seg_page_size(&sps, 12, 0); 5281 kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01); 5282 kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L); 5283 5284 /* If running as a nested hypervisor, we don't support HPT guests */ 5285 if (kvmhv_on_pseries()) 5286 info->flags |= KVM_PPC_NO_HASH; 5287 5288 return 0; 5289 } 5290 5291 /* 5292 * Get (and clear) the dirty memory log for a memory slot. 5293 */ 5294 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm, 5295 struct kvm_dirty_log *log) 5296 { 5297 struct kvm_memslots *slots; 5298 struct kvm_memory_slot *memslot; 5299 int r; 5300 unsigned long n, i; 5301 unsigned long *buf, *p; 5302 struct kvm_vcpu *vcpu; 5303 5304 mutex_lock(&kvm->slots_lock); 5305 5306 r = -EINVAL; 5307 if (log->slot >= KVM_USER_MEM_SLOTS) 5308 goto out; 5309 5310 slots = kvm_memslots(kvm); 5311 memslot = id_to_memslot(slots, log->slot); 5312 r = -ENOENT; 5313 if (!memslot || !memslot->dirty_bitmap) 5314 goto out; 5315 5316 /* 5317 * Use second half of bitmap area because both HPT and radix 5318 * accumulate bits in the first half. 5319 */ 5320 n = kvm_dirty_bitmap_bytes(memslot); 5321 buf = memslot->dirty_bitmap + n / sizeof(long); 5322 memset(buf, 0, n); 5323 5324 if (kvm_is_radix(kvm)) 5325 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf); 5326 else 5327 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf); 5328 if (r) 5329 goto out; 5330 5331 /* 5332 * We accumulate dirty bits in the first half of the 5333 * memslot's dirty_bitmap area, for when pages are paged 5334 * out or modified by the host directly. Pick up these 5335 * bits and add them to the map. 5336 */ 5337 p = memslot->dirty_bitmap; 5338 for (i = 0; i < n / sizeof(long); ++i) 5339 buf[i] |= xchg(&p[i], 0); 5340 5341 /* Harvest dirty bits from VPA and DTL updates */ 5342 /* Note: we never modify the SLB shadow buffer areas */ 5343 kvm_for_each_vcpu(i, vcpu, kvm) { 5344 spin_lock(&vcpu->arch.vpa_update_lock); 5345 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf); 5346 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf); 5347 spin_unlock(&vcpu->arch.vpa_update_lock); 5348 } 5349 5350 r = -EFAULT; 5351 if (copy_to_user(log->dirty_bitmap, buf, n)) 5352 goto out; 5353 5354 r = 0; 5355 out: 5356 mutex_unlock(&kvm->slots_lock); 5357 return r; 5358 } 5359 5360 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot) 5361 { 5362 vfree(slot->arch.rmap); 5363 slot->arch.rmap = NULL; 5364 } 5365 5366 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm, 5367 const struct kvm_memory_slot *old, 5368 struct kvm_memory_slot *new, 5369 enum kvm_mr_change change) 5370 { 5371 if (change == KVM_MR_CREATE) { 5372 unsigned long size = array_size(new->npages, sizeof(*new->arch.rmap)); 5373 5374 if ((size >> PAGE_SHIFT) > totalram_pages()) 5375 return -ENOMEM; 5376 5377 new->arch.rmap = vzalloc(size); 5378 if (!new->arch.rmap) 5379 return -ENOMEM; 5380 } else if (change != KVM_MR_DELETE) { 5381 new->arch.rmap = old->arch.rmap; 5382 } 5383 5384 return 0; 5385 } 5386 5387 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm, 5388 struct kvm_memory_slot *old, 5389 const struct kvm_memory_slot *new, 5390 enum kvm_mr_change change) 5391 { 5392 /* 5393 * If we are creating or modifying a memslot, it might make 5394 * some address that was previously cached as emulated 5395 * MMIO be no longer emulated MMIO, so invalidate 5396 * all the caches of emulated MMIO translations. 5397 */ 5398 if (change != KVM_MR_DELETE) 5399 atomic64_inc(&kvm->arch.mmio_update); 5400 5401 /* 5402 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels 5403 * have already called kvm_arch_flush_shadow_memslot() to 5404 * flush shadow mappings. For KVM_MR_CREATE we have no 5405 * previous mappings. So the only case to handle is 5406 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit 5407 * has been changed. 5408 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES 5409 * to get rid of any THP PTEs in the partition-scoped page tables 5410 * so we can track dirtiness at the page level; we flush when 5411 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to 5412 * using THP PTEs. 5413 */ 5414 if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) && 5415 ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES)) 5416 kvmppc_radix_flush_memslot(kvm, old); 5417 /* 5418 * If UV hasn't yet called H_SVM_INIT_START, don't register memslots. 5419 */ 5420 if (!kvm->arch.secure_guest) 5421 return; 5422 5423 switch (change) { 5424 case KVM_MR_CREATE: 5425 /* 5426 * @TODO kvmppc_uvmem_memslot_create() can fail and 5427 * return error. Fix this. 5428 */ 5429 kvmppc_uvmem_memslot_create(kvm, new); 5430 break; 5431 case KVM_MR_DELETE: 5432 kvmppc_uvmem_memslot_delete(kvm, old); 5433 break; 5434 default: 5435 /* TODO: Handle KVM_MR_MOVE */ 5436 break; 5437 } 5438 } 5439 5440 /* 5441 * Update LPCR values in kvm->arch and in vcores. 5442 * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion 5443 * of kvm->arch.lpcr update). 5444 */ 5445 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask) 5446 { 5447 long int i; 5448 u32 cores_done = 0; 5449 5450 if ((kvm->arch.lpcr & mask) == lpcr) 5451 return; 5452 5453 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr; 5454 5455 for (i = 0; i < KVM_MAX_VCORES; ++i) { 5456 struct kvmppc_vcore *vc = kvm->arch.vcores[i]; 5457 if (!vc) 5458 continue; 5459 5460 spin_lock(&vc->lock); 5461 vc->lpcr = (vc->lpcr & ~mask) | lpcr; 5462 verify_lpcr(kvm, vc->lpcr); 5463 spin_unlock(&vc->lock); 5464 if (++cores_done >= kvm->arch.online_vcores) 5465 break; 5466 } 5467 5468 if (kvmhv_is_nestedv2()) { 5469 struct kvm_vcpu *vcpu; 5470 5471 kvm_for_each_vcpu(i, vcpu, kvm) { 5472 kvmhv_nestedv2_mark_dirty(vcpu, KVMPPC_GSID_LPCR); 5473 } 5474 } 5475 } 5476 5477 void kvmppc_setup_partition_table(struct kvm *kvm) 5478 { 5479 unsigned long dw0, dw1; 5480 5481 if (!kvm_is_radix(kvm)) { 5482 /* PS field - page size for VRMA */ 5483 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) | 5484 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1); 5485 /* HTABSIZE and HTABORG fields */ 5486 dw0 |= kvm->arch.sdr1; 5487 5488 /* Second dword as set by userspace */ 5489 dw1 = kvm->arch.process_table; 5490 } else { 5491 dw0 = PATB_HR | radix__get_tree_size() | 5492 __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE; 5493 dw1 = PATB_GR | kvm->arch.process_table; 5494 } 5495 kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1); 5496 } 5497 5498 /* 5499 * Set up HPT (hashed page table) and RMA (real-mode area). 5500 * Must be called with kvm->arch.mmu_setup_lock held. 5501 */ 5502 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu) 5503 { 5504 int err = 0; 5505 struct kvm *kvm = vcpu->kvm; 5506 unsigned long hva; 5507 struct kvm_memory_slot *memslot; 5508 struct vm_area_struct *vma; 5509 unsigned long lpcr = 0, senc; 5510 unsigned long psize, porder; 5511 int srcu_idx; 5512 5513 /* Allocate hashed page table (if not done already) and reset it */ 5514 if (!kvm->arch.hpt.virt) { 5515 int order = KVM_DEFAULT_HPT_ORDER; 5516 struct kvm_hpt_info info; 5517 5518 err = kvmppc_allocate_hpt(&info, order); 5519 /* If we get here, it means userspace didn't specify a 5520 * size explicitly. So, try successively smaller 5521 * sizes if the default failed. */ 5522 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER) 5523 err = kvmppc_allocate_hpt(&info, order); 5524 5525 if (err < 0) { 5526 pr_err("KVM: Couldn't alloc HPT\n"); 5527 goto out; 5528 } 5529 5530 kvmppc_set_hpt(kvm, &info); 5531 } 5532 5533 /* Look up the memslot for guest physical address 0 */ 5534 srcu_idx = srcu_read_lock(&kvm->srcu); 5535 memslot = gfn_to_memslot(kvm, 0); 5536 5537 /* We must have some memory at 0 by now */ 5538 err = -EINVAL; 5539 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) 5540 goto out_srcu; 5541 5542 /* Look up the VMA for the start of this memory slot */ 5543 hva = memslot->userspace_addr; 5544 mmap_read_lock(kvm->mm); 5545 vma = vma_lookup(kvm->mm, hva); 5546 if (!vma || (vma->vm_flags & VM_IO)) 5547 goto up_out; 5548 5549 psize = vma_kernel_pagesize(vma); 5550 5551 mmap_read_unlock(kvm->mm); 5552 5553 /* We can handle 4k, 64k or 16M pages in the VRMA */ 5554 if (psize >= 0x1000000) 5555 psize = 0x1000000; 5556 else if (psize >= 0x10000) 5557 psize = 0x10000; 5558 else 5559 psize = 0x1000; 5560 porder = __ilog2(psize); 5561 5562 senc = slb_pgsize_encoding(psize); 5563 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T | 5564 (VRMA_VSID << SLB_VSID_SHIFT_1T); 5565 /* Create HPTEs in the hash page table for the VRMA */ 5566 kvmppc_map_vrma(vcpu, memslot, porder); 5567 5568 /* Update VRMASD field in the LPCR */ 5569 if (!cpu_has_feature(CPU_FTR_ARCH_300)) { 5570 /* the -4 is to account for senc values starting at 0x10 */ 5571 lpcr = senc << (LPCR_VRMASD_SH - 4); 5572 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD); 5573 } 5574 5575 /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */ 5576 smp_wmb(); 5577 err = 0; 5578 out_srcu: 5579 srcu_read_unlock(&kvm->srcu, srcu_idx); 5580 out: 5581 return err; 5582 5583 up_out: 5584 mmap_read_unlock(kvm->mm); 5585 goto out_srcu; 5586 } 5587 5588 /* 5589 * Must be called with kvm->arch.mmu_setup_lock held and 5590 * mmu_ready = 0 and no vcpus running. 5591 */ 5592 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm) 5593 { 5594 unsigned long lpcr, lpcr_mask; 5595 5596 if (nesting_enabled(kvm)) 5597 kvmhv_release_all_nested(kvm); 5598 kvmppc_rmap_reset(kvm); 5599 kvm->arch.process_table = 0; 5600 /* Mutual exclusion with kvm_unmap_gfn_range etc. */ 5601 spin_lock(&kvm->mmu_lock); 5602 kvm->arch.radix = 0; 5603 spin_unlock(&kvm->mmu_lock); 5604 kvmppc_free_radix(kvm); 5605 5606 lpcr = LPCR_VPM1; 5607 lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR; 5608 if (cpu_has_feature(CPU_FTR_ARCH_31)) 5609 lpcr_mask |= LPCR_HAIL; 5610 kvmppc_update_lpcr(kvm, lpcr, lpcr_mask); 5611 5612 return 0; 5613 } 5614 5615 /* 5616 * Must be called with kvm->arch.mmu_setup_lock held and 5617 * mmu_ready = 0 and no vcpus running. 5618 */ 5619 int kvmppc_switch_mmu_to_radix(struct kvm *kvm) 5620 { 5621 unsigned long lpcr, lpcr_mask; 5622 int err; 5623 5624 err = kvmppc_init_vm_radix(kvm); 5625 if (err) 5626 return err; 5627 kvmppc_rmap_reset(kvm); 5628 /* Mutual exclusion with kvm_unmap_gfn_range etc. */ 5629 spin_lock(&kvm->mmu_lock); 5630 kvm->arch.radix = 1; 5631 spin_unlock(&kvm->mmu_lock); 5632 kvmppc_free_hpt(&kvm->arch.hpt); 5633 5634 lpcr = LPCR_UPRT | LPCR_GTSE | LPCR_HR; 5635 lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR; 5636 if (cpu_has_feature(CPU_FTR_ARCH_31)) { 5637 lpcr_mask |= LPCR_HAIL; 5638 if (cpu_has_feature(CPU_FTR_HVMODE) && 5639 (kvm->arch.host_lpcr & LPCR_HAIL)) 5640 lpcr |= LPCR_HAIL; 5641 } 5642 kvmppc_update_lpcr(kvm, lpcr, lpcr_mask); 5643 5644 return 0; 5645 } 5646 5647 #ifdef CONFIG_KVM_XICS 5648 /* 5649 * Allocate a per-core structure for managing state about which cores are 5650 * running in the host versus the guest and for exchanging data between 5651 * real mode KVM and CPU running in the host. 5652 * This is only done for the first VM. 5653 * The allocated structure stays even if all VMs have stopped. 5654 * It is only freed when the kvm-hv module is unloaded. 5655 * It's OK for this routine to fail, we just don't support host 5656 * core operations like redirecting H_IPI wakeups. 5657 */ 5658 void kvmppc_alloc_host_rm_ops(void) 5659 { 5660 struct kvmppc_host_rm_ops *ops; 5661 unsigned long l_ops; 5662 int cpu, core; 5663 int size; 5664 5665 if (cpu_has_feature(CPU_FTR_ARCH_300)) 5666 return; 5667 5668 /* Not the first time here ? */ 5669 if (kvmppc_host_rm_ops_hv != NULL) 5670 return; 5671 5672 ops = kzalloc_obj(struct kvmppc_host_rm_ops); 5673 if (!ops) 5674 return; 5675 5676 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core); 5677 ops->rm_core = kzalloc(size, GFP_KERNEL); 5678 5679 if (!ops->rm_core) { 5680 kfree(ops); 5681 return; 5682 } 5683 5684 cpus_read_lock(); 5685 5686 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) { 5687 if (!cpu_online(cpu)) 5688 continue; 5689 5690 core = cpu >> threads_shift; 5691 ops->rm_core[core].rm_state.in_host = 1; 5692 } 5693 5694 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv; 5695 5696 /* 5697 * Make the contents of the kvmppc_host_rm_ops structure visible 5698 * to other CPUs before we assign it to the global variable. 5699 * Do an atomic assignment (no locks used here), but if someone 5700 * beats us to it, just free our copy and return. 5701 */ 5702 smp_wmb(); 5703 l_ops = (unsigned long) ops; 5704 5705 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) { 5706 cpus_read_unlock(); 5707 kfree(ops->rm_core); 5708 kfree(ops); 5709 return; 5710 } 5711 5712 cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE, 5713 "ppc/kvm_book3s:prepare", 5714 kvmppc_set_host_core, 5715 kvmppc_clear_host_core); 5716 cpus_read_unlock(); 5717 } 5718 5719 void kvmppc_free_host_rm_ops(void) 5720 { 5721 if (kvmppc_host_rm_ops_hv) { 5722 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE); 5723 kfree(kvmppc_host_rm_ops_hv->rm_core); 5724 kfree(kvmppc_host_rm_ops_hv); 5725 kvmppc_host_rm_ops_hv = NULL; 5726 } 5727 } 5728 #endif 5729 5730 static int kvmppc_core_init_vm_hv(struct kvm *kvm) 5731 { 5732 unsigned long lpcr, lpid; 5733 int ret; 5734 5735 mutex_init(&kvm->arch.uvmem_lock); 5736 INIT_LIST_HEAD(&kvm->arch.uvmem_pfns); 5737 mutex_init(&kvm->arch.mmu_setup_lock); 5738 5739 /* Allocate the guest's logical partition ID */ 5740 5741 if (!kvmhv_is_nestedv2()) { 5742 lpid = kvmppc_alloc_lpid(); 5743 if ((long)lpid < 0) 5744 return -ENOMEM; 5745 kvm->arch.lpid = lpid; 5746 } 5747 5748 kvmppc_alloc_host_rm_ops(); 5749 5750 kvmhv_vm_nested_init(kvm); 5751 5752 if (kvmhv_is_nestedv2()) { 5753 long rc; 5754 unsigned long guest_id; 5755 5756 rc = plpar_guest_create(0, &guest_id); 5757 5758 if (rc != H_SUCCESS) 5759 pr_err("KVM: Create Guest hcall failed, rc=%ld\n", rc); 5760 5761 switch (rc) { 5762 case H_PARAMETER: 5763 case H_FUNCTION: 5764 case H_STATE: 5765 return -EINVAL; 5766 case H_NOT_ENOUGH_RESOURCES: 5767 case H_ABORTED: 5768 return -ENOMEM; 5769 case H_AUTHORITY: 5770 return -EPERM; 5771 case H_NOT_AVAILABLE: 5772 return -EBUSY; 5773 } 5774 kvm->arch.lpid = guest_id; 5775 } 5776 5777 5778 /* 5779 * Since we don't flush the TLB when tearing down a VM, 5780 * and this lpid might have previously been used, 5781 * make sure we flush on each core before running the new VM. 5782 * On POWER9, the tlbie in mmu_partition_table_set_entry() 5783 * does this flush for us. 5784 */ 5785 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 5786 cpumask_setall(&kvm->arch.need_tlb_flush); 5787 5788 /* Start out with the default set of hcalls enabled */ 5789 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls, 5790 sizeof(kvm->arch.enabled_hcalls)); 5791 5792 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 5793 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1); 5794 5795 /* Init LPCR for virtual RMA mode */ 5796 if (cpu_has_feature(CPU_FTR_HVMODE)) { 5797 kvm->arch.host_lpid = mfspr(SPRN_LPID); 5798 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR); 5799 lpcr &= LPCR_PECE | LPCR_LPES; 5800 } else { 5801 /* 5802 * The L2 LPES mode will be set by the L0 according to whether 5803 * or not it needs to take external interrupts in HV mode. 5804 */ 5805 lpcr = 0; 5806 } 5807 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE | 5808 LPCR_VPM0 | LPCR_VPM1; 5809 kvm->arch.vrma_slb_v = SLB_VSID_B_1T | 5810 (VRMA_VSID << SLB_VSID_SHIFT_1T); 5811 /* On POWER8 turn on online bit to enable PURR/SPURR */ 5812 if (cpu_has_feature(CPU_FTR_ARCH_207S)) 5813 lpcr |= LPCR_ONL; 5814 /* 5815 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed) 5816 * Set HVICE bit to enable hypervisor virtualization interrupts. 5817 * Set HEIC to prevent OS interrupts to go to hypervisor (should 5818 * be unnecessary but better safe than sorry in case we re-enable 5819 * EE in HV mode with this LPCR still set) 5820 */ 5821 if (cpu_has_feature(CPU_FTR_ARCH_300)) { 5822 lpcr &= ~LPCR_VPM0; 5823 lpcr |= LPCR_HVICE | LPCR_HEIC; 5824 5825 /* 5826 * If xive is enabled, we route 0x500 interrupts directly 5827 * to the guest. 5828 */ 5829 if (xics_on_xive()) 5830 lpcr |= LPCR_LPES; 5831 } 5832 5833 /* 5834 * If the host uses radix, the guest starts out as radix. 5835 */ 5836 if (radix_enabled()) { 5837 kvm->arch.radix = 1; 5838 kvm->arch.mmu_ready = 1; 5839 lpcr &= ~LPCR_VPM1; 5840 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR; 5841 if (cpu_has_feature(CPU_FTR_HVMODE) && 5842 cpu_has_feature(CPU_FTR_ARCH_31) && 5843 (kvm->arch.host_lpcr & LPCR_HAIL)) 5844 lpcr |= LPCR_HAIL; 5845 ret = kvmppc_init_vm_radix(kvm); 5846 if (ret) { 5847 if (kvmhv_is_nestedv2()) 5848 plpar_guest_delete(0, kvm->arch.lpid); 5849 else 5850 kvmppc_free_lpid(kvm->arch.lpid); 5851 return ret; 5852 } 5853 kvmppc_setup_partition_table(kvm); 5854 } 5855 5856 verify_lpcr(kvm, lpcr); 5857 kvm->arch.lpcr = lpcr; 5858 5859 /* Initialization for future HPT resizes */ 5860 kvm->arch.resize_hpt = NULL; 5861 5862 /* 5863 * Work out how many sets the TLB has, for the use of 5864 * the TLB invalidation loop in book3s_hv_rmhandlers.S. 5865 */ 5866 if (cpu_has_feature(CPU_FTR_ARCH_31)) { 5867 /* 5868 * P10 will flush all the congruence class with a single tlbiel 5869 */ 5870 kvm->arch.tlb_sets = 1; 5871 } else if (radix_enabled()) 5872 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */ 5873 else if (cpu_has_feature(CPU_FTR_ARCH_300)) 5874 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */ 5875 else if (cpu_has_feature(CPU_FTR_ARCH_207S)) 5876 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */ 5877 else 5878 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */ 5879 5880 /* 5881 * Track that we now have a HV mode VM active. This blocks secondary 5882 * CPU threads from coming online. 5883 */ 5884 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 5885 kvm_hv_vm_activated(); 5886 5887 /* 5888 * Initialize smt_mode depending on processor. 5889 * POWER8 and earlier have to use "strict" threading, where 5890 * all vCPUs in a vcore have to run on the same (sub)core, 5891 * whereas on POWER9 the threads can each run a different 5892 * guest. 5893 */ 5894 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 5895 kvm->arch.smt_mode = threads_per_subcore; 5896 else 5897 kvm->arch.smt_mode = 1; 5898 kvm->arch.emul_smt_mode = 1; 5899 5900 return 0; 5901 } 5902 5903 static int kvmppc_arch_create_vm_debugfs_hv(struct kvm *kvm) 5904 { 5905 kvmppc_mmu_debugfs_init(kvm); 5906 if (radix_enabled()) 5907 kvmhv_radix_debugfs_init(kvm); 5908 return 0; 5909 } 5910 5911 static void kvmppc_free_vcores(struct kvm *kvm) 5912 { 5913 long int i; 5914 5915 for (i = 0; i < KVM_MAX_VCORES; ++i) 5916 kfree(kvm->arch.vcores[i]); 5917 kvm->arch.online_vcores = 0; 5918 } 5919 5920 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm) 5921 { 5922 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 5923 kvm_hv_vm_deactivated(); 5924 5925 kvmppc_free_vcores(kvm); 5926 5927 5928 if (kvm_is_radix(kvm)) 5929 kvmppc_free_radix(kvm); 5930 else 5931 kvmppc_free_hpt(&kvm->arch.hpt); 5932 5933 /* Perform global invalidation and return lpid to the pool */ 5934 if (cpu_has_feature(CPU_FTR_ARCH_300)) { 5935 if (nesting_enabled(kvm)) 5936 kvmhv_release_all_nested(kvm); 5937 kvm->arch.process_table = 0; 5938 if (kvm->arch.secure_guest) 5939 uv_svm_terminate(kvm->arch.lpid); 5940 if (!kvmhv_is_nestedv2()) 5941 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0); 5942 } 5943 5944 if (kvmhv_is_nestedv2()) { 5945 kvmhv_flush_lpid(kvm->arch.lpid); 5946 plpar_guest_delete(0, kvm->arch.lpid); 5947 } else { 5948 kvmppc_free_lpid(kvm->arch.lpid); 5949 } 5950 5951 kvmppc_free_pimap(kvm); 5952 } 5953 5954 /* We don't need to emulate any privileged instructions or dcbz */ 5955 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu, 5956 unsigned int inst, int *advance) 5957 { 5958 return EMULATE_FAIL; 5959 } 5960 5961 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn, 5962 ulong spr_val) 5963 { 5964 return EMULATE_FAIL; 5965 } 5966 5967 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn, 5968 ulong *spr_val) 5969 { 5970 return EMULATE_FAIL; 5971 } 5972 5973 static int kvmppc_core_check_processor_compat_hv(void) 5974 { 5975 if (cpu_has_feature(CPU_FTR_HVMODE) && 5976 cpu_has_feature(CPU_FTR_ARCH_206)) 5977 return 0; 5978 5979 /* POWER9 in radix mode is capable of being a nested hypervisor. */ 5980 if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled()) 5981 return 0; 5982 5983 return -EIO; 5984 } 5985 5986 #ifdef CONFIG_KVM_XICS 5987 5988 void kvmppc_free_pimap(struct kvm *kvm) 5989 { 5990 kfree(kvm->arch.pimap); 5991 } 5992 5993 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void) 5994 { 5995 return kzalloc_obj(struct kvmppc_passthru_irqmap); 5996 } 5997 5998 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi) 5999 { 6000 struct irq_desc *desc; 6001 struct kvmppc_irq_map *irq_map; 6002 struct kvmppc_passthru_irqmap *pimap; 6003 struct irq_chip *chip; 6004 int i, rc = 0; 6005 struct irq_data *host_data; 6006 6007 if (!kvm_irq_bypass) 6008 return 1; 6009 6010 desc = irq_to_desc(host_irq); 6011 if (!desc) 6012 return -EIO; 6013 6014 mutex_lock(&kvm->lock); 6015 6016 pimap = kvm->arch.pimap; 6017 if (pimap == NULL) { 6018 /* First call, allocate structure to hold IRQ map */ 6019 pimap = kvmppc_alloc_pimap(); 6020 if (pimap == NULL) { 6021 mutex_unlock(&kvm->lock); 6022 return -ENOMEM; 6023 } 6024 kvm->arch.pimap = pimap; 6025 } 6026 6027 /* 6028 * For now, we only support interrupts for which the EOI operation 6029 * is an OPAL call followed by a write to XIRR, since that's 6030 * what our real-mode EOI code does, or a XIVE interrupt 6031 */ 6032 chip = irq_data_get_irq_chip(&desc->irq_data); 6033 if (!chip || !is_pnv_opal_msi(chip)) { 6034 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n", 6035 host_irq, guest_gsi); 6036 mutex_unlock(&kvm->lock); 6037 return -ENOENT; 6038 } 6039 6040 /* 6041 * See if we already have an entry for this guest IRQ number. 6042 * If it's mapped to a hardware IRQ number, that's an error, 6043 * otherwise re-use this entry. 6044 */ 6045 for (i = 0; i < pimap->n_mapped; i++) { 6046 if (guest_gsi == pimap->mapped[i].v_hwirq) { 6047 if (pimap->mapped[i].r_hwirq) { 6048 mutex_unlock(&kvm->lock); 6049 return -EINVAL; 6050 } 6051 break; 6052 } 6053 } 6054 6055 if (i == KVMPPC_PIRQ_MAPPED) { 6056 mutex_unlock(&kvm->lock); 6057 return -EAGAIN; /* table is full */ 6058 } 6059 6060 irq_map = &pimap->mapped[i]; 6061 6062 irq_map->v_hwirq = guest_gsi; 6063 irq_map->desc = desc; 6064 6065 /* 6066 * Order the above two stores before the next to serialize with 6067 * the KVM real mode handler. 6068 */ 6069 smp_wmb(); 6070 6071 /* 6072 * The 'host_irq' number is mapped in the PCI-MSI domain but 6073 * the underlying calls, which will EOI the interrupt in real 6074 * mode, need an HW IRQ number mapped in the XICS IRQ domain. 6075 */ 6076 host_data = irq_domain_get_irq_data(irq_get_default_domain(), host_irq); 6077 irq_map->r_hwirq = (unsigned int)irqd_to_hwirq(host_data); 6078 6079 if (i == pimap->n_mapped) 6080 pimap->n_mapped++; 6081 6082 if (xics_on_xive()) 6083 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, host_irq); 6084 else 6085 kvmppc_xics_set_mapped(kvm, guest_gsi, irq_map->r_hwirq); 6086 if (rc) 6087 irq_map->r_hwirq = 0; 6088 6089 mutex_unlock(&kvm->lock); 6090 6091 return 0; 6092 } 6093 6094 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi) 6095 { 6096 struct irq_desc *desc; 6097 struct kvmppc_passthru_irqmap *pimap; 6098 int i, rc = 0; 6099 6100 if (!kvm_irq_bypass) 6101 return 0; 6102 6103 desc = irq_to_desc(host_irq); 6104 if (!desc) 6105 return -EIO; 6106 6107 mutex_lock(&kvm->lock); 6108 if (!kvm->arch.pimap) 6109 goto unlock; 6110 6111 pimap = kvm->arch.pimap; 6112 6113 for (i = 0; i < pimap->n_mapped; i++) { 6114 if (guest_gsi == pimap->mapped[i].v_hwirq) 6115 break; 6116 } 6117 6118 if (i == pimap->n_mapped) { 6119 mutex_unlock(&kvm->lock); 6120 return -ENODEV; 6121 } 6122 6123 if (xics_on_xive()) 6124 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, host_irq); 6125 else 6126 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq); 6127 6128 /* invalidate the entry (what to do on error from the above ?) */ 6129 pimap->mapped[i].r_hwirq = 0; 6130 6131 /* 6132 * We don't free this structure even when the count goes to 6133 * zero. The structure is freed when we destroy the VM. 6134 */ 6135 unlock: 6136 mutex_unlock(&kvm->lock); 6137 return rc; 6138 } 6139 6140 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons, 6141 struct irq_bypass_producer *prod) 6142 { 6143 int ret = 0; 6144 struct kvm_kernel_irqfd *irqfd = 6145 container_of(cons, struct kvm_kernel_irqfd, consumer); 6146 6147 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi); 6148 if (ret) 6149 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n", 6150 prod->irq, irqfd->gsi, ret); 6151 else 6152 irqfd->producer = prod; 6153 6154 return ret; 6155 } 6156 6157 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons, 6158 struct irq_bypass_producer *prod) 6159 { 6160 int ret; 6161 struct kvm_kernel_irqfd *irqfd = 6162 container_of(cons, struct kvm_kernel_irqfd, consumer); 6163 6164 irqfd->producer = NULL; 6165 6166 /* 6167 * When producer of consumer is unregistered, we change back to 6168 * default external interrupt handling mode - KVM real mode 6169 * will switch back to host. 6170 */ 6171 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi); 6172 if (ret) 6173 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n", 6174 prod->irq, irqfd->gsi, ret); 6175 } 6176 #endif 6177 6178 static int kvm_arch_vm_ioctl_hv(struct file *filp, 6179 unsigned int ioctl, unsigned long arg) 6180 { 6181 struct kvm *kvm __maybe_unused = filp->private_data; 6182 void __user *argp = (void __user *)arg; 6183 int r; 6184 6185 switch (ioctl) { 6186 6187 case KVM_PPC_ALLOCATE_HTAB: { 6188 u32 htab_order; 6189 6190 /* If we're a nested hypervisor, we currently only support radix */ 6191 if (kvmhv_on_pseries()) { 6192 r = -EOPNOTSUPP; 6193 break; 6194 } 6195 6196 r = -EFAULT; 6197 if (get_user(htab_order, (u32 __user *)argp)) 6198 break; 6199 r = kvmppc_alloc_reset_hpt(kvm, htab_order); 6200 if (r) 6201 break; 6202 r = 0; 6203 break; 6204 } 6205 6206 case KVM_PPC_GET_HTAB_FD: { 6207 struct kvm_get_htab_fd ghf; 6208 6209 r = -EFAULT; 6210 if (copy_from_user(&ghf, argp, sizeof(ghf))) 6211 break; 6212 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf); 6213 break; 6214 } 6215 6216 case KVM_PPC_RESIZE_HPT_PREPARE: { 6217 struct kvm_ppc_resize_hpt rhpt; 6218 6219 r = -EFAULT; 6220 if (copy_from_user(&rhpt, argp, sizeof(rhpt))) 6221 break; 6222 6223 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt); 6224 break; 6225 } 6226 6227 case KVM_PPC_RESIZE_HPT_COMMIT: { 6228 struct kvm_ppc_resize_hpt rhpt; 6229 6230 r = -EFAULT; 6231 if (copy_from_user(&rhpt, argp, sizeof(rhpt))) 6232 break; 6233 6234 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt); 6235 break; 6236 } 6237 6238 default: 6239 r = -ENOTTY; 6240 } 6241 6242 return r; 6243 } 6244 6245 /* 6246 * List of hcall numbers to enable by default. 6247 * For compatibility with old userspace, we enable by default 6248 * all hcalls that were implemented before the hcall-enabling 6249 * facility was added. Note this list should not include H_RTAS. 6250 */ 6251 static unsigned int default_hcall_list[] = { 6252 H_REMOVE, 6253 H_ENTER, 6254 H_READ, 6255 H_PROTECT, 6256 H_BULK_REMOVE, 6257 #ifdef CONFIG_SPAPR_TCE_IOMMU 6258 H_GET_TCE, 6259 H_PUT_TCE, 6260 #endif 6261 H_SET_DABR, 6262 H_SET_XDABR, 6263 H_CEDE, 6264 H_PROD, 6265 H_CONFER, 6266 H_REGISTER_VPA, 6267 #ifdef CONFIG_KVM_XICS 6268 H_EOI, 6269 H_CPPR, 6270 H_IPI, 6271 H_IPOLL, 6272 H_XIRR, 6273 H_XIRR_X, 6274 #endif 6275 0 6276 }; 6277 6278 static void init_default_hcalls(void) 6279 { 6280 int i; 6281 unsigned int hcall; 6282 6283 for (i = 0; default_hcall_list[i]; ++i) { 6284 hcall = default_hcall_list[i]; 6285 WARN_ON(!kvmppc_hcall_impl_hv(hcall)); 6286 __set_bit(hcall / 4, default_enabled_hcalls); 6287 } 6288 } 6289 6290 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg) 6291 { 6292 unsigned long lpcr; 6293 int radix; 6294 int err; 6295 6296 /* If not on a POWER9, reject it */ 6297 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 6298 return -ENODEV; 6299 6300 /* If any unknown flags set, reject it */ 6301 if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE)) 6302 return -EINVAL; 6303 6304 /* GR (guest radix) bit in process_table field must match */ 6305 radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX); 6306 if (!!(cfg->process_table & PATB_GR) != radix) 6307 return -EINVAL; 6308 6309 /* Process table size field must be reasonable, i.e. <= 24 */ 6310 if ((cfg->process_table & PRTS_MASK) > 24) 6311 return -EINVAL; 6312 6313 /* We can change a guest to/from radix now, if the host is radix */ 6314 if (radix && !radix_enabled()) 6315 return -EINVAL; 6316 6317 /* If we're a nested hypervisor, we currently only support radix */ 6318 if (kvmhv_on_pseries() && !radix) 6319 return -EINVAL; 6320 6321 mutex_lock(&kvm->arch.mmu_setup_lock); 6322 if (radix != kvm_is_radix(kvm)) { 6323 if (kvm->arch.mmu_ready) { 6324 kvm->arch.mmu_ready = 0; 6325 /* order mmu_ready vs. vcpus_running */ 6326 smp_mb(); 6327 if (atomic_read(&kvm->arch.vcpus_running)) { 6328 kvm->arch.mmu_ready = 1; 6329 err = -EBUSY; 6330 goto out_unlock; 6331 } 6332 } 6333 if (radix) 6334 err = kvmppc_switch_mmu_to_radix(kvm); 6335 else 6336 err = kvmppc_switch_mmu_to_hpt(kvm); 6337 if (err) 6338 goto out_unlock; 6339 } 6340 6341 kvm->arch.process_table = cfg->process_table; 6342 kvmppc_setup_partition_table(kvm); 6343 6344 lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0; 6345 kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE); 6346 err = 0; 6347 6348 out_unlock: 6349 mutex_unlock(&kvm->arch.mmu_setup_lock); 6350 return err; 6351 } 6352 6353 static int kvmhv_enable_nested(struct kvm *kvm) 6354 { 6355 if (!nested) 6356 return -EPERM; 6357 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 6358 return -ENODEV; 6359 if (!radix_enabled()) 6360 return -ENODEV; 6361 if (kvmhv_is_nestedv2()) 6362 return -ENODEV; 6363 6364 /* kvm == NULL means the caller is testing if the capability exists */ 6365 if (kvm) 6366 kvm->arch.nested_enable = true; 6367 return 0; 6368 } 6369 6370 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr, 6371 int size) 6372 { 6373 int rc = -EINVAL; 6374 6375 if (kvmhv_vcpu_is_radix(vcpu)) { 6376 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size); 6377 6378 if (rc > 0) 6379 rc = -EINVAL; 6380 } 6381 6382 /* For now quadrants are the only way to access nested guest memory */ 6383 if (rc && vcpu->arch.nested) 6384 rc = -EAGAIN; 6385 6386 return rc; 6387 } 6388 6389 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr, 6390 int size) 6391 { 6392 int rc = -EINVAL; 6393 6394 if (kvmhv_vcpu_is_radix(vcpu)) { 6395 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size); 6396 6397 if (rc > 0) 6398 rc = -EINVAL; 6399 } 6400 6401 /* For now quadrants are the only way to access nested guest memory */ 6402 if (rc && vcpu->arch.nested) 6403 rc = -EAGAIN; 6404 6405 return rc; 6406 } 6407 6408 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa) 6409 { 6410 unpin_vpa(kvm, vpa); 6411 vpa->gpa = 0; 6412 vpa->pinned_addr = NULL; 6413 vpa->dirty = false; 6414 vpa->update_pending = 0; 6415 } 6416 6417 /* 6418 * Enable a guest to become a secure VM, or test whether 6419 * that could be enabled. 6420 * Called when the KVM_CAP_PPC_SECURE_GUEST capability is 6421 * tested (kvm == NULL) or enabled (kvm != NULL). 6422 */ 6423 static int kvmhv_enable_svm(struct kvm *kvm) 6424 { 6425 if (!kvmppc_uvmem_available()) 6426 return -EINVAL; 6427 if (kvm) 6428 kvm->arch.svm_enabled = 1; 6429 return 0; 6430 } 6431 6432 /* 6433 * IOCTL handler to turn off secure mode of guest 6434 * 6435 * - Release all device pages 6436 * - Issue ucall to terminate the guest on the UV side 6437 * - Unpin the VPA pages. 6438 * - Reinit the partition scoped page tables 6439 */ 6440 static int kvmhv_svm_off(struct kvm *kvm) 6441 { 6442 struct kvm_vcpu *vcpu; 6443 int mmu_was_ready; 6444 int srcu_idx; 6445 int ret = 0; 6446 unsigned long i; 6447 6448 if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START)) 6449 return ret; 6450 6451 mutex_lock(&kvm->arch.mmu_setup_lock); 6452 mmu_was_ready = kvm->arch.mmu_ready; 6453 if (kvm->arch.mmu_ready) { 6454 kvm->arch.mmu_ready = 0; 6455 /* order mmu_ready vs. vcpus_running */ 6456 smp_mb(); 6457 if (atomic_read(&kvm->arch.vcpus_running)) { 6458 kvm->arch.mmu_ready = 1; 6459 ret = -EBUSY; 6460 goto out; 6461 } 6462 } 6463 6464 srcu_idx = srcu_read_lock(&kvm->srcu); 6465 for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) { 6466 struct kvm_memory_slot *memslot; 6467 struct kvm_memslots *slots = __kvm_memslots(kvm, i); 6468 int bkt; 6469 6470 if (!slots) 6471 continue; 6472 6473 kvm_for_each_memslot(memslot, bkt, slots) { 6474 kvmppc_uvmem_drop_pages(memslot, kvm, true); 6475 uv_unregister_mem_slot(kvm->arch.lpid, memslot->id); 6476 } 6477 } 6478 srcu_read_unlock(&kvm->srcu, srcu_idx); 6479 6480 ret = uv_svm_terminate(kvm->arch.lpid); 6481 if (ret != U_SUCCESS) { 6482 ret = -EINVAL; 6483 goto out; 6484 } 6485 6486 /* 6487 * When secure guest is reset, all the guest pages are sent 6488 * to UV via UV_PAGE_IN before the non-boot vcpus get a 6489 * chance to run and unpin their VPA pages. Unpinning of all 6490 * VPA pages is done here explicitly so that VPA pages 6491 * can be migrated to the secure side. 6492 * 6493 * This is required to for the secure SMP guest to reboot 6494 * correctly. 6495 */ 6496 kvm_for_each_vcpu(i, vcpu, kvm) { 6497 spin_lock(&vcpu->arch.vpa_update_lock); 6498 unpin_vpa_reset(kvm, &vcpu->arch.dtl); 6499 unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow); 6500 unpin_vpa_reset(kvm, &vcpu->arch.vpa); 6501 spin_unlock(&vcpu->arch.vpa_update_lock); 6502 } 6503 6504 kvmppc_setup_partition_table(kvm); 6505 kvm->arch.secure_guest = 0; 6506 kvm->arch.mmu_ready = mmu_was_ready; 6507 out: 6508 mutex_unlock(&kvm->arch.mmu_setup_lock); 6509 return ret; 6510 } 6511 6512 static int kvmhv_enable_dawr1(struct kvm *kvm) 6513 { 6514 if (!cpu_has_feature(CPU_FTR_DAWR1)) 6515 return -ENODEV; 6516 6517 /* kvm == NULL means the caller is testing if the capability exists */ 6518 if (kvm) 6519 kvm->arch.dawr1_enabled = true; 6520 return 0; 6521 } 6522 6523 static bool kvmppc_hash_v3_possible(void) 6524 { 6525 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 6526 return false; 6527 6528 if (!cpu_has_feature(CPU_FTR_HVMODE)) 6529 return false; 6530 6531 /* 6532 * POWER9 chips before version 2.02 can't have some threads in 6533 * HPT mode and some in radix mode on the same core. 6534 */ 6535 if (radix_enabled()) { 6536 unsigned int pvr = mfspr(SPRN_PVR); 6537 if ((pvr >> 16) == PVR_POWER9 && 6538 (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) || 6539 ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101))) 6540 return false; 6541 } 6542 6543 return true; 6544 } 6545 6546 static int kvmppc_map_compat_capabilities(u32 cpu_version, 6547 unsigned long *capabilities) 6548 { 6549 switch (cpu_version) { 6550 case PVR_ARCH_31_P11: 6551 *capabilities |= KVM_PPC_COMPAT_CAP_POWER11; 6552 fallthrough; 6553 case PVR_ARCH_31: 6554 *capabilities |= KVM_PPC_COMPAT_CAP_POWER10; 6555 fallthrough; 6556 case PVR_ARCH_300: 6557 *capabilities |= KVM_PPC_COMPAT_CAP_POWER9; 6558 break; 6559 default: 6560 return -EINVAL; 6561 } 6562 6563 return 0; 6564 } 6565 6566 static int kvmppc_get_compat_caps(struct kvm_ppc_compat_caps *host_caps) 6567 { 6568 struct device_node *np; 6569 unsigned long capabilities = 0; 6570 long rc = -EINVAL; 6571 u32 cpu_version = 0; 6572 6573 if (kvmhv_on_pseries()) { 6574 if (kvmhv_is_nestedv2()) { 6575 WARN_ON_ONCE(!nested_capabilities); 6576 capabilities = nested_capabilities; 6577 rc = 0; 6578 } else { 6579 for_each_node_by_type(np, "cpu") { 6580 if (!of_property_read_u32(np, "cpu-version", 6581 &cpu_version)) { 6582 of_node_put(np); 6583 break; 6584 } 6585 } 6586 if (!cpu_version) 6587 return -EINVAL; 6588 rc = kvmppc_map_compat_capabilities(cpu_version, 6589 &capabilities); 6590 } 6591 } 6592 6593 if (rc < 0) 6594 return rc; 6595 6596 host_caps->compat_capabilities = capabilities & KVM_PPC_COMPAT_BITMASK; 6597 6598 return rc; 6599 } 6600 6601 static struct kvmppc_ops kvm_ops_hv = { 6602 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv, 6603 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv, 6604 .get_one_reg = kvmppc_get_one_reg_hv, 6605 .set_one_reg = kvmppc_set_one_reg_hv, 6606 .vcpu_load = kvmppc_core_vcpu_load_hv, 6607 .vcpu_put = kvmppc_core_vcpu_put_hv, 6608 .inject_interrupt = kvmppc_inject_interrupt_hv, 6609 .set_msr = kvmppc_set_msr_hv, 6610 .vcpu_run = kvmppc_vcpu_run_hv, 6611 .vcpu_create = kvmppc_core_vcpu_create_hv, 6612 .vcpu_free = kvmppc_core_vcpu_free_hv, 6613 .check_requests = kvmppc_core_check_requests_hv, 6614 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv, 6615 .flush_memslot = kvmppc_core_flush_memslot_hv, 6616 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv, 6617 .commit_memory_region = kvmppc_core_commit_memory_region_hv, 6618 .unmap_gfn_range = kvm_unmap_gfn_range_hv, 6619 .age_gfn = kvm_age_gfn_hv, 6620 .test_age_gfn = kvm_test_age_gfn_hv, 6621 .free_memslot = kvmppc_core_free_memslot_hv, 6622 .init_vm = kvmppc_core_init_vm_hv, 6623 .destroy_vm = kvmppc_core_destroy_vm_hv, 6624 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv, 6625 .emulate_op = kvmppc_core_emulate_op_hv, 6626 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv, 6627 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv, 6628 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv, 6629 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv, 6630 .hcall_implemented = kvmppc_hcall_impl_hv, 6631 .configure_mmu = kvmhv_configure_mmu, 6632 .get_rmmu_info = kvmhv_get_rmmu_info, 6633 .set_smt_mode = kvmhv_set_smt_mode, 6634 .enable_nested = kvmhv_enable_nested, 6635 .load_from_eaddr = kvmhv_load_from_eaddr, 6636 .store_to_eaddr = kvmhv_store_to_eaddr, 6637 .enable_svm = kvmhv_enable_svm, 6638 .svm_off = kvmhv_svm_off, 6639 .enable_dawr1 = kvmhv_enable_dawr1, 6640 .hash_v3_possible = kvmppc_hash_v3_possible, 6641 .create_vcpu_debugfs = kvmppc_arch_create_vcpu_debugfs_hv, 6642 .create_vm_debugfs = kvmppc_arch_create_vm_debugfs_hv, 6643 .get_compat_caps = kvmppc_get_compat_caps, 6644 }; 6645 6646 static int kvm_init_subcore_bitmap(void) 6647 { 6648 int i, j; 6649 int nr_cores = cpu_nr_cores(); 6650 struct sibling_subcore_state *sibling_subcore_state; 6651 6652 for (i = 0; i < nr_cores; i++) { 6653 int first_cpu = i * threads_per_core; 6654 int node = cpu_to_node(first_cpu); 6655 6656 /* Ignore if it is already allocated. */ 6657 if (paca_ptrs[first_cpu]->sibling_subcore_state) 6658 continue; 6659 6660 sibling_subcore_state = 6661 kzalloc_node(sizeof(struct sibling_subcore_state), 6662 GFP_KERNEL, node); 6663 if (!sibling_subcore_state) 6664 return -ENOMEM; 6665 6666 6667 for (j = 0; j < threads_per_core; j++) { 6668 int cpu = first_cpu + j; 6669 6670 paca_ptrs[cpu]->sibling_subcore_state = 6671 sibling_subcore_state; 6672 } 6673 } 6674 return 0; 6675 } 6676 6677 static int kvmppc_radix_possible(void) 6678 { 6679 return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled(); 6680 } 6681 6682 static int kvmppc_book3s_init_hv(void) 6683 { 6684 int r; 6685 6686 if (!tlbie_capable) { 6687 pr_err("KVM-HV: Host does not support TLBIE\n"); 6688 return -ENODEV; 6689 } 6690 6691 /* 6692 * FIXME!! Do we need to check on all cpus ? 6693 */ 6694 r = kvmppc_core_check_processor_compat_hv(); 6695 if (r < 0) 6696 return -ENODEV; 6697 6698 r = kvmhv_nested_init(); 6699 if (r) 6700 return r; 6701 6702 if (!cpu_has_feature(CPU_FTR_ARCH_300)) { 6703 r = kvm_init_subcore_bitmap(); 6704 if (r) 6705 goto err; 6706 } 6707 6708 /* 6709 * We need a way of accessing the XICS interrupt controller, 6710 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or 6711 * indirectly, via OPAL. 6712 */ 6713 #ifdef CONFIG_SMP 6714 if (!xics_on_xive() && !kvmhv_on_pseries() && 6715 !local_paca->kvm_hstate.xics_phys) { 6716 struct device_node *np; 6717 6718 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc"); 6719 if (!np) { 6720 pr_err("KVM-HV: Cannot determine method for accessing XICS\n"); 6721 r = -ENODEV; 6722 goto err; 6723 } 6724 /* presence of intc confirmed - node can be dropped again */ 6725 of_node_put(np); 6726 } 6727 #endif 6728 6729 init_default_hcalls(); 6730 6731 init_vcore_lists(); 6732 6733 r = kvmppc_mmu_hv_init(); 6734 if (r) 6735 goto err; 6736 6737 if (kvmppc_radix_possible()) { 6738 r = kvmppc_radix_init(); 6739 if (r) 6740 goto err; 6741 } 6742 6743 r = kvmppc_uvmem_init(); 6744 if (r < 0) { 6745 pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r); 6746 return r; 6747 } 6748 6749 #if defined(CONFIG_KVM_XICS) 6750 /* 6751 * IRQ bypass is supported only for interrupts whose EOI operations are 6752 * handled via OPAL calls. Therefore, register IRQ bypass handlers 6753 * exclusively for PowerNV KVM when booted with 'xive=off', indicating 6754 * the use of the emulated XICS interrupt controller. 6755 */ 6756 if (!kvmhv_on_pseries()) { 6757 pr_info("KVM-HV: Enabling IRQ bypass\n"); 6758 kvm_ops_hv.irq_bypass_add_producer = 6759 kvmppc_irq_bypass_add_producer_hv; 6760 kvm_ops_hv.irq_bypass_del_producer = 6761 kvmppc_irq_bypass_del_producer_hv; 6762 } 6763 #endif 6764 6765 kvm_ops_hv.owner = THIS_MODULE; 6766 kvmppc_hv_ops = &kvm_ops_hv; 6767 6768 return 0; 6769 6770 err: 6771 kvmhv_nested_exit(); 6772 kvmppc_radix_exit(); 6773 6774 return r; 6775 } 6776 6777 static void kvmppc_book3s_exit_hv(void) 6778 { 6779 kvmppc_uvmem_free(); 6780 kvmppc_free_host_rm_ops(); 6781 if (kvmppc_radix_possible()) 6782 kvmppc_radix_exit(); 6783 kvmppc_hv_ops = NULL; 6784 kvmhv_nested_exit(); 6785 } 6786 6787 module_init(kvmppc_book3s_init_hv); 6788 module_exit(kvmppc_book3s_exit_hv); 6789 MODULE_DESCRIPTION("KVM on Book3S (POWER8 and later) in hypervisor mode"); 6790 MODULE_LICENSE("GPL"); 6791 MODULE_ALIAS_MISCDEV(KVM_MINOR); 6792 MODULE_ALIAS("devname:kvm"); 6793