1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com> 4 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved. 5 * 6 * Authors: 7 * Paul Mackerras <paulus@au1.ibm.com> 8 * Alexander Graf <agraf@suse.de> 9 * Kevin Wolf <mail@kevin-wolf.de> 10 * 11 * Description: KVM functions specific to running on Book 3S 12 * processors in hypervisor mode (specifically POWER7 and later). 13 * 14 * This file is derived from arch/powerpc/kvm/book3s.c, 15 * by Alexander Graf <agraf@suse.de>. 16 */ 17 18 #include <linux/kvm_host.h> 19 #include <linux/kernel.h> 20 #include <linux/err.h> 21 #include <linux/slab.h> 22 #include <linux/preempt.h> 23 #include <linux/sched/signal.h> 24 #include <linux/sched/stat.h> 25 #include <linux/delay.h> 26 #include <linux/export.h> 27 #include <linux/fs.h> 28 #include <linux/anon_inodes.h> 29 #include <linux/cpu.h> 30 #include <linux/cpumask.h> 31 #include <linux/spinlock.h> 32 #include <linux/page-flags.h> 33 #include <linux/srcu.h> 34 #include <linux/miscdevice.h> 35 #include <linux/debugfs.h> 36 #include <linux/gfp.h> 37 #include <linux/vmalloc.h> 38 #include <linux/highmem.h> 39 #include <linux/hugetlb.h> 40 #include <linux/kvm_irqfd.h> 41 #include <linux/irqbypass.h> 42 #include <linux/module.h> 43 #include <linux/compiler.h> 44 #include <linux/of.h> 45 #include <linux/irqdomain.h> 46 #include <linux/smp.h> 47 48 #include <asm/ftrace.h> 49 #include <asm/reg.h> 50 #include <asm/ppc-opcode.h> 51 #include <asm/asm-prototypes.h> 52 #include <asm/archrandom.h> 53 #include <asm/debug.h> 54 #include <asm/disassemble.h> 55 #include <asm/cputable.h> 56 #include <asm/cacheflush.h> 57 #include <linux/uaccess.h> 58 #include <asm/interrupt.h> 59 #include <asm/io.h> 60 #include <asm/kvm_ppc.h> 61 #include <asm/kvm_book3s.h> 62 #include <asm/mmu_context.h> 63 #include <asm/lppaca.h> 64 #include <asm/pmc.h> 65 #include <asm/processor.h> 66 #include <asm/cputhreads.h> 67 #include <asm/page.h> 68 #include <asm/hvcall.h> 69 #include <asm/switch_to.h> 70 #include <asm/smp.h> 71 #include <asm/dbell.h> 72 #include <asm/hmi.h> 73 #include <asm/pnv-pci.h> 74 #include <asm/mmu.h> 75 #include <asm/opal.h> 76 #include <asm/xics.h> 77 #include <asm/xive.h> 78 #include <asm/hw_breakpoint.h> 79 #include <asm/kvm_book3s_uvmem.h> 80 #include <asm/ultravisor.h> 81 #include <asm/dtl.h> 82 #include <asm/plpar_wrappers.h> 83 84 #include <trace/events/ipi.h> 85 86 #include "book3s.h" 87 #include "book3s_hv.h" 88 89 #define CREATE_TRACE_POINTS 90 #include "trace_hv.h" 91 92 /* #define EXIT_DEBUG */ 93 /* #define EXIT_DEBUG_SIMPLE */ 94 /* #define EXIT_DEBUG_INT */ 95 96 /* Used to indicate that a guest page fault needs to be handled */ 97 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1) 98 /* Used to indicate that a guest passthrough interrupt needs to be handled */ 99 #define RESUME_PASSTHROUGH (RESUME_GUEST | RESUME_FLAG_ARCH2) 100 101 /* Used as a "null" value for timebase values */ 102 #define TB_NIL (~(u64)0) 103 104 static DECLARE_BITMAP(default_enabled_hcalls, MAX_HCALL_OPCODE/4 + 1); 105 106 static int dynamic_mt_modes = 6; 107 module_param(dynamic_mt_modes, int, 0644); 108 MODULE_PARM_DESC(dynamic_mt_modes, "Set of allowed dynamic micro-threading modes: 0 (= none), 2, 4, or 6 (= 2 or 4)"); 109 static int target_smt_mode; 110 module_param(target_smt_mode, int, 0644); 111 MODULE_PARM_DESC(target_smt_mode, "Target threads per core (0 = max)"); 112 113 static bool one_vm_per_core; 114 module_param(one_vm_per_core, bool, S_IRUGO | S_IWUSR); 115 MODULE_PARM_DESC(one_vm_per_core, "Only run vCPUs from the same VM on a core (requires POWER8 or older)"); 116 117 #ifdef CONFIG_KVM_XICS 118 static const struct kernel_param_ops module_param_ops = { 119 .set = param_set_int, 120 .get = param_get_int, 121 }; 122 123 module_param_cb(kvm_irq_bypass, &module_param_ops, &kvm_irq_bypass, 0644); 124 MODULE_PARM_DESC(kvm_irq_bypass, "Bypass passthrough interrupt optimization"); 125 126 module_param_cb(h_ipi_redirect, &module_param_ops, &h_ipi_redirect, 0644); 127 MODULE_PARM_DESC(h_ipi_redirect, "Redirect H_IPI wakeup to a free host core"); 128 #endif 129 130 /* If set, guests are allowed to create and control nested guests */ 131 static bool nested = true; 132 module_param(nested, bool, S_IRUGO | S_IWUSR); 133 MODULE_PARM_DESC(nested, "Enable nested virtualization (only on POWER9)"); 134 135 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu); 136 137 /* 138 * RWMR values for POWER8. These control the rate at which PURR 139 * and SPURR count and should be set according to the number of 140 * online threads in the vcore being run. 141 */ 142 #define RWMR_RPA_P8_1THREAD 0x164520C62609AECAUL 143 #define RWMR_RPA_P8_2THREAD 0x7FFF2908450D8DA9UL 144 #define RWMR_RPA_P8_3THREAD 0x164520C62609AECAUL 145 #define RWMR_RPA_P8_4THREAD 0x199A421245058DA9UL 146 #define RWMR_RPA_P8_5THREAD 0x164520C62609AECAUL 147 #define RWMR_RPA_P8_6THREAD 0x164520C62609AECAUL 148 #define RWMR_RPA_P8_7THREAD 0x164520C62609AECAUL 149 #define RWMR_RPA_P8_8THREAD 0x164520C62609AECAUL 150 151 static unsigned long p8_rwmr_values[MAX_SMT_THREADS + 1] = { 152 RWMR_RPA_P8_1THREAD, 153 RWMR_RPA_P8_1THREAD, 154 RWMR_RPA_P8_2THREAD, 155 RWMR_RPA_P8_3THREAD, 156 RWMR_RPA_P8_4THREAD, 157 RWMR_RPA_P8_5THREAD, 158 RWMR_RPA_P8_6THREAD, 159 RWMR_RPA_P8_7THREAD, 160 RWMR_RPA_P8_8THREAD, 161 }; 162 163 static inline struct kvm_vcpu *next_runnable_thread(struct kvmppc_vcore *vc, 164 int *ip) 165 { 166 int i = *ip; 167 struct kvm_vcpu *vcpu; 168 169 while (++i < MAX_SMT_THREADS) { 170 vcpu = READ_ONCE(vc->runnable_threads[i]); 171 if (vcpu) { 172 *ip = i; 173 return vcpu; 174 } 175 } 176 return NULL; 177 } 178 179 /* Used to traverse the list of runnable threads for a given vcore */ 180 #define for_each_runnable_thread(i, vcpu, vc) \ 181 for (i = -1; (vcpu = next_runnable_thread(vc, &i)); ) 182 183 static bool kvmppc_ipi_thread(int cpu) 184 { 185 unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER); 186 187 /* If we're a nested hypervisor, fall back to ordinary IPIs for now */ 188 if (kvmhv_on_pseries()) 189 return false; 190 191 /* On POWER9 we can use msgsnd to IPI any cpu */ 192 if (cpu_has_feature(CPU_FTR_ARCH_300)) { 193 msg |= get_hard_smp_processor_id(cpu); 194 smp_mb(); 195 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg)); 196 return true; 197 } 198 199 /* On POWER8 for IPIs to threads in the same core, use msgsnd */ 200 if (cpu_has_feature(CPU_FTR_ARCH_207S)) { 201 preempt_disable(); 202 if (cpu_first_thread_sibling(cpu) == 203 cpu_first_thread_sibling(smp_processor_id())) { 204 msg |= cpu_thread_in_core(cpu); 205 smp_mb(); 206 __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg)); 207 preempt_enable(); 208 return true; 209 } 210 preempt_enable(); 211 } 212 213 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP) 214 if (cpu >= 0 && cpu < nr_cpu_ids) { 215 if (paca_ptrs[cpu]->kvm_hstate.xics_phys) { 216 xics_wake_cpu(cpu); 217 return true; 218 } 219 opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY); 220 return true; 221 } 222 #endif 223 224 return false; 225 } 226 227 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu) 228 { 229 int cpu; 230 struct rcuwait *waitp; 231 232 /* 233 * rcuwait_wake_up contains smp_mb() which orders prior stores that 234 * create pending work vs below loads of cpu fields. The other side 235 * is the barrier in vcpu run that orders setting the cpu fields vs 236 * testing for pending work. 237 */ 238 239 waitp = kvm_arch_vcpu_get_wait(vcpu); 240 if (rcuwait_wake_up(waitp)) 241 ++vcpu->stat.generic.halt_wakeup; 242 243 cpu = READ_ONCE(vcpu->arch.thread_cpu); 244 if (cpu >= 0 && kvmppc_ipi_thread(cpu)) 245 return; 246 247 /* CPU points to the first thread of the core */ 248 cpu = vcpu->cpu; 249 if (cpu >= 0 && cpu < nr_cpu_ids && cpu_online(cpu)) 250 smp_send_reschedule(cpu); 251 } 252 253 /* 254 * We use the vcpu_load/put functions to measure stolen time. 255 * 256 * Stolen time is counted as time when either the vcpu is able to 257 * run as part of a virtual core, but the task running the vcore 258 * is preempted or sleeping, or when the vcpu needs something done 259 * in the kernel by the task running the vcpu, but that task is 260 * preempted or sleeping. Those two things have to be counted 261 * separately, since one of the vcpu tasks will take on the job 262 * of running the core, and the other vcpu tasks in the vcore will 263 * sleep waiting for it to do that, but that sleep shouldn't count 264 * as stolen time. 265 * 266 * Hence we accumulate stolen time when the vcpu can run as part of 267 * a vcore using vc->stolen_tb, and the stolen time when the vcpu 268 * needs its task to do other things in the kernel (for example, 269 * service a page fault) in busy_stolen. We don't accumulate 270 * stolen time for a vcore when it is inactive, or for a vcpu 271 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of 272 * a misnomer; it means that the vcpu task is not executing in 273 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in 274 * the kernel. We don't have any way of dividing up that time 275 * between time that the vcpu is genuinely stopped, time that 276 * the task is actively working on behalf of the vcpu, and time 277 * that the task is preempted, so we don't count any of it as 278 * stolen. 279 * 280 * Updates to busy_stolen are protected by arch.tbacct_lock; 281 * updates to vc->stolen_tb are protected by the vcore->stoltb_lock 282 * lock. The stolen times are measured in units of timebase ticks. 283 * (Note that the != TB_NIL checks below are purely defensive; 284 * they should never fail.) 285 * 286 * The POWER9 path is simpler, one vcpu per virtual core so the 287 * former case does not exist. If a vcpu is preempted when it is 288 * BUSY_IN_HOST and not ceded or otherwise blocked, then accumulate 289 * the stolen cycles in busy_stolen. RUNNING is not a preemptible 290 * state in the P9 path. 291 */ 292 293 static void kvmppc_core_start_stolen(struct kvmppc_vcore *vc, u64 tb) 294 { 295 unsigned long flags; 296 297 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)); 298 299 spin_lock_irqsave(&vc->stoltb_lock, flags); 300 vc->preempt_tb = tb; 301 spin_unlock_irqrestore(&vc->stoltb_lock, flags); 302 } 303 304 static void kvmppc_core_end_stolen(struct kvmppc_vcore *vc, u64 tb) 305 { 306 unsigned long flags; 307 308 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)); 309 310 spin_lock_irqsave(&vc->stoltb_lock, flags); 311 if (vc->preempt_tb != TB_NIL) { 312 vc->stolen_tb += tb - vc->preempt_tb; 313 vc->preempt_tb = TB_NIL; 314 } 315 spin_unlock_irqrestore(&vc->stoltb_lock, flags); 316 } 317 318 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu) 319 { 320 struct kvmppc_vcore *vc = vcpu->arch.vcore; 321 unsigned long flags; 322 u64 now; 323 324 if (cpu_has_feature(CPU_FTR_ARCH_300)) { 325 if (vcpu->arch.busy_preempt != TB_NIL) { 326 WARN_ON_ONCE(vcpu->arch.state != KVMPPC_VCPU_BUSY_IN_HOST); 327 vc->stolen_tb += mftb() - vcpu->arch.busy_preempt; 328 vcpu->arch.busy_preempt = TB_NIL; 329 } 330 return; 331 } 332 333 now = mftb(); 334 335 /* 336 * We can test vc->runner without taking the vcore lock, 337 * because only this task ever sets vc->runner to this 338 * vcpu, and once it is set to this vcpu, only this task 339 * ever sets it to NULL. 340 */ 341 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING) 342 kvmppc_core_end_stolen(vc, now); 343 344 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags); 345 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST && 346 vcpu->arch.busy_preempt != TB_NIL) { 347 vcpu->arch.busy_stolen += now - vcpu->arch.busy_preempt; 348 vcpu->arch.busy_preempt = TB_NIL; 349 } 350 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags); 351 } 352 353 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu) 354 { 355 struct kvmppc_vcore *vc = vcpu->arch.vcore; 356 unsigned long flags; 357 u64 now; 358 359 if (cpu_has_feature(CPU_FTR_ARCH_300)) { 360 /* 361 * In the P9 path, RUNNABLE is not preemptible 362 * (nor takes host interrupts) 363 */ 364 WARN_ON_ONCE(vcpu->arch.state == KVMPPC_VCPU_RUNNABLE); 365 /* 366 * Account stolen time when preempted while the vcpu task is 367 * running in the kernel (but not in qemu, which is INACTIVE). 368 */ 369 if (task_is_running(current) && 370 vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST) 371 vcpu->arch.busy_preempt = mftb(); 372 return; 373 } 374 375 now = mftb(); 376 377 if (vc->runner == vcpu && vc->vcore_state >= VCORE_SLEEPING) 378 kvmppc_core_start_stolen(vc, now); 379 380 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags); 381 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST) 382 vcpu->arch.busy_preempt = now; 383 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags); 384 } 385 386 static void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr) 387 { 388 vcpu->arch.pvr = pvr; 389 } 390 391 /* Dummy value used in computing PCR value below */ 392 #define PCR_ARCH_31 (PCR_ARCH_300 << 1) 393 394 static int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat) 395 { 396 unsigned long host_pcr_bit = 0, guest_pcr_bit = 0; 397 struct kvmppc_vcore *vc = vcpu->arch.vcore; 398 399 /* We can (emulate) our own architecture version and anything older */ 400 if (cpu_has_feature(CPU_FTR_ARCH_31)) 401 host_pcr_bit = PCR_ARCH_31; 402 else if (cpu_has_feature(CPU_FTR_ARCH_300)) 403 host_pcr_bit = PCR_ARCH_300; 404 else if (cpu_has_feature(CPU_FTR_ARCH_207S)) 405 host_pcr_bit = PCR_ARCH_207; 406 else if (cpu_has_feature(CPU_FTR_ARCH_206)) 407 host_pcr_bit = PCR_ARCH_206; 408 else 409 host_pcr_bit = PCR_ARCH_205; 410 411 /* Determine lowest PCR bit needed to run guest in given PVR level */ 412 guest_pcr_bit = host_pcr_bit; 413 if (arch_compat) { 414 switch (arch_compat) { 415 case PVR_ARCH_205: 416 guest_pcr_bit = PCR_ARCH_205; 417 break; 418 case PVR_ARCH_206: 419 case PVR_ARCH_206p: 420 guest_pcr_bit = PCR_ARCH_206; 421 break; 422 case PVR_ARCH_207: 423 guest_pcr_bit = PCR_ARCH_207; 424 break; 425 case PVR_ARCH_300: 426 guest_pcr_bit = PCR_ARCH_300; 427 break; 428 case PVR_ARCH_31: 429 guest_pcr_bit = PCR_ARCH_31; 430 break; 431 default: 432 return -EINVAL; 433 } 434 } 435 436 /* Check requested PCR bits don't exceed our capabilities */ 437 if (guest_pcr_bit > host_pcr_bit) 438 return -EINVAL; 439 440 spin_lock(&vc->lock); 441 vc->arch_compat = arch_compat; 442 /* 443 * Set all PCR bits for which guest_pcr_bit <= bit < host_pcr_bit 444 * Also set all reserved PCR bits 445 */ 446 vc->pcr = (host_pcr_bit - guest_pcr_bit) | PCR_MASK; 447 spin_unlock(&vc->lock); 448 449 return 0; 450 } 451 452 static void kvmppc_dump_regs(struct kvm_vcpu *vcpu) 453 { 454 int r; 455 456 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id); 457 pr_err("pc = %.16lx msr = %.16llx trap = %x\n", 458 vcpu->arch.regs.nip, vcpu->arch.shregs.msr, vcpu->arch.trap); 459 for (r = 0; r < 16; ++r) 460 pr_err("r%2d = %.16lx r%d = %.16lx\n", 461 r, kvmppc_get_gpr(vcpu, r), 462 r+16, kvmppc_get_gpr(vcpu, r+16)); 463 pr_err("ctr = %.16lx lr = %.16lx\n", 464 vcpu->arch.regs.ctr, vcpu->arch.regs.link); 465 pr_err("srr0 = %.16llx srr1 = %.16llx\n", 466 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1); 467 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n", 468 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1); 469 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n", 470 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3); 471 pr_err("cr = %.8lx xer = %.16lx dsisr = %.8x\n", 472 vcpu->arch.regs.ccr, vcpu->arch.regs.xer, vcpu->arch.shregs.dsisr); 473 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar); 474 pr_err("fault dar = %.16lx dsisr = %.8x\n", 475 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr); 476 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max); 477 for (r = 0; r < vcpu->arch.slb_max; ++r) 478 pr_err(" ESID = %.16llx VSID = %.16llx\n", 479 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv); 480 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n", 481 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1, 482 vcpu->arch.last_inst); 483 } 484 485 static struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id) 486 { 487 return kvm_get_vcpu_by_id(kvm, id); 488 } 489 490 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa) 491 { 492 vpa->__old_status |= LPPACA_OLD_SHARED_PROC; 493 vpa->yield_count = cpu_to_be32(1); 494 } 495 496 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v, 497 unsigned long addr, unsigned long len) 498 { 499 /* check address is cacheline aligned */ 500 if (addr & (L1_CACHE_BYTES - 1)) 501 return -EINVAL; 502 spin_lock(&vcpu->arch.vpa_update_lock); 503 if (v->next_gpa != addr || v->len != len) { 504 v->next_gpa = addr; 505 v->len = addr ? len : 0; 506 v->update_pending = 1; 507 } 508 spin_unlock(&vcpu->arch.vpa_update_lock); 509 return 0; 510 } 511 512 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */ 513 struct reg_vpa { 514 u32 dummy; 515 union { 516 __be16 hword; 517 __be32 word; 518 } length; 519 }; 520 521 static int vpa_is_registered(struct kvmppc_vpa *vpap) 522 { 523 if (vpap->update_pending) 524 return vpap->next_gpa != 0; 525 return vpap->pinned_addr != NULL; 526 } 527 528 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu, 529 unsigned long flags, 530 unsigned long vcpuid, unsigned long vpa) 531 { 532 struct kvm *kvm = vcpu->kvm; 533 unsigned long len, nb; 534 void *va; 535 struct kvm_vcpu *tvcpu; 536 int err; 537 int subfunc; 538 struct kvmppc_vpa *vpap; 539 540 tvcpu = kvmppc_find_vcpu(kvm, vcpuid); 541 if (!tvcpu) 542 return H_PARAMETER; 543 544 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK; 545 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL || 546 subfunc == H_VPA_REG_SLB) { 547 /* Registering new area - address must be cache-line aligned */ 548 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa) 549 return H_PARAMETER; 550 551 /* convert logical addr to kernel addr and read length */ 552 va = kvmppc_pin_guest_page(kvm, vpa, &nb); 553 if (va == NULL) 554 return H_PARAMETER; 555 if (subfunc == H_VPA_REG_VPA) 556 len = be16_to_cpu(((struct reg_vpa *)va)->length.hword); 557 else 558 len = be32_to_cpu(((struct reg_vpa *)va)->length.word); 559 kvmppc_unpin_guest_page(kvm, va, vpa, false); 560 561 /* Check length */ 562 if (len > nb || len < sizeof(struct reg_vpa)) 563 return H_PARAMETER; 564 } else { 565 vpa = 0; 566 len = 0; 567 } 568 569 err = H_PARAMETER; 570 vpap = NULL; 571 spin_lock(&tvcpu->arch.vpa_update_lock); 572 573 switch (subfunc) { 574 case H_VPA_REG_VPA: /* register VPA */ 575 /* 576 * The size of our lppaca is 1kB because of the way we align 577 * it for the guest to avoid crossing a 4kB boundary. We only 578 * use 640 bytes of the structure though, so we should accept 579 * clients that set a size of 640. 580 */ 581 BUILD_BUG_ON(sizeof(struct lppaca) != 640); 582 if (len < sizeof(struct lppaca)) 583 break; 584 vpap = &tvcpu->arch.vpa; 585 err = 0; 586 break; 587 588 case H_VPA_REG_DTL: /* register DTL */ 589 if (len < sizeof(struct dtl_entry)) 590 break; 591 len -= len % sizeof(struct dtl_entry); 592 593 /* Check that they have previously registered a VPA */ 594 err = H_RESOURCE; 595 if (!vpa_is_registered(&tvcpu->arch.vpa)) 596 break; 597 598 vpap = &tvcpu->arch.dtl; 599 err = 0; 600 break; 601 602 case H_VPA_REG_SLB: /* register SLB shadow buffer */ 603 /* Check that they have previously registered a VPA */ 604 err = H_RESOURCE; 605 if (!vpa_is_registered(&tvcpu->arch.vpa)) 606 break; 607 608 vpap = &tvcpu->arch.slb_shadow; 609 err = 0; 610 break; 611 612 case H_VPA_DEREG_VPA: /* deregister VPA */ 613 /* Check they don't still have a DTL or SLB buf registered */ 614 err = H_RESOURCE; 615 if (vpa_is_registered(&tvcpu->arch.dtl) || 616 vpa_is_registered(&tvcpu->arch.slb_shadow)) 617 break; 618 619 vpap = &tvcpu->arch.vpa; 620 err = 0; 621 break; 622 623 case H_VPA_DEREG_DTL: /* deregister DTL */ 624 vpap = &tvcpu->arch.dtl; 625 err = 0; 626 break; 627 628 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */ 629 vpap = &tvcpu->arch.slb_shadow; 630 err = 0; 631 break; 632 } 633 634 if (vpap) { 635 vpap->next_gpa = vpa; 636 vpap->len = len; 637 vpap->update_pending = 1; 638 } 639 640 spin_unlock(&tvcpu->arch.vpa_update_lock); 641 642 return err; 643 } 644 645 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap) 646 { 647 struct kvm *kvm = vcpu->kvm; 648 void *va; 649 unsigned long nb; 650 unsigned long gpa; 651 652 /* 653 * We need to pin the page pointed to by vpap->next_gpa, 654 * but we can't call kvmppc_pin_guest_page under the lock 655 * as it does get_user_pages() and down_read(). So we 656 * have to drop the lock, pin the page, then get the lock 657 * again and check that a new area didn't get registered 658 * in the meantime. 659 */ 660 for (;;) { 661 gpa = vpap->next_gpa; 662 spin_unlock(&vcpu->arch.vpa_update_lock); 663 va = NULL; 664 nb = 0; 665 if (gpa) 666 va = kvmppc_pin_guest_page(kvm, gpa, &nb); 667 spin_lock(&vcpu->arch.vpa_update_lock); 668 if (gpa == vpap->next_gpa) 669 break; 670 /* sigh... unpin that one and try again */ 671 if (va) 672 kvmppc_unpin_guest_page(kvm, va, gpa, false); 673 } 674 675 vpap->update_pending = 0; 676 if (va && nb < vpap->len) { 677 /* 678 * If it's now too short, it must be that userspace 679 * has changed the mappings underlying guest memory, 680 * so unregister the region. 681 */ 682 kvmppc_unpin_guest_page(kvm, va, gpa, false); 683 va = NULL; 684 } 685 if (vpap->pinned_addr) 686 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa, 687 vpap->dirty); 688 vpap->gpa = gpa; 689 vpap->pinned_addr = va; 690 vpap->dirty = false; 691 if (va) 692 vpap->pinned_end = va + vpap->len; 693 } 694 695 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu) 696 { 697 if (!(vcpu->arch.vpa.update_pending || 698 vcpu->arch.slb_shadow.update_pending || 699 vcpu->arch.dtl.update_pending)) 700 return; 701 702 spin_lock(&vcpu->arch.vpa_update_lock); 703 if (vcpu->arch.vpa.update_pending) { 704 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa); 705 if (vcpu->arch.vpa.pinned_addr) 706 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr); 707 } 708 if (vcpu->arch.dtl.update_pending) { 709 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl); 710 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr; 711 vcpu->arch.dtl_index = 0; 712 } 713 if (vcpu->arch.slb_shadow.update_pending) 714 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow); 715 spin_unlock(&vcpu->arch.vpa_update_lock); 716 } 717 718 /* 719 * Return the accumulated stolen time for the vcore up until `now'. 720 * The caller should hold the vcore lock. 721 */ 722 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now) 723 { 724 u64 p; 725 unsigned long flags; 726 727 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)); 728 729 spin_lock_irqsave(&vc->stoltb_lock, flags); 730 p = vc->stolen_tb; 731 if (vc->vcore_state != VCORE_INACTIVE && 732 vc->preempt_tb != TB_NIL) 733 p += now - vc->preempt_tb; 734 spin_unlock_irqrestore(&vc->stoltb_lock, flags); 735 return p; 736 } 737 738 static void __kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu, 739 struct lppaca *vpa, 740 unsigned int pcpu, u64 now, 741 unsigned long stolen) 742 { 743 struct dtl_entry *dt; 744 745 dt = vcpu->arch.dtl_ptr; 746 747 if (!dt) 748 return; 749 750 dt->dispatch_reason = 7; 751 dt->preempt_reason = 0; 752 dt->processor_id = cpu_to_be16(pcpu + vcpu->arch.ptid); 753 dt->enqueue_to_dispatch_time = cpu_to_be32(stolen); 754 dt->ready_to_enqueue_time = 0; 755 dt->waiting_to_ready_time = 0; 756 dt->timebase = cpu_to_be64(now); 757 dt->fault_addr = 0; 758 dt->srr0 = cpu_to_be64(kvmppc_get_pc(vcpu)); 759 dt->srr1 = cpu_to_be64(vcpu->arch.shregs.msr); 760 761 ++dt; 762 if (dt == vcpu->arch.dtl.pinned_end) 763 dt = vcpu->arch.dtl.pinned_addr; 764 vcpu->arch.dtl_ptr = dt; 765 /* order writing *dt vs. writing vpa->dtl_idx */ 766 smp_wmb(); 767 vpa->dtl_idx = cpu_to_be64(++vcpu->arch.dtl_index); 768 769 /* vcpu->arch.dtl.dirty is set by the caller */ 770 } 771 772 static void kvmppc_update_vpa_dispatch(struct kvm_vcpu *vcpu, 773 struct kvmppc_vcore *vc) 774 { 775 struct lppaca *vpa; 776 unsigned long stolen; 777 unsigned long core_stolen; 778 u64 now; 779 unsigned long flags; 780 781 vpa = vcpu->arch.vpa.pinned_addr; 782 if (!vpa) 783 return; 784 785 now = mftb(); 786 787 core_stolen = vcore_stolen_time(vc, now); 788 stolen = core_stolen - vcpu->arch.stolen_logged; 789 vcpu->arch.stolen_logged = core_stolen; 790 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags); 791 stolen += vcpu->arch.busy_stolen; 792 vcpu->arch.busy_stolen = 0; 793 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags); 794 795 vpa->enqueue_dispatch_tb = cpu_to_be64(be64_to_cpu(vpa->enqueue_dispatch_tb) + stolen); 796 797 __kvmppc_create_dtl_entry(vcpu, vpa, vc->pcpu, now + vc->tb_offset, stolen); 798 799 vcpu->arch.vpa.dirty = true; 800 } 801 802 static void kvmppc_update_vpa_dispatch_p9(struct kvm_vcpu *vcpu, 803 struct kvmppc_vcore *vc, 804 u64 now) 805 { 806 struct lppaca *vpa; 807 unsigned long stolen; 808 unsigned long stolen_delta; 809 810 vpa = vcpu->arch.vpa.pinned_addr; 811 if (!vpa) 812 return; 813 814 stolen = vc->stolen_tb; 815 stolen_delta = stolen - vcpu->arch.stolen_logged; 816 vcpu->arch.stolen_logged = stolen; 817 818 vpa->enqueue_dispatch_tb = cpu_to_be64(stolen); 819 820 __kvmppc_create_dtl_entry(vcpu, vpa, vc->pcpu, now, stolen_delta); 821 822 vcpu->arch.vpa.dirty = true; 823 } 824 825 /* See if there is a doorbell interrupt pending for a vcpu */ 826 static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu) 827 { 828 int thr; 829 struct kvmppc_vcore *vc; 830 831 if (vcpu->arch.doorbell_request) 832 return true; 833 if (cpu_has_feature(CPU_FTR_ARCH_300)) 834 return false; 835 /* 836 * Ensure that the read of vcore->dpdes comes after the read 837 * of vcpu->doorbell_request. This barrier matches the 838 * smp_wmb() in kvmppc_guest_entry_inject(). 839 */ 840 smp_rmb(); 841 vc = vcpu->arch.vcore; 842 thr = vcpu->vcpu_id - vc->first_vcpuid; 843 return !!(vc->dpdes & (1 << thr)); 844 } 845 846 static bool kvmppc_power8_compatible(struct kvm_vcpu *vcpu) 847 { 848 if (vcpu->arch.vcore->arch_compat >= PVR_ARCH_207) 849 return true; 850 if ((!vcpu->arch.vcore->arch_compat) && 851 cpu_has_feature(CPU_FTR_ARCH_207S)) 852 return true; 853 return false; 854 } 855 856 static int kvmppc_h_set_mode(struct kvm_vcpu *vcpu, unsigned long mflags, 857 unsigned long resource, unsigned long value1, 858 unsigned long value2) 859 { 860 switch (resource) { 861 case H_SET_MODE_RESOURCE_SET_CIABR: 862 if (!kvmppc_power8_compatible(vcpu)) 863 return H_P2; 864 if (value2) 865 return H_P4; 866 if (mflags) 867 return H_UNSUPPORTED_FLAG_START; 868 /* Guests can't breakpoint the hypervisor */ 869 if ((value1 & CIABR_PRIV) == CIABR_PRIV_HYPER) 870 return H_P3; 871 vcpu->arch.ciabr = value1; 872 return H_SUCCESS; 873 case H_SET_MODE_RESOURCE_SET_DAWR0: 874 if (!kvmppc_power8_compatible(vcpu)) 875 return H_P2; 876 if (!ppc_breakpoint_available()) 877 return H_P2; 878 if (mflags) 879 return H_UNSUPPORTED_FLAG_START; 880 if (value2 & DABRX_HYP) 881 return H_P4; 882 vcpu->arch.dawr0 = value1; 883 vcpu->arch.dawrx0 = value2; 884 return H_SUCCESS; 885 case H_SET_MODE_RESOURCE_SET_DAWR1: 886 if (!kvmppc_power8_compatible(vcpu)) 887 return H_P2; 888 if (!ppc_breakpoint_available()) 889 return H_P2; 890 if (!cpu_has_feature(CPU_FTR_DAWR1)) 891 return H_P2; 892 if (!vcpu->kvm->arch.dawr1_enabled) 893 return H_FUNCTION; 894 if (mflags) 895 return H_UNSUPPORTED_FLAG_START; 896 if (value2 & DABRX_HYP) 897 return H_P4; 898 vcpu->arch.dawr1 = value1; 899 vcpu->arch.dawrx1 = value2; 900 return H_SUCCESS; 901 case H_SET_MODE_RESOURCE_ADDR_TRANS_MODE: 902 /* 903 * KVM does not support mflags=2 (AIL=2) and AIL=1 is reserved. 904 * Keep this in synch with kvmppc_filter_guest_lpcr_hv. 905 */ 906 if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) && 907 kvmhv_vcpu_is_radix(vcpu) && mflags == 3) 908 return H_UNSUPPORTED_FLAG_START; 909 return H_TOO_HARD; 910 default: 911 return H_TOO_HARD; 912 } 913 } 914 915 /* Copy guest memory in place - must reside within a single memslot */ 916 static int kvmppc_copy_guest(struct kvm *kvm, gpa_t to, gpa_t from, 917 unsigned long len) 918 { 919 struct kvm_memory_slot *to_memslot = NULL; 920 struct kvm_memory_slot *from_memslot = NULL; 921 unsigned long to_addr, from_addr; 922 int r; 923 924 /* Get HPA for from address */ 925 from_memslot = gfn_to_memslot(kvm, from >> PAGE_SHIFT); 926 if (!from_memslot) 927 return -EFAULT; 928 if ((from + len) >= ((from_memslot->base_gfn + from_memslot->npages) 929 << PAGE_SHIFT)) 930 return -EINVAL; 931 from_addr = gfn_to_hva_memslot(from_memslot, from >> PAGE_SHIFT); 932 if (kvm_is_error_hva(from_addr)) 933 return -EFAULT; 934 from_addr |= (from & (PAGE_SIZE - 1)); 935 936 /* Get HPA for to address */ 937 to_memslot = gfn_to_memslot(kvm, to >> PAGE_SHIFT); 938 if (!to_memslot) 939 return -EFAULT; 940 if ((to + len) >= ((to_memslot->base_gfn + to_memslot->npages) 941 << PAGE_SHIFT)) 942 return -EINVAL; 943 to_addr = gfn_to_hva_memslot(to_memslot, to >> PAGE_SHIFT); 944 if (kvm_is_error_hva(to_addr)) 945 return -EFAULT; 946 to_addr |= (to & (PAGE_SIZE - 1)); 947 948 /* Perform copy */ 949 r = raw_copy_in_user((void __user *)to_addr, (void __user *)from_addr, 950 len); 951 if (r) 952 return -EFAULT; 953 mark_page_dirty(kvm, to >> PAGE_SHIFT); 954 return 0; 955 } 956 957 static long kvmppc_h_page_init(struct kvm_vcpu *vcpu, unsigned long flags, 958 unsigned long dest, unsigned long src) 959 { 960 u64 pg_sz = SZ_4K; /* 4K page size */ 961 u64 pg_mask = SZ_4K - 1; 962 int ret; 963 964 /* Check for invalid flags (H_PAGE_SET_LOANED covers all CMO flags) */ 965 if (flags & ~(H_ICACHE_INVALIDATE | H_ICACHE_SYNCHRONIZE | 966 H_ZERO_PAGE | H_COPY_PAGE | H_PAGE_SET_LOANED)) 967 return H_PARAMETER; 968 969 /* dest (and src if copy_page flag set) must be page aligned */ 970 if ((dest & pg_mask) || ((flags & H_COPY_PAGE) && (src & pg_mask))) 971 return H_PARAMETER; 972 973 /* zero and/or copy the page as determined by the flags */ 974 if (flags & H_COPY_PAGE) { 975 ret = kvmppc_copy_guest(vcpu->kvm, dest, src, pg_sz); 976 if (ret < 0) 977 return H_PARAMETER; 978 } else if (flags & H_ZERO_PAGE) { 979 ret = kvm_clear_guest(vcpu->kvm, dest, pg_sz); 980 if (ret < 0) 981 return H_PARAMETER; 982 } 983 984 /* We can ignore the remaining flags */ 985 986 return H_SUCCESS; 987 } 988 989 static int kvm_arch_vcpu_yield_to(struct kvm_vcpu *target) 990 { 991 struct kvmppc_vcore *vcore = target->arch.vcore; 992 993 /* 994 * We expect to have been called by the real mode handler 995 * (kvmppc_rm_h_confer()) which would have directly returned 996 * H_SUCCESS if the source vcore wasn't idle (e.g. if it may 997 * have useful work to do and should not confer) so we don't 998 * recheck that here. 999 * 1000 * In the case of the P9 single vcpu per vcore case, the real 1001 * mode handler is not called but no other threads are in the 1002 * source vcore. 1003 */ 1004 if (!cpu_has_feature(CPU_FTR_ARCH_300)) { 1005 spin_lock(&vcore->lock); 1006 if (target->arch.state == KVMPPC_VCPU_RUNNABLE && 1007 vcore->vcore_state != VCORE_INACTIVE && 1008 vcore->runner) 1009 target = vcore->runner; 1010 spin_unlock(&vcore->lock); 1011 } 1012 1013 return kvm_vcpu_yield_to(target); 1014 } 1015 1016 static int kvmppc_get_yield_count(struct kvm_vcpu *vcpu) 1017 { 1018 int yield_count = 0; 1019 struct lppaca *lppaca; 1020 1021 spin_lock(&vcpu->arch.vpa_update_lock); 1022 lppaca = (struct lppaca *)vcpu->arch.vpa.pinned_addr; 1023 if (lppaca) 1024 yield_count = be32_to_cpu(lppaca->yield_count); 1025 spin_unlock(&vcpu->arch.vpa_update_lock); 1026 return yield_count; 1027 } 1028 1029 /* 1030 * H_RPT_INVALIDATE hcall handler for nested guests. 1031 * 1032 * Handles only nested process-scoped invalidation requests in L0. 1033 */ 1034 static int kvmppc_nested_h_rpt_invalidate(struct kvm_vcpu *vcpu) 1035 { 1036 unsigned long type = kvmppc_get_gpr(vcpu, 6); 1037 unsigned long pid, pg_sizes, start, end; 1038 1039 /* 1040 * The partition-scoped invalidations aren't handled here in L0. 1041 */ 1042 if (type & H_RPTI_TYPE_NESTED) 1043 return RESUME_HOST; 1044 1045 pid = kvmppc_get_gpr(vcpu, 4); 1046 pg_sizes = kvmppc_get_gpr(vcpu, 7); 1047 start = kvmppc_get_gpr(vcpu, 8); 1048 end = kvmppc_get_gpr(vcpu, 9); 1049 1050 do_h_rpt_invalidate_prt(pid, vcpu->arch.nested->shadow_lpid, 1051 type, pg_sizes, start, end); 1052 1053 kvmppc_set_gpr(vcpu, 3, H_SUCCESS); 1054 return RESUME_GUEST; 1055 } 1056 1057 static long kvmppc_h_rpt_invalidate(struct kvm_vcpu *vcpu, 1058 unsigned long id, unsigned long target, 1059 unsigned long type, unsigned long pg_sizes, 1060 unsigned long start, unsigned long end) 1061 { 1062 if (!kvm_is_radix(vcpu->kvm)) 1063 return H_UNSUPPORTED; 1064 1065 if (end < start) 1066 return H_P5; 1067 1068 /* 1069 * Partition-scoped invalidation for nested guests. 1070 */ 1071 if (type & H_RPTI_TYPE_NESTED) { 1072 if (!nesting_enabled(vcpu->kvm)) 1073 return H_FUNCTION; 1074 1075 /* Support only cores as target */ 1076 if (target != H_RPTI_TARGET_CMMU) 1077 return H_P2; 1078 1079 return do_h_rpt_invalidate_pat(vcpu, id, type, pg_sizes, 1080 start, end); 1081 } 1082 1083 /* 1084 * Process-scoped invalidation for L1 guests. 1085 */ 1086 do_h_rpt_invalidate_prt(id, vcpu->kvm->arch.lpid, 1087 type, pg_sizes, start, end); 1088 return H_SUCCESS; 1089 } 1090 1091 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu) 1092 { 1093 struct kvm *kvm = vcpu->kvm; 1094 unsigned long req = kvmppc_get_gpr(vcpu, 3); 1095 unsigned long target, ret = H_SUCCESS; 1096 int yield_count; 1097 struct kvm_vcpu *tvcpu; 1098 int idx, rc; 1099 1100 if (req <= MAX_HCALL_OPCODE && 1101 !test_bit(req/4, vcpu->kvm->arch.enabled_hcalls)) 1102 return RESUME_HOST; 1103 1104 switch (req) { 1105 case H_REMOVE: 1106 ret = kvmppc_h_remove(vcpu, kvmppc_get_gpr(vcpu, 4), 1107 kvmppc_get_gpr(vcpu, 5), 1108 kvmppc_get_gpr(vcpu, 6)); 1109 if (ret == H_TOO_HARD) 1110 return RESUME_HOST; 1111 break; 1112 case H_ENTER: 1113 ret = kvmppc_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4), 1114 kvmppc_get_gpr(vcpu, 5), 1115 kvmppc_get_gpr(vcpu, 6), 1116 kvmppc_get_gpr(vcpu, 7)); 1117 if (ret == H_TOO_HARD) 1118 return RESUME_HOST; 1119 break; 1120 case H_READ: 1121 ret = kvmppc_h_read(vcpu, kvmppc_get_gpr(vcpu, 4), 1122 kvmppc_get_gpr(vcpu, 5)); 1123 if (ret == H_TOO_HARD) 1124 return RESUME_HOST; 1125 break; 1126 case H_CLEAR_MOD: 1127 ret = kvmppc_h_clear_mod(vcpu, kvmppc_get_gpr(vcpu, 4), 1128 kvmppc_get_gpr(vcpu, 5)); 1129 if (ret == H_TOO_HARD) 1130 return RESUME_HOST; 1131 break; 1132 case H_CLEAR_REF: 1133 ret = kvmppc_h_clear_ref(vcpu, kvmppc_get_gpr(vcpu, 4), 1134 kvmppc_get_gpr(vcpu, 5)); 1135 if (ret == H_TOO_HARD) 1136 return RESUME_HOST; 1137 break; 1138 case H_PROTECT: 1139 ret = kvmppc_h_protect(vcpu, kvmppc_get_gpr(vcpu, 4), 1140 kvmppc_get_gpr(vcpu, 5), 1141 kvmppc_get_gpr(vcpu, 6)); 1142 if (ret == H_TOO_HARD) 1143 return RESUME_HOST; 1144 break; 1145 case H_BULK_REMOVE: 1146 ret = kvmppc_h_bulk_remove(vcpu); 1147 if (ret == H_TOO_HARD) 1148 return RESUME_HOST; 1149 break; 1150 1151 case H_CEDE: 1152 break; 1153 case H_PROD: 1154 target = kvmppc_get_gpr(vcpu, 4); 1155 tvcpu = kvmppc_find_vcpu(kvm, target); 1156 if (!tvcpu) { 1157 ret = H_PARAMETER; 1158 break; 1159 } 1160 tvcpu->arch.prodded = 1; 1161 smp_mb(); /* This orders prodded store vs ceded load */ 1162 if (tvcpu->arch.ceded) 1163 kvmppc_fast_vcpu_kick_hv(tvcpu); 1164 break; 1165 case H_CONFER: 1166 target = kvmppc_get_gpr(vcpu, 4); 1167 if (target == -1) 1168 break; 1169 tvcpu = kvmppc_find_vcpu(kvm, target); 1170 if (!tvcpu) { 1171 ret = H_PARAMETER; 1172 break; 1173 } 1174 yield_count = kvmppc_get_gpr(vcpu, 5); 1175 if (kvmppc_get_yield_count(tvcpu) != yield_count) 1176 break; 1177 kvm_arch_vcpu_yield_to(tvcpu); 1178 break; 1179 case H_REGISTER_VPA: 1180 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4), 1181 kvmppc_get_gpr(vcpu, 5), 1182 kvmppc_get_gpr(vcpu, 6)); 1183 break; 1184 case H_RTAS: 1185 if (list_empty(&kvm->arch.rtas_tokens)) 1186 return RESUME_HOST; 1187 1188 idx = srcu_read_lock(&kvm->srcu); 1189 rc = kvmppc_rtas_hcall(vcpu); 1190 srcu_read_unlock(&kvm->srcu, idx); 1191 1192 if (rc == -ENOENT) 1193 return RESUME_HOST; 1194 else if (rc == 0) 1195 break; 1196 1197 /* Send the error out to userspace via KVM_RUN */ 1198 return rc; 1199 case H_LOGICAL_CI_LOAD: 1200 ret = kvmppc_h_logical_ci_load(vcpu); 1201 if (ret == H_TOO_HARD) 1202 return RESUME_HOST; 1203 break; 1204 case H_LOGICAL_CI_STORE: 1205 ret = kvmppc_h_logical_ci_store(vcpu); 1206 if (ret == H_TOO_HARD) 1207 return RESUME_HOST; 1208 break; 1209 case H_SET_MODE: 1210 ret = kvmppc_h_set_mode(vcpu, kvmppc_get_gpr(vcpu, 4), 1211 kvmppc_get_gpr(vcpu, 5), 1212 kvmppc_get_gpr(vcpu, 6), 1213 kvmppc_get_gpr(vcpu, 7)); 1214 if (ret == H_TOO_HARD) 1215 return RESUME_HOST; 1216 break; 1217 case H_XIRR: 1218 case H_CPPR: 1219 case H_EOI: 1220 case H_IPI: 1221 case H_IPOLL: 1222 case H_XIRR_X: 1223 if (kvmppc_xics_enabled(vcpu)) { 1224 if (xics_on_xive()) { 1225 ret = H_NOT_AVAILABLE; 1226 return RESUME_GUEST; 1227 } 1228 ret = kvmppc_xics_hcall(vcpu, req); 1229 break; 1230 } 1231 return RESUME_HOST; 1232 case H_SET_DABR: 1233 ret = kvmppc_h_set_dabr(vcpu, kvmppc_get_gpr(vcpu, 4)); 1234 break; 1235 case H_SET_XDABR: 1236 ret = kvmppc_h_set_xdabr(vcpu, kvmppc_get_gpr(vcpu, 4), 1237 kvmppc_get_gpr(vcpu, 5)); 1238 break; 1239 #ifdef CONFIG_SPAPR_TCE_IOMMU 1240 case H_GET_TCE: 1241 ret = kvmppc_h_get_tce(vcpu, kvmppc_get_gpr(vcpu, 4), 1242 kvmppc_get_gpr(vcpu, 5)); 1243 if (ret == H_TOO_HARD) 1244 return RESUME_HOST; 1245 break; 1246 case H_PUT_TCE: 1247 ret = kvmppc_h_put_tce(vcpu, kvmppc_get_gpr(vcpu, 4), 1248 kvmppc_get_gpr(vcpu, 5), 1249 kvmppc_get_gpr(vcpu, 6)); 1250 if (ret == H_TOO_HARD) 1251 return RESUME_HOST; 1252 break; 1253 case H_PUT_TCE_INDIRECT: 1254 ret = kvmppc_h_put_tce_indirect(vcpu, kvmppc_get_gpr(vcpu, 4), 1255 kvmppc_get_gpr(vcpu, 5), 1256 kvmppc_get_gpr(vcpu, 6), 1257 kvmppc_get_gpr(vcpu, 7)); 1258 if (ret == H_TOO_HARD) 1259 return RESUME_HOST; 1260 break; 1261 case H_STUFF_TCE: 1262 ret = kvmppc_h_stuff_tce(vcpu, kvmppc_get_gpr(vcpu, 4), 1263 kvmppc_get_gpr(vcpu, 5), 1264 kvmppc_get_gpr(vcpu, 6), 1265 kvmppc_get_gpr(vcpu, 7)); 1266 if (ret == H_TOO_HARD) 1267 return RESUME_HOST; 1268 break; 1269 #endif 1270 case H_RANDOM: 1271 if (!arch_get_random_seed_longs(&vcpu->arch.regs.gpr[4], 1)) 1272 ret = H_HARDWARE; 1273 break; 1274 case H_RPT_INVALIDATE: 1275 ret = kvmppc_h_rpt_invalidate(vcpu, kvmppc_get_gpr(vcpu, 4), 1276 kvmppc_get_gpr(vcpu, 5), 1277 kvmppc_get_gpr(vcpu, 6), 1278 kvmppc_get_gpr(vcpu, 7), 1279 kvmppc_get_gpr(vcpu, 8), 1280 kvmppc_get_gpr(vcpu, 9)); 1281 break; 1282 1283 case H_SET_PARTITION_TABLE: 1284 ret = H_FUNCTION; 1285 if (nesting_enabled(kvm)) 1286 ret = kvmhv_set_partition_table(vcpu); 1287 break; 1288 case H_ENTER_NESTED: 1289 ret = H_FUNCTION; 1290 if (!nesting_enabled(kvm)) 1291 break; 1292 ret = kvmhv_enter_nested_guest(vcpu); 1293 if (ret == H_INTERRUPT) { 1294 kvmppc_set_gpr(vcpu, 3, 0); 1295 vcpu->arch.hcall_needed = 0; 1296 return -EINTR; 1297 } else if (ret == H_TOO_HARD) { 1298 kvmppc_set_gpr(vcpu, 3, 0); 1299 vcpu->arch.hcall_needed = 0; 1300 return RESUME_HOST; 1301 } 1302 break; 1303 case H_TLB_INVALIDATE: 1304 ret = H_FUNCTION; 1305 if (nesting_enabled(kvm)) 1306 ret = kvmhv_do_nested_tlbie(vcpu); 1307 break; 1308 case H_COPY_TOFROM_GUEST: 1309 ret = H_FUNCTION; 1310 if (nesting_enabled(kvm)) 1311 ret = kvmhv_copy_tofrom_guest_nested(vcpu); 1312 break; 1313 case H_PAGE_INIT: 1314 ret = kvmppc_h_page_init(vcpu, kvmppc_get_gpr(vcpu, 4), 1315 kvmppc_get_gpr(vcpu, 5), 1316 kvmppc_get_gpr(vcpu, 6)); 1317 break; 1318 case H_SVM_PAGE_IN: 1319 ret = H_UNSUPPORTED; 1320 if (kvmppc_get_srr1(vcpu) & MSR_S) 1321 ret = kvmppc_h_svm_page_in(kvm, 1322 kvmppc_get_gpr(vcpu, 4), 1323 kvmppc_get_gpr(vcpu, 5), 1324 kvmppc_get_gpr(vcpu, 6)); 1325 break; 1326 case H_SVM_PAGE_OUT: 1327 ret = H_UNSUPPORTED; 1328 if (kvmppc_get_srr1(vcpu) & MSR_S) 1329 ret = kvmppc_h_svm_page_out(kvm, 1330 kvmppc_get_gpr(vcpu, 4), 1331 kvmppc_get_gpr(vcpu, 5), 1332 kvmppc_get_gpr(vcpu, 6)); 1333 break; 1334 case H_SVM_INIT_START: 1335 ret = H_UNSUPPORTED; 1336 if (kvmppc_get_srr1(vcpu) & MSR_S) 1337 ret = kvmppc_h_svm_init_start(kvm); 1338 break; 1339 case H_SVM_INIT_DONE: 1340 ret = H_UNSUPPORTED; 1341 if (kvmppc_get_srr1(vcpu) & MSR_S) 1342 ret = kvmppc_h_svm_init_done(kvm); 1343 break; 1344 case H_SVM_INIT_ABORT: 1345 /* 1346 * Even if that call is made by the Ultravisor, the SSR1 value 1347 * is the guest context one, with the secure bit clear as it has 1348 * not yet been secured. So we can't check it here. 1349 * Instead the kvm->arch.secure_guest flag is checked inside 1350 * kvmppc_h_svm_init_abort(). 1351 */ 1352 ret = kvmppc_h_svm_init_abort(kvm); 1353 break; 1354 1355 default: 1356 return RESUME_HOST; 1357 } 1358 WARN_ON_ONCE(ret == H_TOO_HARD); 1359 kvmppc_set_gpr(vcpu, 3, ret); 1360 vcpu->arch.hcall_needed = 0; 1361 return RESUME_GUEST; 1362 } 1363 1364 /* 1365 * Handle H_CEDE in the P9 path where we don't call the real-mode hcall 1366 * handlers in book3s_hv_rmhandlers.S. 1367 * 1368 * This has to be done early, not in kvmppc_pseries_do_hcall(), so 1369 * that the cede logic in kvmppc_run_single_vcpu() works properly. 1370 */ 1371 static void kvmppc_cede(struct kvm_vcpu *vcpu) 1372 { 1373 vcpu->arch.shregs.msr |= MSR_EE; 1374 vcpu->arch.ceded = 1; 1375 smp_mb(); 1376 if (vcpu->arch.prodded) { 1377 vcpu->arch.prodded = 0; 1378 smp_mb(); 1379 vcpu->arch.ceded = 0; 1380 } 1381 } 1382 1383 static int kvmppc_hcall_impl_hv(unsigned long cmd) 1384 { 1385 switch (cmd) { 1386 case H_CEDE: 1387 case H_PROD: 1388 case H_CONFER: 1389 case H_REGISTER_VPA: 1390 case H_SET_MODE: 1391 #ifdef CONFIG_SPAPR_TCE_IOMMU 1392 case H_GET_TCE: 1393 case H_PUT_TCE: 1394 case H_PUT_TCE_INDIRECT: 1395 case H_STUFF_TCE: 1396 #endif 1397 case H_LOGICAL_CI_LOAD: 1398 case H_LOGICAL_CI_STORE: 1399 #ifdef CONFIG_KVM_XICS 1400 case H_XIRR: 1401 case H_CPPR: 1402 case H_EOI: 1403 case H_IPI: 1404 case H_IPOLL: 1405 case H_XIRR_X: 1406 #endif 1407 case H_PAGE_INIT: 1408 case H_RPT_INVALIDATE: 1409 return 1; 1410 } 1411 1412 /* See if it's in the real-mode table */ 1413 return kvmppc_hcall_impl_hv_realmode(cmd); 1414 } 1415 1416 static int kvmppc_emulate_debug_inst(struct kvm_vcpu *vcpu) 1417 { 1418 u32 last_inst; 1419 1420 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) != 1421 EMULATE_DONE) { 1422 /* 1423 * Fetch failed, so return to guest and 1424 * try executing it again. 1425 */ 1426 return RESUME_GUEST; 1427 } 1428 1429 if (last_inst == KVMPPC_INST_SW_BREAKPOINT) { 1430 vcpu->run->exit_reason = KVM_EXIT_DEBUG; 1431 vcpu->run->debug.arch.address = kvmppc_get_pc(vcpu); 1432 return RESUME_HOST; 1433 } else { 1434 kvmppc_core_queue_program(vcpu, SRR1_PROGILL); 1435 return RESUME_GUEST; 1436 } 1437 } 1438 1439 static void do_nothing(void *x) 1440 { 1441 } 1442 1443 static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu) 1444 { 1445 int thr, cpu, pcpu, nthreads; 1446 struct kvm_vcpu *v; 1447 unsigned long dpdes; 1448 1449 nthreads = vcpu->kvm->arch.emul_smt_mode; 1450 dpdes = 0; 1451 cpu = vcpu->vcpu_id & ~(nthreads - 1); 1452 for (thr = 0; thr < nthreads; ++thr, ++cpu) { 1453 v = kvmppc_find_vcpu(vcpu->kvm, cpu); 1454 if (!v) 1455 continue; 1456 /* 1457 * If the vcpu is currently running on a physical cpu thread, 1458 * interrupt it in order to pull it out of the guest briefly, 1459 * which will update its vcore->dpdes value. 1460 */ 1461 pcpu = READ_ONCE(v->cpu); 1462 if (pcpu >= 0) 1463 smp_call_function_single(pcpu, do_nothing, NULL, 1); 1464 if (kvmppc_doorbell_pending(v)) 1465 dpdes |= 1 << thr; 1466 } 1467 return dpdes; 1468 } 1469 1470 /* 1471 * On POWER9, emulate doorbell-related instructions in order to 1472 * give the guest the illusion of running on a multi-threaded core. 1473 * The instructions emulated are msgsndp, msgclrp, mfspr TIR, 1474 * and mfspr DPDES. 1475 */ 1476 static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu) 1477 { 1478 u32 inst, rb, thr; 1479 unsigned long arg; 1480 struct kvm *kvm = vcpu->kvm; 1481 struct kvm_vcpu *tvcpu; 1482 1483 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE) 1484 return RESUME_GUEST; 1485 if (get_op(inst) != 31) 1486 return EMULATE_FAIL; 1487 rb = get_rb(inst); 1488 thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1); 1489 switch (get_xop(inst)) { 1490 case OP_31_XOP_MSGSNDP: 1491 arg = kvmppc_get_gpr(vcpu, rb); 1492 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER) 1493 break; 1494 arg &= 0x7f; 1495 if (arg >= kvm->arch.emul_smt_mode) 1496 break; 1497 tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg); 1498 if (!tvcpu) 1499 break; 1500 if (!tvcpu->arch.doorbell_request) { 1501 tvcpu->arch.doorbell_request = 1; 1502 kvmppc_fast_vcpu_kick_hv(tvcpu); 1503 } 1504 break; 1505 case OP_31_XOP_MSGCLRP: 1506 arg = kvmppc_get_gpr(vcpu, rb); 1507 if (((arg >> 27) & 0x1f) != PPC_DBELL_SERVER) 1508 break; 1509 vcpu->arch.vcore->dpdes = 0; 1510 vcpu->arch.doorbell_request = 0; 1511 break; 1512 case OP_31_XOP_MFSPR: 1513 switch (get_sprn(inst)) { 1514 case SPRN_TIR: 1515 arg = thr; 1516 break; 1517 case SPRN_DPDES: 1518 arg = kvmppc_read_dpdes(vcpu); 1519 break; 1520 default: 1521 return EMULATE_FAIL; 1522 } 1523 kvmppc_set_gpr(vcpu, get_rt(inst), arg); 1524 break; 1525 default: 1526 return EMULATE_FAIL; 1527 } 1528 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4); 1529 return RESUME_GUEST; 1530 } 1531 1532 /* 1533 * If the lppaca had pmcregs_in_use clear when we exited the guest, then 1534 * HFSCR_PM is cleared for next entry. If the guest then tries to access 1535 * the PMU SPRs, we get this facility unavailable interrupt. Putting HFSCR_PM 1536 * back in the guest HFSCR will cause the next entry to load the PMU SPRs and 1537 * allow the guest access to continue. 1538 */ 1539 static int kvmppc_pmu_unavailable(struct kvm_vcpu *vcpu) 1540 { 1541 if (!(vcpu->arch.hfscr_permitted & HFSCR_PM)) 1542 return EMULATE_FAIL; 1543 1544 vcpu->arch.hfscr |= HFSCR_PM; 1545 1546 return RESUME_GUEST; 1547 } 1548 1549 static int kvmppc_ebb_unavailable(struct kvm_vcpu *vcpu) 1550 { 1551 if (!(vcpu->arch.hfscr_permitted & HFSCR_EBB)) 1552 return EMULATE_FAIL; 1553 1554 vcpu->arch.hfscr |= HFSCR_EBB; 1555 1556 return RESUME_GUEST; 1557 } 1558 1559 static int kvmppc_tm_unavailable(struct kvm_vcpu *vcpu) 1560 { 1561 if (!(vcpu->arch.hfscr_permitted & HFSCR_TM)) 1562 return EMULATE_FAIL; 1563 1564 vcpu->arch.hfscr |= HFSCR_TM; 1565 1566 return RESUME_GUEST; 1567 } 1568 1569 static int kvmppc_handle_exit_hv(struct kvm_vcpu *vcpu, 1570 struct task_struct *tsk) 1571 { 1572 struct kvm_run *run = vcpu->run; 1573 int r = RESUME_HOST; 1574 1575 vcpu->stat.sum_exits++; 1576 1577 /* 1578 * This can happen if an interrupt occurs in the last stages 1579 * of guest entry or the first stages of guest exit (i.e. after 1580 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV 1581 * and before setting it to KVM_GUEST_MODE_HOST_HV). 1582 * That can happen due to a bug, or due to a machine check 1583 * occurring at just the wrong time. 1584 */ 1585 if (vcpu->arch.shregs.msr & MSR_HV) { 1586 printk(KERN_EMERG "KVM trap in HV mode!\n"); 1587 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n", 1588 vcpu->arch.trap, kvmppc_get_pc(vcpu), 1589 vcpu->arch.shregs.msr); 1590 kvmppc_dump_regs(vcpu); 1591 run->exit_reason = KVM_EXIT_INTERNAL_ERROR; 1592 run->hw.hardware_exit_reason = vcpu->arch.trap; 1593 return RESUME_HOST; 1594 } 1595 run->exit_reason = KVM_EXIT_UNKNOWN; 1596 run->ready_for_interrupt_injection = 1; 1597 switch (vcpu->arch.trap) { 1598 /* We're good on these - the host merely wanted to get our attention */ 1599 case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER: 1600 WARN_ON_ONCE(1); /* Should never happen */ 1601 vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER; 1602 fallthrough; 1603 case BOOK3S_INTERRUPT_HV_DECREMENTER: 1604 vcpu->stat.dec_exits++; 1605 r = RESUME_GUEST; 1606 break; 1607 case BOOK3S_INTERRUPT_EXTERNAL: 1608 case BOOK3S_INTERRUPT_H_DOORBELL: 1609 case BOOK3S_INTERRUPT_H_VIRT: 1610 vcpu->stat.ext_intr_exits++; 1611 r = RESUME_GUEST; 1612 break; 1613 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/ 1614 case BOOK3S_INTERRUPT_HMI: 1615 case BOOK3S_INTERRUPT_PERFMON: 1616 case BOOK3S_INTERRUPT_SYSTEM_RESET: 1617 r = RESUME_GUEST; 1618 break; 1619 case BOOK3S_INTERRUPT_MACHINE_CHECK: { 1620 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL, 1621 DEFAULT_RATELIMIT_BURST); 1622 /* 1623 * Print the MCE event to host console. Ratelimit so the guest 1624 * can't flood the host log. 1625 */ 1626 if (__ratelimit(&rs)) 1627 machine_check_print_event_info(&vcpu->arch.mce_evt,false, true); 1628 1629 /* 1630 * If the guest can do FWNMI, exit to userspace so it can 1631 * deliver a FWNMI to the guest. 1632 * Otherwise we synthesize a machine check for the guest 1633 * so that it knows that the machine check occurred. 1634 */ 1635 if (!vcpu->kvm->arch.fwnmi_enabled) { 1636 ulong flags = vcpu->arch.shregs.msr & 0x083c0000; 1637 kvmppc_core_queue_machine_check(vcpu, flags); 1638 r = RESUME_GUEST; 1639 break; 1640 } 1641 1642 /* Exit to guest with KVM_EXIT_NMI as exit reason */ 1643 run->exit_reason = KVM_EXIT_NMI; 1644 run->hw.hardware_exit_reason = vcpu->arch.trap; 1645 /* Clear out the old NMI status from run->flags */ 1646 run->flags &= ~KVM_RUN_PPC_NMI_DISP_MASK; 1647 /* Now set the NMI status */ 1648 if (vcpu->arch.mce_evt.disposition == MCE_DISPOSITION_RECOVERED) 1649 run->flags |= KVM_RUN_PPC_NMI_DISP_FULLY_RECOV; 1650 else 1651 run->flags |= KVM_RUN_PPC_NMI_DISP_NOT_RECOV; 1652 1653 r = RESUME_HOST; 1654 break; 1655 } 1656 case BOOK3S_INTERRUPT_PROGRAM: 1657 { 1658 ulong flags; 1659 /* 1660 * Normally program interrupts are delivered directly 1661 * to the guest by the hardware, but we can get here 1662 * as a result of a hypervisor emulation interrupt 1663 * (e40) getting turned into a 700 by BML RTAS. 1664 */ 1665 flags = vcpu->arch.shregs.msr & 0x1f0000ull; 1666 kvmppc_core_queue_program(vcpu, flags); 1667 r = RESUME_GUEST; 1668 break; 1669 } 1670 case BOOK3S_INTERRUPT_SYSCALL: 1671 { 1672 int i; 1673 1674 if (unlikely(vcpu->arch.shregs.msr & MSR_PR)) { 1675 /* 1676 * Guest userspace executed sc 1. This can only be 1677 * reached by the P9 path because the old path 1678 * handles this case in realmode hcall handlers. 1679 */ 1680 if (!kvmhv_vcpu_is_radix(vcpu)) { 1681 /* 1682 * A guest could be running PR KVM, so this 1683 * may be a PR KVM hcall. It must be reflected 1684 * to the guest kernel as a sc interrupt. 1685 */ 1686 kvmppc_core_queue_syscall(vcpu); 1687 } else { 1688 /* 1689 * Radix guests can not run PR KVM or nested HV 1690 * hash guests which might run PR KVM, so this 1691 * is always a privilege fault. Send a program 1692 * check to guest kernel. 1693 */ 1694 kvmppc_core_queue_program(vcpu, SRR1_PROGPRIV); 1695 } 1696 r = RESUME_GUEST; 1697 break; 1698 } 1699 1700 /* 1701 * hcall - gather args and set exit_reason. This will next be 1702 * handled by kvmppc_pseries_do_hcall which may be able to deal 1703 * with it and resume guest, or may punt to userspace. 1704 */ 1705 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3); 1706 for (i = 0; i < 9; ++i) 1707 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i); 1708 run->exit_reason = KVM_EXIT_PAPR_HCALL; 1709 vcpu->arch.hcall_needed = 1; 1710 r = RESUME_HOST; 1711 break; 1712 } 1713 /* 1714 * We get these next two if the guest accesses a page which it thinks 1715 * it has mapped but which is not actually present, either because 1716 * it is for an emulated I/O device or because the corresonding 1717 * host page has been paged out. 1718 * 1719 * Any other HDSI/HISI interrupts have been handled already for P7/8 1720 * guests. For POWER9 hash guests not using rmhandlers, basic hash 1721 * fault handling is done here. 1722 */ 1723 case BOOK3S_INTERRUPT_H_DATA_STORAGE: { 1724 unsigned long vsid; 1725 long err; 1726 1727 if (cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG) && 1728 unlikely(vcpu->arch.fault_dsisr == HDSISR_CANARY)) { 1729 r = RESUME_GUEST; /* Just retry if it's the canary */ 1730 break; 1731 } 1732 1733 if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) { 1734 /* 1735 * Radix doesn't require anything, and pre-ISAv3.0 hash 1736 * already attempted to handle this in rmhandlers. The 1737 * hash fault handling below is v3 only (it uses ASDR 1738 * via fault_gpa). 1739 */ 1740 r = RESUME_PAGE_FAULT; 1741 break; 1742 } 1743 1744 if (!(vcpu->arch.fault_dsisr & (DSISR_NOHPTE | DSISR_PROTFAULT))) { 1745 kvmppc_core_queue_data_storage(vcpu, 1746 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr); 1747 r = RESUME_GUEST; 1748 break; 1749 } 1750 1751 if (!(vcpu->arch.shregs.msr & MSR_DR)) 1752 vsid = vcpu->kvm->arch.vrma_slb_v; 1753 else 1754 vsid = vcpu->arch.fault_gpa; 1755 1756 err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar, 1757 vsid, vcpu->arch.fault_dsisr, true); 1758 if (err == 0) { 1759 r = RESUME_GUEST; 1760 } else if (err == -1 || err == -2) { 1761 r = RESUME_PAGE_FAULT; 1762 } else { 1763 kvmppc_core_queue_data_storage(vcpu, 1764 vcpu->arch.fault_dar, err); 1765 r = RESUME_GUEST; 1766 } 1767 break; 1768 } 1769 case BOOK3S_INTERRUPT_H_INST_STORAGE: { 1770 unsigned long vsid; 1771 long err; 1772 1773 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu); 1774 vcpu->arch.fault_dsisr = vcpu->arch.shregs.msr & 1775 DSISR_SRR1_MATCH_64S; 1776 if (kvm_is_radix(vcpu->kvm) || !cpu_has_feature(CPU_FTR_ARCH_300)) { 1777 /* 1778 * Radix doesn't require anything, and pre-ISAv3.0 hash 1779 * already attempted to handle this in rmhandlers. The 1780 * hash fault handling below is v3 only (it uses ASDR 1781 * via fault_gpa). 1782 */ 1783 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE) 1784 vcpu->arch.fault_dsisr |= DSISR_ISSTORE; 1785 r = RESUME_PAGE_FAULT; 1786 break; 1787 } 1788 1789 if (!(vcpu->arch.fault_dsisr & SRR1_ISI_NOPT)) { 1790 kvmppc_core_queue_inst_storage(vcpu, 1791 vcpu->arch.fault_dsisr); 1792 r = RESUME_GUEST; 1793 break; 1794 } 1795 1796 if (!(vcpu->arch.shregs.msr & MSR_IR)) 1797 vsid = vcpu->kvm->arch.vrma_slb_v; 1798 else 1799 vsid = vcpu->arch.fault_gpa; 1800 1801 err = kvmppc_hpte_hv_fault(vcpu, vcpu->arch.fault_dar, 1802 vsid, vcpu->arch.fault_dsisr, false); 1803 if (err == 0) { 1804 r = RESUME_GUEST; 1805 } else if (err == -1) { 1806 r = RESUME_PAGE_FAULT; 1807 } else { 1808 kvmppc_core_queue_inst_storage(vcpu, err); 1809 r = RESUME_GUEST; 1810 } 1811 break; 1812 } 1813 1814 /* 1815 * This occurs if the guest executes an illegal instruction. 1816 * If the guest debug is disabled, generate a program interrupt 1817 * to the guest. If guest debug is enabled, we need to check 1818 * whether the instruction is a software breakpoint instruction. 1819 * Accordingly return to Guest or Host. 1820 */ 1821 case BOOK3S_INTERRUPT_H_EMUL_ASSIST: 1822 if (vcpu->arch.emul_inst != KVM_INST_FETCH_FAILED) 1823 vcpu->arch.last_inst = kvmppc_need_byteswap(vcpu) ? 1824 swab32(vcpu->arch.emul_inst) : 1825 vcpu->arch.emul_inst; 1826 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP) { 1827 r = kvmppc_emulate_debug_inst(vcpu); 1828 } else { 1829 kvmppc_core_queue_program(vcpu, SRR1_PROGILL); 1830 r = RESUME_GUEST; 1831 } 1832 break; 1833 1834 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 1835 case BOOK3S_INTERRUPT_HV_SOFTPATCH: 1836 /* 1837 * This occurs for various TM-related instructions that 1838 * we need to emulate on POWER9 DD2.2. We have already 1839 * handled the cases where the guest was in real-suspend 1840 * mode and was transitioning to transactional state. 1841 */ 1842 r = kvmhv_p9_tm_emulation(vcpu); 1843 if (r != -1) 1844 break; 1845 fallthrough; /* go to facility unavailable handler */ 1846 #endif 1847 1848 /* 1849 * This occurs if the guest (kernel or userspace), does something that 1850 * is prohibited by HFSCR. 1851 * On POWER9, this could be a doorbell instruction that we need 1852 * to emulate. 1853 * Otherwise, we just generate a program interrupt to the guest. 1854 */ 1855 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: { 1856 u64 cause = vcpu->arch.hfscr >> 56; 1857 1858 r = EMULATE_FAIL; 1859 if (cpu_has_feature(CPU_FTR_ARCH_300)) { 1860 if (cause == FSCR_MSGP_LG) 1861 r = kvmppc_emulate_doorbell_instr(vcpu); 1862 if (cause == FSCR_PM_LG) 1863 r = kvmppc_pmu_unavailable(vcpu); 1864 if (cause == FSCR_EBB_LG) 1865 r = kvmppc_ebb_unavailable(vcpu); 1866 if (cause == FSCR_TM_LG) 1867 r = kvmppc_tm_unavailable(vcpu); 1868 } 1869 if (r == EMULATE_FAIL) { 1870 kvmppc_core_queue_program(vcpu, SRR1_PROGILL); 1871 r = RESUME_GUEST; 1872 } 1873 break; 1874 } 1875 1876 case BOOK3S_INTERRUPT_HV_RM_HARD: 1877 r = RESUME_PASSTHROUGH; 1878 break; 1879 default: 1880 kvmppc_dump_regs(vcpu); 1881 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n", 1882 vcpu->arch.trap, kvmppc_get_pc(vcpu), 1883 vcpu->arch.shregs.msr); 1884 run->hw.hardware_exit_reason = vcpu->arch.trap; 1885 r = RESUME_HOST; 1886 break; 1887 } 1888 1889 return r; 1890 } 1891 1892 static int kvmppc_handle_nested_exit(struct kvm_vcpu *vcpu) 1893 { 1894 int r; 1895 int srcu_idx; 1896 1897 vcpu->stat.sum_exits++; 1898 1899 /* 1900 * This can happen if an interrupt occurs in the last stages 1901 * of guest entry or the first stages of guest exit (i.e. after 1902 * setting paca->kvm_hstate.in_guest to KVM_GUEST_MODE_GUEST_HV 1903 * and before setting it to KVM_GUEST_MODE_HOST_HV). 1904 * That can happen due to a bug, or due to a machine check 1905 * occurring at just the wrong time. 1906 */ 1907 if (vcpu->arch.shregs.msr & MSR_HV) { 1908 pr_emerg("KVM trap in HV mode while nested!\n"); 1909 pr_emerg("trap=0x%x | pc=0x%lx | msr=0x%llx\n", 1910 vcpu->arch.trap, kvmppc_get_pc(vcpu), 1911 vcpu->arch.shregs.msr); 1912 kvmppc_dump_regs(vcpu); 1913 return RESUME_HOST; 1914 } 1915 switch (vcpu->arch.trap) { 1916 /* We're good on these - the host merely wanted to get our attention */ 1917 case BOOK3S_INTERRUPT_HV_DECREMENTER: 1918 vcpu->stat.dec_exits++; 1919 r = RESUME_GUEST; 1920 break; 1921 case BOOK3S_INTERRUPT_EXTERNAL: 1922 vcpu->stat.ext_intr_exits++; 1923 r = RESUME_HOST; 1924 break; 1925 case BOOK3S_INTERRUPT_H_DOORBELL: 1926 case BOOK3S_INTERRUPT_H_VIRT: 1927 vcpu->stat.ext_intr_exits++; 1928 r = RESUME_GUEST; 1929 break; 1930 /* These need to go to the nested HV */ 1931 case BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER: 1932 vcpu->arch.trap = BOOK3S_INTERRUPT_HV_DECREMENTER; 1933 vcpu->stat.dec_exits++; 1934 r = RESUME_HOST; 1935 break; 1936 /* SR/HMI/PMI are HV interrupts that host has handled. Resume guest.*/ 1937 case BOOK3S_INTERRUPT_HMI: 1938 case BOOK3S_INTERRUPT_PERFMON: 1939 case BOOK3S_INTERRUPT_SYSTEM_RESET: 1940 r = RESUME_GUEST; 1941 break; 1942 case BOOK3S_INTERRUPT_MACHINE_CHECK: 1943 { 1944 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL, 1945 DEFAULT_RATELIMIT_BURST); 1946 /* Pass the machine check to the L1 guest */ 1947 r = RESUME_HOST; 1948 /* Print the MCE event to host console. */ 1949 if (__ratelimit(&rs)) 1950 machine_check_print_event_info(&vcpu->arch.mce_evt, false, true); 1951 break; 1952 } 1953 /* 1954 * We get these next two if the guest accesses a page which it thinks 1955 * it has mapped but which is not actually present, either because 1956 * it is for an emulated I/O device or because the corresonding 1957 * host page has been paged out. 1958 */ 1959 case BOOK3S_INTERRUPT_H_DATA_STORAGE: 1960 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); 1961 r = kvmhv_nested_page_fault(vcpu); 1962 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx); 1963 break; 1964 case BOOK3S_INTERRUPT_H_INST_STORAGE: 1965 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu); 1966 vcpu->arch.fault_dsisr = kvmppc_get_msr(vcpu) & 1967 DSISR_SRR1_MATCH_64S; 1968 if (vcpu->arch.shregs.msr & HSRR1_HISI_WRITE) 1969 vcpu->arch.fault_dsisr |= DSISR_ISSTORE; 1970 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); 1971 r = kvmhv_nested_page_fault(vcpu); 1972 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx); 1973 break; 1974 1975 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 1976 case BOOK3S_INTERRUPT_HV_SOFTPATCH: 1977 /* 1978 * This occurs for various TM-related instructions that 1979 * we need to emulate on POWER9 DD2.2. We have already 1980 * handled the cases where the guest was in real-suspend 1981 * mode and was transitioning to transactional state. 1982 */ 1983 r = kvmhv_p9_tm_emulation(vcpu); 1984 if (r != -1) 1985 break; 1986 fallthrough; /* go to facility unavailable handler */ 1987 #endif 1988 1989 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: { 1990 u64 cause = vcpu->arch.hfscr >> 56; 1991 1992 /* 1993 * Only pass HFU interrupts to the L1 if the facility is 1994 * permitted but disabled by the L1's HFSCR, otherwise 1995 * the interrupt does not make sense to the L1 so turn 1996 * it into a HEAI. 1997 */ 1998 if (!(vcpu->arch.hfscr_permitted & (1UL << cause)) || 1999 (vcpu->arch.nested_hfscr & (1UL << cause))) { 2000 vcpu->arch.trap = BOOK3S_INTERRUPT_H_EMUL_ASSIST; 2001 2002 /* 2003 * If the fetch failed, return to guest and 2004 * try executing it again. 2005 */ 2006 r = kvmppc_get_last_inst(vcpu, INST_GENERIC, 2007 &vcpu->arch.emul_inst); 2008 if (r != EMULATE_DONE) 2009 r = RESUME_GUEST; 2010 else 2011 r = RESUME_HOST; 2012 } else { 2013 r = RESUME_HOST; 2014 } 2015 2016 break; 2017 } 2018 2019 case BOOK3S_INTERRUPT_HV_RM_HARD: 2020 vcpu->arch.trap = 0; 2021 r = RESUME_GUEST; 2022 if (!xics_on_xive()) 2023 kvmppc_xics_rm_complete(vcpu, 0); 2024 break; 2025 case BOOK3S_INTERRUPT_SYSCALL: 2026 { 2027 unsigned long req = kvmppc_get_gpr(vcpu, 3); 2028 2029 /* 2030 * The H_RPT_INVALIDATE hcalls issued by nested 2031 * guests for process-scoped invalidations when 2032 * GTSE=0, are handled here in L0. 2033 */ 2034 if (req == H_RPT_INVALIDATE) { 2035 r = kvmppc_nested_h_rpt_invalidate(vcpu); 2036 break; 2037 } 2038 2039 r = RESUME_HOST; 2040 break; 2041 } 2042 default: 2043 r = RESUME_HOST; 2044 break; 2045 } 2046 2047 return r; 2048 } 2049 2050 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu, 2051 struct kvm_sregs *sregs) 2052 { 2053 int i; 2054 2055 memset(sregs, 0, sizeof(struct kvm_sregs)); 2056 sregs->pvr = vcpu->arch.pvr; 2057 for (i = 0; i < vcpu->arch.slb_max; i++) { 2058 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige; 2059 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv; 2060 } 2061 2062 return 0; 2063 } 2064 2065 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu, 2066 struct kvm_sregs *sregs) 2067 { 2068 int i, j; 2069 2070 /* Only accept the same PVR as the host's, since we can't spoof it */ 2071 if (sregs->pvr != vcpu->arch.pvr) 2072 return -EINVAL; 2073 2074 j = 0; 2075 for (i = 0; i < vcpu->arch.slb_nr; i++) { 2076 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) { 2077 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe; 2078 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv; 2079 ++j; 2080 } 2081 } 2082 vcpu->arch.slb_max = j; 2083 2084 return 0; 2085 } 2086 2087 /* 2088 * Enforce limits on guest LPCR values based on hardware availability, 2089 * guest configuration, and possibly hypervisor support and security 2090 * concerns. 2091 */ 2092 unsigned long kvmppc_filter_lpcr_hv(struct kvm *kvm, unsigned long lpcr) 2093 { 2094 /* LPCR_TC only applies to HPT guests */ 2095 if (kvm_is_radix(kvm)) 2096 lpcr &= ~LPCR_TC; 2097 2098 /* On POWER8 and above, userspace can modify AIL */ 2099 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) 2100 lpcr &= ~LPCR_AIL; 2101 if ((lpcr & LPCR_AIL) != LPCR_AIL_3) 2102 lpcr &= ~LPCR_AIL; /* LPCR[AIL]=1/2 is disallowed */ 2103 /* 2104 * On some POWER9s we force AIL off for radix guests to prevent 2105 * executing in MSR[HV]=1 mode with the MMU enabled and PIDR set to 2106 * guest, which can result in Q0 translations with LPID=0 PID=PIDR to 2107 * be cached, which the host TLB management does not expect. 2108 */ 2109 if (kvm_is_radix(kvm) && cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG)) 2110 lpcr &= ~LPCR_AIL; 2111 2112 /* 2113 * On POWER9, allow userspace to enable large decrementer for the 2114 * guest, whether or not the host has it enabled. 2115 */ 2116 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 2117 lpcr &= ~LPCR_LD; 2118 2119 return lpcr; 2120 } 2121 2122 static void verify_lpcr(struct kvm *kvm, unsigned long lpcr) 2123 { 2124 if (lpcr != kvmppc_filter_lpcr_hv(kvm, lpcr)) { 2125 WARN_ONCE(1, "lpcr 0x%lx differs from filtered 0x%lx\n", 2126 lpcr, kvmppc_filter_lpcr_hv(kvm, lpcr)); 2127 } 2128 } 2129 2130 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr, 2131 bool preserve_top32) 2132 { 2133 struct kvm *kvm = vcpu->kvm; 2134 struct kvmppc_vcore *vc = vcpu->arch.vcore; 2135 u64 mask; 2136 2137 spin_lock(&vc->lock); 2138 2139 /* 2140 * Userspace can only modify 2141 * DPFD (default prefetch depth), ILE (interrupt little-endian), 2142 * TC (translation control), AIL (alternate interrupt location), 2143 * LD (large decrementer). 2144 * These are subject to restrictions from kvmppc_filter_lcpr_hv(). 2145 */ 2146 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC | LPCR_AIL | LPCR_LD; 2147 2148 /* Broken 32-bit version of LPCR must not clear top bits */ 2149 if (preserve_top32) 2150 mask &= 0xFFFFFFFF; 2151 2152 new_lpcr = kvmppc_filter_lpcr_hv(kvm, 2153 (vc->lpcr & ~mask) | (new_lpcr & mask)); 2154 2155 /* 2156 * If ILE (interrupt little-endian) has changed, update the 2157 * MSR_LE bit in the intr_msr for each vcpu in this vcore. 2158 */ 2159 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) { 2160 struct kvm_vcpu *vcpu; 2161 unsigned long i; 2162 2163 kvm_for_each_vcpu(i, vcpu, kvm) { 2164 if (vcpu->arch.vcore != vc) 2165 continue; 2166 if (new_lpcr & LPCR_ILE) 2167 vcpu->arch.intr_msr |= MSR_LE; 2168 else 2169 vcpu->arch.intr_msr &= ~MSR_LE; 2170 } 2171 } 2172 2173 vc->lpcr = new_lpcr; 2174 2175 spin_unlock(&vc->lock); 2176 } 2177 2178 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id, 2179 union kvmppc_one_reg *val) 2180 { 2181 int r = 0; 2182 long int i; 2183 2184 switch (id) { 2185 case KVM_REG_PPC_DEBUG_INST: 2186 *val = get_reg_val(id, KVMPPC_INST_SW_BREAKPOINT); 2187 break; 2188 case KVM_REG_PPC_HIOR: 2189 *val = get_reg_val(id, 0); 2190 break; 2191 case KVM_REG_PPC_DABR: 2192 *val = get_reg_val(id, vcpu->arch.dabr); 2193 break; 2194 case KVM_REG_PPC_DABRX: 2195 *val = get_reg_val(id, vcpu->arch.dabrx); 2196 break; 2197 case KVM_REG_PPC_DSCR: 2198 *val = get_reg_val(id, vcpu->arch.dscr); 2199 break; 2200 case KVM_REG_PPC_PURR: 2201 *val = get_reg_val(id, vcpu->arch.purr); 2202 break; 2203 case KVM_REG_PPC_SPURR: 2204 *val = get_reg_val(id, vcpu->arch.spurr); 2205 break; 2206 case KVM_REG_PPC_AMR: 2207 *val = get_reg_val(id, vcpu->arch.amr); 2208 break; 2209 case KVM_REG_PPC_UAMOR: 2210 *val = get_reg_val(id, vcpu->arch.uamor); 2211 break; 2212 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1: 2213 i = id - KVM_REG_PPC_MMCR0; 2214 *val = get_reg_val(id, vcpu->arch.mmcr[i]); 2215 break; 2216 case KVM_REG_PPC_MMCR2: 2217 *val = get_reg_val(id, vcpu->arch.mmcr[2]); 2218 break; 2219 case KVM_REG_PPC_MMCRA: 2220 *val = get_reg_val(id, vcpu->arch.mmcra); 2221 break; 2222 case KVM_REG_PPC_MMCRS: 2223 *val = get_reg_val(id, vcpu->arch.mmcrs); 2224 break; 2225 case KVM_REG_PPC_MMCR3: 2226 *val = get_reg_val(id, vcpu->arch.mmcr[3]); 2227 break; 2228 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8: 2229 i = id - KVM_REG_PPC_PMC1; 2230 *val = get_reg_val(id, vcpu->arch.pmc[i]); 2231 break; 2232 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2: 2233 i = id - KVM_REG_PPC_SPMC1; 2234 *val = get_reg_val(id, vcpu->arch.spmc[i]); 2235 break; 2236 case KVM_REG_PPC_SIAR: 2237 *val = get_reg_val(id, vcpu->arch.siar); 2238 break; 2239 case KVM_REG_PPC_SDAR: 2240 *val = get_reg_val(id, vcpu->arch.sdar); 2241 break; 2242 case KVM_REG_PPC_SIER: 2243 *val = get_reg_val(id, vcpu->arch.sier[0]); 2244 break; 2245 case KVM_REG_PPC_SIER2: 2246 *val = get_reg_val(id, vcpu->arch.sier[1]); 2247 break; 2248 case KVM_REG_PPC_SIER3: 2249 *val = get_reg_val(id, vcpu->arch.sier[2]); 2250 break; 2251 case KVM_REG_PPC_IAMR: 2252 *val = get_reg_val(id, vcpu->arch.iamr); 2253 break; 2254 case KVM_REG_PPC_PSPB: 2255 *val = get_reg_val(id, vcpu->arch.pspb); 2256 break; 2257 case KVM_REG_PPC_DPDES: 2258 /* 2259 * On POWER9, where we are emulating msgsndp etc., 2260 * we return 1 bit for each vcpu, which can come from 2261 * either vcore->dpdes or doorbell_request. 2262 * On POWER8, doorbell_request is 0. 2263 */ 2264 if (cpu_has_feature(CPU_FTR_ARCH_300)) 2265 *val = get_reg_val(id, vcpu->arch.doorbell_request); 2266 else 2267 *val = get_reg_val(id, vcpu->arch.vcore->dpdes); 2268 break; 2269 case KVM_REG_PPC_VTB: 2270 *val = get_reg_val(id, vcpu->arch.vcore->vtb); 2271 break; 2272 case KVM_REG_PPC_DAWR: 2273 *val = get_reg_val(id, vcpu->arch.dawr0); 2274 break; 2275 case KVM_REG_PPC_DAWRX: 2276 *val = get_reg_val(id, vcpu->arch.dawrx0); 2277 break; 2278 case KVM_REG_PPC_DAWR1: 2279 *val = get_reg_val(id, vcpu->arch.dawr1); 2280 break; 2281 case KVM_REG_PPC_DAWRX1: 2282 *val = get_reg_val(id, vcpu->arch.dawrx1); 2283 break; 2284 case KVM_REG_PPC_CIABR: 2285 *val = get_reg_val(id, vcpu->arch.ciabr); 2286 break; 2287 case KVM_REG_PPC_CSIGR: 2288 *val = get_reg_val(id, vcpu->arch.csigr); 2289 break; 2290 case KVM_REG_PPC_TACR: 2291 *val = get_reg_val(id, vcpu->arch.tacr); 2292 break; 2293 case KVM_REG_PPC_TCSCR: 2294 *val = get_reg_val(id, vcpu->arch.tcscr); 2295 break; 2296 case KVM_REG_PPC_PID: 2297 *val = get_reg_val(id, vcpu->arch.pid); 2298 break; 2299 case KVM_REG_PPC_ACOP: 2300 *val = get_reg_val(id, vcpu->arch.acop); 2301 break; 2302 case KVM_REG_PPC_WORT: 2303 *val = get_reg_val(id, vcpu->arch.wort); 2304 break; 2305 case KVM_REG_PPC_TIDR: 2306 *val = get_reg_val(id, vcpu->arch.tid); 2307 break; 2308 case KVM_REG_PPC_PSSCR: 2309 *val = get_reg_val(id, vcpu->arch.psscr); 2310 break; 2311 case KVM_REG_PPC_VPA_ADDR: 2312 spin_lock(&vcpu->arch.vpa_update_lock); 2313 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa); 2314 spin_unlock(&vcpu->arch.vpa_update_lock); 2315 break; 2316 case KVM_REG_PPC_VPA_SLB: 2317 spin_lock(&vcpu->arch.vpa_update_lock); 2318 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa; 2319 val->vpaval.length = vcpu->arch.slb_shadow.len; 2320 spin_unlock(&vcpu->arch.vpa_update_lock); 2321 break; 2322 case KVM_REG_PPC_VPA_DTL: 2323 spin_lock(&vcpu->arch.vpa_update_lock); 2324 val->vpaval.addr = vcpu->arch.dtl.next_gpa; 2325 val->vpaval.length = vcpu->arch.dtl.len; 2326 spin_unlock(&vcpu->arch.vpa_update_lock); 2327 break; 2328 case KVM_REG_PPC_TB_OFFSET: 2329 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset); 2330 break; 2331 case KVM_REG_PPC_LPCR: 2332 case KVM_REG_PPC_LPCR_64: 2333 *val = get_reg_val(id, vcpu->arch.vcore->lpcr); 2334 break; 2335 case KVM_REG_PPC_PPR: 2336 *val = get_reg_val(id, vcpu->arch.ppr); 2337 break; 2338 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 2339 case KVM_REG_PPC_TFHAR: 2340 *val = get_reg_val(id, vcpu->arch.tfhar); 2341 break; 2342 case KVM_REG_PPC_TFIAR: 2343 *val = get_reg_val(id, vcpu->arch.tfiar); 2344 break; 2345 case KVM_REG_PPC_TEXASR: 2346 *val = get_reg_val(id, vcpu->arch.texasr); 2347 break; 2348 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31: 2349 i = id - KVM_REG_PPC_TM_GPR0; 2350 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]); 2351 break; 2352 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63: 2353 { 2354 int j; 2355 i = id - KVM_REG_PPC_TM_VSR0; 2356 if (i < 32) 2357 for (j = 0; j < TS_FPRWIDTH; j++) 2358 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j]; 2359 else { 2360 if (cpu_has_feature(CPU_FTR_ALTIVEC)) 2361 val->vval = vcpu->arch.vr_tm.vr[i-32]; 2362 else 2363 r = -ENXIO; 2364 } 2365 break; 2366 } 2367 case KVM_REG_PPC_TM_CR: 2368 *val = get_reg_val(id, vcpu->arch.cr_tm); 2369 break; 2370 case KVM_REG_PPC_TM_XER: 2371 *val = get_reg_val(id, vcpu->arch.xer_tm); 2372 break; 2373 case KVM_REG_PPC_TM_LR: 2374 *val = get_reg_val(id, vcpu->arch.lr_tm); 2375 break; 2376 case KVM_REG_PPC_TM_CTR: 2377 *val = get_reg_val(id, vcpu->arch.ctr_tm); 2378 break; 2379 case KVM_REG_PPC_TM_FPSCR: 2380 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr); 2381 break; 2382 case KVM_REG_PPC_TM_AMR: 2383 *val = get_reg_val(id, vcpu->arch.amr_tm); 2384 break; 2385 case KVM_REG_PPC_TM_PPR: 2386 *val = get_reg_val(id, vcpu->arch.ppr_tm); 2387 break; 2388 case KVM_REG_PPC_TM_VRSAVE: 2389 *val = get_reg_val(id, vcpu->arch.vrsave_tm); 2390 break; 2391 case KVM_REG_PPC_TM_VSCR: 2392 if (cpu_has_feature(CPU_FTR_ALTIVEC)) 2393 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]); 2394 else 2395 r = -ENXIO; 2396 break; 2397 case KVM_REG_PPC_TM_DSCR: 2398 *val = get_reg_val(id, vcpu->arch.dscr_tm); 2399 break; 2400 case KVM_REG_PPC_TM_TAR: 2401 *val = get_reg_val(id, vcpu->arch.tar_tm); 2402 break; 2403 #endif 2404 case KVM_REG_PPC_ARCH_COMPAT: 2405 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat); 2406 break; 2407 case KVM_REG_PPC_DEC_EXPIRY: 2408 *val = get_reg_val(id, vcpu->arch.dec_expires); 2409 break; 2410 case KVM_REG_PPC_ONLINE: 2411 *val = get_reg_val(id, vcpu->arch.online); 2412 break; 2413 case KVM_REG_PPC_PTCR: 2414 *val = get_reg_val(id, vcpu->kvm->arch.l1_ptcr); 2415 break; 2416 default: 2417 r = -EINVAL; 2418 break; 2419 } 2420 2421 return r; 2422 } 2423 2424 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id, 2425 union kvmppc_one_reg *val) 2426 { 2427 int r = 0; 2428 long int i; 2429 unsigned long addr, len; 2430 2431 switch (id) { 2432 case KVM_REG_PPC_HIOR: 2433 /* Only allow this to be set to zero */ 2434 if (set_reg_val(id, *val)) 2435 r = -EINVAL; 2436 break; 2437 case KVM_REG_PPC_DABR: 2438 vcpu->arch.dabr = set_reg_val(id, *val); 2439 break; 2440 case KVM_REG_PPC_DABRX: 2441 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP; 2442 break; 2443 case KVM_REG_PPC_DSCR: 2444 vcpu->arch.dscr = set_reg_val(id, *val); 2445 break; 2446 case KVM_REG_PPC_PURR: 2447 vcpu->arch.purr = set_reg_val(id, *val); 2448 break; 2449 case KVM_REG_PPC_SPURR: 2450 vcpu->arch.spurr = set_reg_val(id, *val); 2451 break; 2452 case KVM_REG_PPC_AMR: 2453 vcpu->arch.amr = set_reg_val(id, *val); 2454 break; 2455 case KVM_REG_PPC_UAMOR: 2456 vcpu->arch.uamor = set_reg_val(id, *val); 2457 break; 2458 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCR1: 2459 i = id - KVM_REG_PPC_MMCR0; 2460 vcpu->arch.mmcr[i] = set_reg_val(id, *val); 2461 break; 2462 case KVM_REG_PPC_MMCR2: 2463 vcpu->arch.mmcr[2] = set_reg_val(id, *val); 2464 break; 2465 case KVM_REG_PPC_MMCRA: 2466 vcpu->arch.mmcra = set_reg_val(id, *val); 2467 break; 2468 case KVM_REG_PPC_MMCRS: 2469 vcpu->arch.mmcrs = set_reg_val(id, *val); 2470 break; 2471 case KVM_REG_PPC_MMCR3: 2472 *val = get_reg_val(id, vcpu->arch.mmcr[3]); 2473 break; 2474 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8: 2475 i = id - KVM_REG_PPC_PMC1; 2476 vcpu->arch.pmc[i] = set_reg_val(id, *val); 2477 break; 2478 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2: 2479 i = id - KVM_REG_PPC_SPMC1; 2480 vcpu->arch.spmc[i] = set_reg_val(id, *val); 2481 break; 2482 case KVM_REG_PPC_SIAR: 2483 vcpu->arch.siar = set_reg_val(id, *val); 2484 break; 2485 case KVM_REG_PPC_SDAR: 2486 vcpu->arch.sdar = set_reg_val(id, *val); 2487 break; 2488 case KVM_REG_PPC_SIER: 2489 vcpu->arch.sier[0] = set_reg_val(id, *val); 2490 break; 2491 case KVM_REG_PPC_SIER2: 2492 vcpu->arch.sier[1] = set_reg_val(id, *val); 2493 break; 2494 case KVM_REG_PPC_SIER3: 2495 vcpu->arch.sier[2] = set_reg_val(id, *val); 2496 break; 2497 case KVM_REG_PPC_IAMR: 2498 vcpu->arch.iamr = set_reg_val(id, *val); 2499 break; 2500 case KVM_REG_PPC_PSPB: 2501 vcpu->arch.pspb = set_reg_val(id, *val); 2502 break; 2503 case KVM_REG_PPC_DPDES: 2504 if (cpu_has_feature(CPU_FTR_ARCH_300)) 2505 vcpu->arch.doorbell_request = set_reg_val(id, *val) & 1; 2506 else 2507 vcpu->arch.vcore->dpdes = set_reg_val(id, *val); 2508 break; 2509 case KVM_REG_PPC_VTB: 2510 vcpu->arch.vcore->vtb = set_reg_val(id, *val); 2511 break; 2512 case KVM_REG_PPC_DAWR: 2513 vcpu->arch.dawr0 = set_reg_val(id, *val); 2514 break; 2515 case KVM_REG_PPC_DAWRX: 2516 vcpu->arch.dawrx0 = set_reg_val(id, *val) & ~DAWRX_HYP; 2517 break; 2518 case KVM_REG_PPC_DAWR1: 2519 vcpu->arch.dawr1 = set_reg_val(id, *val); 2520 break; 2521 case KVM_REG_PPC_DAWRX1: 2522 vcpu->arch.dawrx1 = set_reg_val(id, *val) & ~DAWRX_HYP; 2523 break; 2524 case KVM_REG_PPC_CIABR: 2525 vcpu->arch.ciabr = set_reg_val(id, *val); 2526 /* Don't allow setting breakpoints in hypervisor code */ 2527 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER) 2528 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */ 2529 break; 2530 case KVM_REG_PPC_CSIGR: 2531 vcpu->arch.csigr = set_reg_val(id, *val); 2532 break; 2533 case KVM_REG_PPC_TACR: 2534 vcpu->arch.tacr = set_reg_val(id, *val); 2535 break; 2536 case KVM_REG_PPC_TCSCR: 2537 vcpu->arch.tcscr = set_reg_val(id, *val); 2538 break; 2539 case KVM_REG_PPC_PID: 2540 vcpu->arch.pid = set_reg_val(id, *val); 2541 break; 2542 case KVM_REG_PPC_ACOP: 2543 vcpu->arch.acop = set_reg_val(id, *val); 2544 break; 2545 case KVM_REG_PPC_WORT: 2546 vcpu->arch.wort = set_reg_val(id, *val); 2547 break; 2548 case KVM_REG_PPC_TIDR: 2549 vcpu->arch.tid = set_reg_val(id, *val); 2550 break; 2551 case KVM_REG_PPC_PSSCR: 2552 vcpu->arch.psscr = set_reg_val(id, *val) & PSSCR_GUEST_VIS; 2553 break; 2554 case KVM_REG_PPC_VPA_ADDR: 2555 addr = set_reg_val(id, *val); 2556 r = -EINVAL; 2557 if (!addr && (vcpu->arch.slb_shadow.next_gpa || 2558 vcpu->arch.dtl.next_gpa)) 2559 break; 2560 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca)); 2561 break; 2562 case KVM_REG_PPC_VPA_SLB: 2563 addr = val->vpaval.addr; 2564 len = val->vpaval.length; 2565 r = -EINVAL; 2566 if (addr && !vcpu->arch.vpa.next_gpa) 2567 break; 2568 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len); 2569 break; 2570 case KVM_REG_PPC_VPA_DTL: 2571 addr = val->vpaval.addr; 2572 len = val->vpaval.length; 2573 r = -EINVAL; 2574 if (addr && (len < sizeof(struct dtl_entry) || 2575 !vcpu->arch.vpa.next_gpa)) 2576 break; 2577 len -= len % sizeof(struct dtl_entry); 2578 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len); 2579 break; 2580 case KVM_REG_PPC_TB_OFFSET: 2581 { 2582 /* round up to multiple of 2^24 */ 2583 u64 tb_offset = ALIGN(set_reg_val(id, *val), 1UL << 24); 2584 2585 /* 2586 * Now that we know the timebase offset, update the 2587 * decrementer expiry with a guest timebase value. If 2588 * the userspace does not set DEC_EXPIRY, this ensures 2589 * a migrated vcpu at least starts with an expired 2590 * decrementer, which is better than a large one that 2591 * causes a hang. 2592 */ 2593 if (!vcpu->arch.dec_expires && tb_offset) 2594 vcpu->arch.dec_expires = get_tb() + tb_offset; 2595 2596 vcpu->arch.vcore->tb_offset = tb_offset; 2597 break; 2598 } 2599 case KVM_REG_PPC_LPCR: 2600 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), true); 2601 break; 2602 case KVM_REG_PPC_LPCR_64: 2603 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val), false); 2604 break; 2605 case KVM_REG_PPC_PPR: 2606 vcpu->arch.ppr = set_reg_val(id, *val); 2607 break; 2608 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 2609 case KVM_REG_PPC_TFHAR: 2610 vcpu->arch.tfhar = set_reg_val(id, *val); 2611 break; 2612 case KVM_REG_PPC_TFIAR: 2613 vcpu->arch.tfiar = set_reg_val(id, *val); 2614 break; 2615 case KVM_REG_PPC_TEXASR: 2616 vcpu->arch.texasr = set_reg_val(id, *val); 2617 break; 2618 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31: 2619 i = id - KVM_REG_PPC_TM_GPR0; 2620 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val); 2621 break; 2622 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63: 2623 { 2624 int j; 2625 i = id - KVM_REG_PPC_TM_VSR0; 2626 if (i < 32) 2627 for (j = 0; j < TS_FPRWIDTH; j++) 2628 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j]; 2629 else 2630 if (cpu_has_feature(CPU_FTR_ALTIVEC)) 2631 vcpu->arch.vr_tm.vr[i-32] = val->vval; 2632 else 2633 r = -ENXIO; 2634 break; 2635 } 2636 case KVM_REG_PPC_TM_CR: 2637 vcpu->arch.cr_tm = set_reg_val(id, *val); 2638 break; 2639 case KVM_REG_PPC_TM_XER: 2640 vcpu->arch.xer_tm = set_reg_val(id, *val); 2641 break; 2642 case KVM_REG_PPC_TM_LR: 2643 vcpu->arch.lr_tm = set_reg_val(id, *val); 2644 break; 2645 case KVM_REG_PPC_TM_CTR: 2646 vcpu->arch.ctr_tm = set_reg_val(id, *val); 2647 break; 2648 case KVM_REG_PPC_TM_FPSCR: 2649 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val); 2650 break; 2651 case KVM_REG_PPC_TM_AMR: 2652 vcpu->arch.amr_tm = set_reg_val(id, *val); 2653 break; 2654 case KVM_REG_PPC_TM_PPR: 2655 vcpu->arch.ppr_tm = set_reg_val(id, *val); 2656 break; 2657 case KVM_REG_PPC_TM_VRSAVE: 2658 vcpu->arch.vrsave_tm = set_reg_val(id, *val); 2659 break; 2660 case KVM_REG_PPC_TM_VSCR: 2661 if (cpu_has_feature(CPU_FTR_ALTIVEC)) 2662 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val); 2663 else 2664 r = - ENXIO; 2665 break; 2666 case KVM_REG_PPC_TM_DSCR: 2667 vcpu->arch.dscr_tm = set_reg_val(id, *val); 2668 break; 2669 case KVM_REG_PPC_TM_TAR: 2670 vcpu->arch.tar_tm = set_reg_val(id, *val); 2671 break; 2672 #endif 2673 case KVM_REG_PPC_ARCH_COMPAT: 2674 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val)); 2675 break; 2676 case KVM_REG_PPC_DEC_EXPIRY: 2677 vcpu->arch.dec_expires = set_reg_val(id, *val); 2678 break; 2679 case KVM_REG_PPC_ONLINE: 2680 i = set_reg_val(id, *val); 2681 if (i && !vcpu->arch.online) 2682 atomic_inc(&vcpu->arch.vcore->online_count); 2683 else if (!i && vcpu->arch.online) 2684 atomic_dec(&vcpu->arch.vcore->online_count); 2685 vcpu->arch.online = i; 2686 break; 2687 case KVM_REG_PPC_PTCR: 2688 vcpu->kvm->arch.l1_ptcr = set_reg_val(id, *val); 2689 break; 2690 default: 2691 r = -EINVAL; 2692 break; 2693 } 2694 2695 return r; 2696 } 2697 2698 /* 2699 * On POWER9, threads are independent and can be in different partitions. 2700 * Therefore we consider each thread to be a subcore. 2701 * There is a restriction that all threads have to be in the same 2702 * MMU mode (radix or HPT), unfortunately, but since we only support 2703 * HPT guests on a HPT host so far, that isn't an impediment yet. 2704 */ 2705 static int threads_per_vcore(struct kvm *kvm) 2706 { 2707 if (cpu_has_feature(CPU_FTR_ARCH_300)) 2708 return 1; 2709 return threads_per_subcore; 2710 } 2711 2712 static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id) 2713 { 2714 struct kvmppc_vcore *vcore; 2715 2716 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL); 2717 2718 if (vcore == NULL) 2719 return NULL; 2720 2721 spin_lock_init(&vcore->lock); 2722 spin_lock_init(&vcore->stoltb_lock); 2723 rcuwait_init(&vcore->wait); 2724 vcore->preempt_tb = TB_NIL; 2725 vcore->lpcr = kvm->arch.lpcr; 2726 vcore->first_vcpuid = id; 2727 vcore->kvm = kvm; 2728 INIT_LIST_HEAD(&vcore->preempt_list); 2729 2730 return vcore; 2731 } 2732 2733 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING 2734 static struct debugfs_timings_element { 2735 const char *name; 2736 size_t offset; 2737 } timings[] = { 2738 #ifdef CONFIG_KVM_BOOK3S_HV_P9_TIMING 2739 {"vcpu_entry", offsetof(struct kvm_vcpu, arch.vcpu_entry)}, 2740 {"guest_entry", offsetof(struct kvm_vcpu, arch.guest_entry)}, 2741 {"in_guest", offsetof(struct kvm_vcpu, arch.in_guest)}, 2742 {"guest_exit", offsetof(struct kvm_vcpu, arch.guest_exit)}, 2743 {"vcpu_exit", offsetof(struct kvm_vcpu, arch.vcpu_exit)}, 2744 {"hypercall", offsetof(struct kvm_vcpu, arch.hcall)}, 2745 {"page_fault", offsetof(struct kvm_vcpu, arch.pg_fault)}, 2746 #else 2747 {"rm_entry", offsetof(struct kvm_vcpu, arch.rm_entry)}, 2748 {"rm_intr", offsetof(struct kvm_vcpu, arch.rm_intr)}, 2749 {"rm_exit", offsetof(struct kvm_vcpu, arch.rm_exit)}, 2750 {"guest", offsetof(struct kvm_vcpu, arch.guest_time)}, 2751 {"cede", offsetof(struct kvm_vcpu, arch.cede_time)}, 2752 #endif 2753 }; 2754 2755 #define N_TIMINGS (ARRAY_SIZE(timings)) 2756 2757 struct debugfs_timings_state { 2758 struct kvm_vcpu *vcpu; 2759 unsigned int buflen; 2760 char buf[N_TIMINGS * 100]; 2761 }; 2762 2763 static int debugfs_timings_open(struct inode *inode, struct file *file) 2764 { 2765 struct kvm_vcpu *vcpu = inode->i_private; 2766 struct debugfs_timings_state *p; 2767 2768 p = kzalloc(sizeof(*p), GFP_KERNEL); 2769 if (!p) 2770 return -ENOMEM; 2771 2772 kvm_get_kvm(vcpu->kvm); 2773 p->vcpu = vcpu; 2774 file->private_data = p; 2775 2776 return nonseekable_open(inode, file); 2777 } 2778 2779 static int debugfs_timings_release(struct inode *inode, struct file *file) 2780 { 2781 struct debugfs_timings_state *p = file->private_data; 2782 2783 kvm_put_kvm(p->vcpu->kvm); 2784 kfree(p); 2785 return 0; 2786 } 2787 2788 static ssize_t debugfs_timings_read(struct file *file, char __user *buf, 2789 size_t len, loff_t *ppos) 2790 { 2791 struct debugfs_timings_state *p = file->private_data; 2792 struct kvm_vcpu *vcpu = p->vcpu; 2793 char *s, *buf_end; 2794 struct kvmhv_tb_accumulator tb; 2795 u64 count; 2796 loff_t pos; 2797 ssize_t n; 2798 int i, loops; 2799 bool ok; 2800 2801 if (!p->buflen) { 2802 s = p->buf; 2803 buf_end = s + sizeof(p->buf); 2804 for (i = 0; i < N_TIMINGS; ++i) { 2805 struct kvmhv_tb_accumulator *acc; 2806 2807 acc = (struct kvmhv_tb_accumulator *) 2808 ((unsigned long)vcpu + timings[i].offset); 2809 ok = false; 2810 for (loops = 0; loops < 1000; ++loops) { 2811 count = acc->seqcount; 2812 if (!(count & 1)) { 2813 smp_rmb(); 2814 tb = *acc; 2815 smp_rmb(); 2816 if (count == acc->seqcount) { 2817 ok = true; 2818 break; 2819 } 2820 } 2821 udelay(1); 2822 } 2823 if (!ok) 2824 snprintf(s, buf_end - s, "%s: stuck\n", 2825 timings[i].name); 2826 else 2827 snprintf(s, buf_end - s, 2828 "%s: %llu %llu %llu %llu\n", 2829 timings[i].name, count / 2, 2830 tb_to_ns(tb.tb_total), 2831 tb_to_ns(tb.tb_min), 2832 tb_to_ns(tb.tb_max)); 2833 s += strlen(s); 2834 } 2835 p->buflen = s - p->buf; 2836 } 2837 2838 pos = *ppos; 2839 if (pos >= p->buflen) 2840 return 0; 2841 if (len > p->buflen - pos) 2842 len = p->buflen - pos; 2843 n = copy_to_user(buf, p->buf + pos, len); 2844 if (n) { 2845 if (n == len) 2846 return -EFAULT; 2847 len -= n; 2848 } 2849 *ppos = pos + len; 2850 return len; 2851 } 2852 2853 static ssize_t debugfs_timings_write(struct file *file, const char __user *buf, 2854 size_t len, loff_t *ppos) 2855 { 2856 return -EACCES; 2857 } 2858 2859 static const struct file_operations debugfs_timings_ops = { 2860 .owner = THIS_MODULE, 2861 .open = debugfs_timings_open, 2862 .release = debugfs_timings_release, 2863 .read = debugfs_timings_read, 2864 .write = debugfs_timings_write, 2865 .llseek = generic_file_llseek, 2866 }; 2867 2868 /* Create a debugfs directory for the vcpu */ 2869 static int kvmppc_arch_create_vcpu_debugfs_hv(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry) 2870 { 2871 if (cpu_has_feature(CPU_FTR_ARCH_300) == IS_ENABLED(CONFIG_KVM_BOOK3S_HV_P9_TIMING)) 2872 debugfs_create_file("timings", 0444, debugfs_dentry, vcpu, 2873 &debugfs_timings_ops); 2874 return 0; 2875 } 2876 2877 #else /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */ 2878 static int kvmppc_arch_create_vcpu_debugfs_hv(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry) 2879 { 2880 return 0; 2881 } 2882 #endif /* CONFIG_KVM_BOOK3S_HV_EXIT_TIMING */ 2883 2884 static int kvmppc_core_vcpu_create_hv(struct kvm_vcpu *vcpu) 2885 { 2886 int err; 2887 int core; 2888 struct kvmppc_vcore *vcore; 2889 struct kvm *kvm; 2890 unsigned int id; 2891 2892 kvm = vcpu->kvm; 2893 id = vcpu->vcpu_id; 2894 2895 vcpu->arch.shared = &vcpu->arch.shregs; 2896 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE 2897 /* 2898 * The shared struct is never shared on HV, 2899 * so we can always use host endianness 2900 */ 2901 #ifdef __BIG_ENDIAN__ 2902 vcpu->arch.shared_big_endian = true; 2903 #else 2904 vcpu->arch.shared_big_endian = false; 2905 #endif 2906 #endif 2907 vcpu->arch.mmcr[0] = MMCR0_FC; 2908 if (cpu_has_feature(CPU_FTR_ARCH_31)) { 2909 vcpu->arch.mmcr[0] |= MMCR0_PMCCEXT; 2910 vcpu->arch.mmcra = MMCRA_BHRB_DISABLE; 2911 } 2912 2913 vcpu->arch.ctrl = CTRL_RUNLATCH; 2914 /* default to host PVR, since we can't spoof it */ 2915 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR)); 2916 spin_lock_init(&vcpu->arch.vpa_update_lock); 2917 spin_lock_init(&vcpu->arch.tbacct_lock); 2918 vcpu->arch.busy_preempt = TB_NIL; 2919 vcpu->arch.shregs.msr = MSR_ME; 2920 vcpu->arch.intr_msr = MSR_SF | MSR_ME; 2921 2922 /* 2923 * Set the default HFSCR for the guest from the host value. 2924 * This value is only used on POWER9. 2925 * On POWER9, we want to virtualize the doorbell facility, so we 2926 * don't set the HFSCR_MSGP bit, and that causes those instructions 2927 * to trap and then we emulate them. 2928 */ 2929 vcpu->arch.hfscr = HFSCR_TAR | HFSCR_EBB | HFSCR_PM | HFSCR_BHRB | 2930 HFSCR_DSCR | HFSCR_VECVSX | HFSCR_FP; 2931 if (cpu_has_feature(CPU_FTR_HVMODE)) { 2932 vcpu->arch.hfscr &= mfspr(SPRN_HFSCR); 2933 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 2934 if (cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) 2935 vcpu->arch.hfscr |= HFSCR_TM; 2936 #endif 2937 } 2938 if (cpu_has_feature(CPU_FTR_TM_COMP)) 2939 vcpu->arch.hfscr |= HFSCR_TM; 2940 2941 vcpu->arch.hfscr_permitted = vcpu->arch.hfscr; 2942 2943 /* 2944 * PM, EBB, TM are demand-faulted so start with it clear. 2945 */ 2946 vcpu->arch.hfscr &= ~(HFSCR_PM | HFSCR_EBB | HFSCR_TM); 2947 2948 kvmppc_mmu_book3s_hv_init(vcpu); 2949 2950 vcpu->arch.state = KVMPPC_VCPU_NOTREADY; 2951 2952 init_waitqueue_head(&vcpu->arch.cpu_run); 2953 2954 mutex_lock(&kvm->lock); 2955 vcore = NULL; 2956 err = -EINVAL; 2957 if (cpu_has_feature(CPU_FTR_ARCH_300)) { 2958 if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode)) { 2959 pr_devel("KVM: VCPU ID too high\n"); 2960 core = KVM_MAX_VCORES; 2961 } else { 2962 BUG_ON(kvm->arch.smt_mode != 1); 2963 core = kvmppc_pack_vcpu_id(kvm, id); 2964 } 2965 } else { 2966 core = id / kvm->arch.smt_mode; 2967 } 2968 if (core < KVM_MAX_VCORES) { 2969 vcore = kvm->arch.vcores[core]; 2970 if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) { 2971 pr_devel("KVM: collision on id %u", id); 2972 vcore = NULL; 2973 } else if (!vcore) { 2974 /* 2975 * Take mmu_setup_lock for mutual exclusion 2976 * with kvmppc_update_lpcr(). 2977 */ 2978 err = -ENOMEM; 2979 vcore = kvmppc_vcore_create(kvm, 2980 id & ~(kvm->arch.smt_mode - 1)); 2981 mutex_lock(&kvm->arch.mmu_setup_lock); 2982 kvm->arch.vcores[core] = vcore; 2983 kvm->arch.online_vcores++; 2984 mutex_unlock(&kvm->arch.mmu_setup_lock); 2985 } 2986 } 2987 mutex_unlock(&kvm->lock); 2988 2989 if (!vcore) 2990 return err; 2991 2992 spin_lock(&vcore->lock); 2993 ++vcore->num_threads; 2994 spin_unlock(&vcore->lock); 2995 vcpu->arch.vcore = vcore; 2996 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid; 2997 vcpu->arch.thread_cpu = -1; 2998 vcpu->arch.prev_cpu = -1; 2999 3000 vcpu->arch.cpu_type = KVM_CPU_3S_64; 3001 kvmppc_sanity_check(vcpu); 3002 3003 return 0; 3004 } 3005 3006 static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode, 3007 unsigned long flags) 3008 { 3009 int err; 3010 int esmt = 0; 3011 3012 if (flags) 3013 return -EINVAL; 3014 if (smt_mode > MAX_SMT_THREADS || !is_power_of_2(smt_mode)) 3015 return -EINVAL; 3016 if (!cpu_has_feature(CPU_FTR_ARCH_300)) { 3017 /* 3018 * On POWER8 (or POWER7), the threading mode is "strict", 3019 * so we pack smt_mode vcpus per vcore. 3020 */ 3021 if (smt_mode > threads_per_subcore) 3022 return -EINVAL; 3023 } else { 3024 /* 3025 * On POWER9, the threading mode is "loose", 3026 * so each vcpu gets its own vcore. 3027 */ 3028 esmt = smt_mode; 3029 smt_mode = 1; 3030 } 3031 mutex_lock(&kvm->lock); 3032 err = -EBUSY; 3033 if (!kvm->arch.online_vcores) { 3034 kvm->arch.smt_mode = smt_mode; 3035 kvm->arch.emul_smt_mode = esmt; 3036 err = 0; 3037 } 3038 mutex_unlock(&kvm->lock); 3039 3040 return err; 3041 } 3042 3043 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa) 3044 { 3045 if (vpa->pinned_addr) 3046 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa, 3047 vpa->dirty); 3048 } 3049 3050 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu) 3051 { 3052 spin_lock(&vcpu->arch.vpa_update_lock); 3053 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl); 3054 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow); 3055 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa); 3056 spin_unlock(&vcpu->arch.vpa_update_lock); 3057 } 3058 3059 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu) 3060 { 3061 /* Indicate we want to get back into the guest */ 3062 return 1; 3063 } 3064 3065 static void kvmppc_set_timer(struct kvm_vcpu *vcpu) 3066 { 3067 unsigned long dec_nsec, now; 3068 3069 now = get_tb(); 3070 if (now > kvmppc_dec_expires_host_tb(vcpu)) { 3071 /* decrementer has already gone negative */ 3072 kvmppc_core_queue_dec(vcpu); 3073 kvmppc_core_prepare_to_enter(vcpu); 3074 return; 3075 } 3076 dec_nsec = tb_to_ns(kvmppc_dec_expires_host_tb(vcpu) - now); 3077 hrtimer_start(&vcpu->arch.dec_timer, dec_nsec, HRTIMER_MODE_REL); 3078 vcpu->arch.timer_running = 1; 3079 } 3080 3081 extern int __kvmppc_vcore_entry(void); 3082 3083 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc, 3084 struct kvm_vcpu *vcpu, u64 tb) 3085 { 3086 u64 now; 3087 3088 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE) 3089 return; 3090 spin_lock_irq(&vcpu->arch.tbacct_lock); 3091 now = tb; 3092 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) - 3093 vcpu->arch.stolen_logged; 3094 vcpu->arch.busy_preempt = now; 3095 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST; 3096 spin_unlock_irq(&vcpu->arch.tbacct_lock); 3097 --vc->n_runnable; 3098 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], NULL); 3099 } 3100 3101 static int kvmppc_grab_hwthread(int cpu) 3102 { 3103 struct paca_struct *tpaca; 3104 long timeout = 10000; 3105 3106 tpaca = paca_ptrs[cpu]; 3107 3108 /* Ensure the thread won't go into the kernel if it wakes */ 3109 tpaca->kvm_hstate.kvm_vcpu = NULL; 3110 tpaca->kvm_hstate.kvm_vcore = NULL; 3111 tpaca->kvm_hstate.napping = 0; 3112 smp_wmb(); 3113 tpaca->kvm_hstate.hwthread_req = 1; 3114 3115 /* 3116 * If the thread is already executing in the kernel (e.g. handling 3117 * a stray interrupt), wait for it to get back to nap mode. 3118 * The smp_mb() is to ensure that our setting of hwthread_req 3119 * is visible before we look at hwthread_state, so if this 3120 * races with the code at system_reset_pSeries and the thread 3121 * misses our setting of hwthread_req, we are sure to see its 3122 * setting of hwthread_state, and vice versa. 3123 */ 3124 smp_mb(); 3125 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) { 3126 if (--timeout <= 0) { 3127 pr_err("KVM: couldn't grab cpu %d\n", cpu); 3128 return -EBUSY; 3129 } 3130 udelay(1); 3131 } 3132 return 0; 3133 } 3134 3135 static void kvmppc_release_hwthread(int cpu) 3136 { 3137 struct paca_struct *tpaca; 3138 3139 tpaca = paca_ptrs[cpu]; 3140 tpaca->kvm_hstate.hwthread_req = 0; 3141 tpaca->kvm_hstate.kvm_vcpu = NULL; 3142 tpaca->kvm_hstate.kvm_vcore = NULL; 3143 tpaca->kvm_hstate.kvm_split_mode = NULL; 3144 } 3145 3146 static DEFINE_PER_CPU(struct kvm *, cpu_in_guest); 3147 3148 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu) 3149 { 3150 struct kvm_nested_guest *nested = vcpu->arch.nested; 3151 cpumask_t *need_tlb_flush; 3152 int i; 3153 3154 if (nested) 3155 need_tlb_flush = &nested->need_tlb_flush; 3156 else 3157 need_tlb_flush = &kvm->arch.need_tlb_flush; 3158 3159 cpu = cpu_first_tlb_thread_sibling(cpu); 3160 for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu); 3161 i += cpu_tlb_thread_sibling_step()) 3162 cpumask_set_cpu(i, need_tlb_flush); 3163 3164 /* 3165 * Make sure setting of bit in need_tlb_flush precedes testing of 3166 * cpu_in_guest. The matching barrier on the other side is hwsync 3167 * when switching to guest MMU mode, which happens between 3168 * cpu_in_guest being set to the guest kvm, and need_tlb_flush bit 3169 * being tested. 3170 */ 3171 smp_mb(); 3172 3173 for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu); 3174 i += cpu_tlb_thread_sibling_step()) { 3175 struct kvm *running = *per_cpu_ptr(&cpu_in_guest, i); 3176 3177 if (running == kvm) 3178 smp_call_function_single(i, do_nothing, NULL, 1); 3179 } 3180 } 3181 3182 static void do_migrate_away_vcpu(void *arg) 3183 { 3184 struct kvm_vcpu *vcpu = arg; 3185 struct kvm *kvm = vcpu->kvm; 3186 3187 /* 3188 * If the guest has GTSE, it may execute tlbie, so do a eieio; tlbsync; 3189 * ptesync sequence on the old CPU before migrating to a new one, in 3190 * case we interrupted the guest between a tlbie ; eieio ; 3191 * tlbsync; ptesync sequence. 3192 * 3193 * Otherwise, ptesync is sufficient for ordering tlbiel sequences. 3194 */ 3195 if (kvm->arch.lpcr & LPCR_GTSE) 3196 asm volatile("eieio; tlbsync; ptesync"); 3197 else 3198 asm volatile("ptesync"); 3199 } 3200 3201 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu) 3202 { 3203 struct kvm_nested_guest *nested = vcpu->arch.nested; 3204 struct kvm *kvm = vcpu->kvm; 3205 int prev_cpu; 3206 3207 if (!cpu_has_feature(CPU_FTR_HVMODE)) 3208 return; 3209 3210 if (nested) 3211 prev_cpu = nested->prev_cpu[vcpu->arch.nested_vcpu_id]; 3212 else 3213 prev_cpu = vcpu->arch.prev_cpu; 3214 3215 /* 3216 * With radix, the guest can do TLB invalidations itself, 3217 * and it could choose to use the local form (tlbiel) if 3218 * it is invalidating a translation that has only ever been 3219 * used on one vcpu. However, that doesn't mean it has 3220 * only ever been used on one physical cpu, since vcpus 3221 * can move around between pcpus. To cope with this, when 3222 * a vcpu moves from one pcpu to another, we need to tell 3223 * any vcpus running on the same core as this vcpu previously 3224 * ran to flush the TLB. 3225 */ 3226 if (prev_cpu != pcpu) { 3227 if (prev_cpu >= 0) { 3228 if (cpu_first_tlb_thread_sibling(prev_cpu) != 3229 cpu_first_tlb_thread_sibling(pcpu)) 3230 radix_flush_cpu(kvm, prev_cpu, vcpu); 3231 3232 smp_call_function_single(prev_cpu, 3233 do_migrate_away_vcpu, vcpu, 1); 3234 } 3235 if (nested) 3236 nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu; 3237 else 3238 vcpu->arch.prev_cpu = pcpu; 3239 } 3240 } 3241 3242 static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc) 3243 { 3244 int cpu; 3245 struct paca_struct *tpaca; 3246 3247 cpu = vc->pcpu; 3248 if (vcpu) { 3249 if (vcpu->arch.timer_running) { 3250 hrtimer_try_to_cancel(&vcpu->arch.dec_timer); 3251 vcpu->arch.timer_running = 0; 3252 } 3253 cpu += vcpu->arch.ptid; 3254 vcpu->cpu = vc->pcpu; 3255 vcpu->arch.thread_cpu = cpu; 3256 } 3257 tpaca = paca_ptrs[cpu]; 3258 tpaca->kvm_hstate.kvm_vcpu = vcpu; 3259 tpaca->kvm_hstate.ptid = cpu - vc->pcpu; 3260 tpaca->kvm_hstate.fake_suspend = 0; 3261 /* Order stores to hstate.kvm_vcpu etc. before store to kvm_vcore */ 3262 smp_wmb(); 3263 tpaca->kvm_hstate.kvm_vcore = vc; 3264 if (cpu != smp_processor_id()) 3265 kvmppc_ipi_thread(cpu); 3266 } 3267 3268 static void kvmppc_wait_for_nap(int n_threads) 3269 { 3270 int cpu = smp_processor_id(); 3271 int i, loops; 3272 3273 if (n_threads <= 1) 3274 return; 3275 for (loops = 0; loops < 1000000; ++loops) { 3276 /* 3277 * Check if all threads are finished. 3278 * We set the vcore pointer when starting a thread 3279 * and the thread clears it when finished, so we look 3280 * for any threads that still have a non-NULL vcore ptr. 3281 */ 3282 for (i = 1; i < n_threads; ++i) 3283 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore) 3284 break; 3285 if (i == n_threads) { 3286 HMT_medium(); 3287 return; 3288 } 3289 HMT_low(); 3290 } 3291 HMT_medium(); 3292 for (i = 1; i < n_threads; ++i) 3293 if (paca_ptrs[cpu + i]->kvm_hstate.kvm_vcore) 3294 pr_err("KVM: CPU %d seems to be stuck\n", cpu + i); 3295 } 3296 3297 /* 3298 * Check that we are on thread 0 and that any other threads in 3299 * this core are off-line. Then grab the threads so they can't 3300 * enter the kernel. 3301 */ 3302 static int on_primary_thread(void) 3303 { 3304 int cpu = smp_processor_id(); 3305 int thr; 3306 3307 /* Are we on a primary subcore? */ 3308 if (cpu_thread_in_subcore(cpu)) 3309 return 0; 3310 3311 thr = 0; 3312 while (++thr < threads_per_subcore) 3313 if (cpu_online(cpu + thr)) 3314 return 0; 3315 3316 /* Grab all hw threads so they can't go into the kernel */ 3317 for (thr = 1; thr < threads_per_subcore; ++thr) { 3318 if (kvmppc_grab_hwthread(cpu + thr)) { 3319 /* Couldn't grab one; let the others go */ 3320 do { 3321 kvmppc_release_hwthread(cpu + thr); 3322 } while (--thr > 0); 3323 return 0; 3324 } 3325 } 3326 return 1; 3327 } 3328 3329 /* 3330 * A list of virtual cores for each physical CPU. 3331 * These are vcores that could run but their runner VCPU tasks are 3332 * (or may be) preempted. 3333 */ 3334 struct preempted_vcore_list { 3335 struct list_head list; 3336 spinlock_t lock; 3337 }; 3338 3339 static DEFINE_PER_CPU(struct preempted_vcore_list, preempted_vcores); 3340 3341 static void init_vcore_lists(void) 3342 { 3343 int cpu; 3344 3345 for_each_possible_cpu(cpu) { 3346 struct preempted_vcore_list *lp = &per_cpu(preempted_vcores, cpu); 3347 spin_lock_init(&lp->lock); 3348 INIT_LIST_HEAD(&lp->list); 3349 } 3350 } 3351 3352 static void kvmppc_vcore_preempt(struct kvmppc_vcore *vc) 3353 { 3354 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores); 3355 3356 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)); 3357 3358 vc->vcore_state = VCORE_PREEMPT; 3359 vc->pcpu = smp_processor_id(); 3360 if (vc->num_threads < threads_per_vcore(vc->kvm)) { 3361 spin_lock(&lp->lock); 3362 list_add_tail(&vc->preempt_list, &lp->list); 3363 spin_unlock(&lp->lock); 3364 } 3365 3366 /* Start accumulating stolen time */ 3367 kvmppc_core_start_stolen(vc, mftb()); 3368 } 3369 3370 static void kvmppc_vcore_end_preempt(struct kvmppc_vcore *vc) 3371 { 3372 struct preempted_vcore_list *lp; 3373 3374 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)); 3375 3376 kvmppc_core_end_stolen(vc, mftb()); 3377 if (!list_empty(&vc->preempt_list)) { 3378 lp = &per_cpu(preempted_vcores, vc->pcpu); 3379 spin_lock(&lp->lock); 3380 list_del_init(&vc->preempt_list); 3381 spin_unlock(&lp->lock); 3382 } 3383 vc->vcore_state = VCORE_INACTIVE; 3384 } 3385 3386 /* 3387 * This stores information about the virtual cores currently 3388 * assigned to a physical core. 3389 */ 3390 struct core_info { 3391 int n_subcores; 3392 int max_subcore_threads; 3393 int total_threads; 3394 int subcore_threads[MAX_SUBCORES]; 3395 struct kvmppc_vcore *vc[MAX_SUBCORES]; 3396 }; 3397 3398 /* 3399 * This mapping means subcores 0 and 1 can use threads 0-3 and 4-7 3400 * respectively in 2-way micro-threading (split-core) mode on POWER8. 3401 */ 3402 static int subcore_thread_map[MAX_SUBCORES] = { 0, 4, 2, 6 }; 3403 3404 static void init_core_info(struct core_info *cip, struct kvmppc_vcore *vc) 3405 { 3406 memset(cip, 0, sizeof(*cip)); 3407 cip->n_subcores = 1; 3408 cip->max_subcore_threads = vc->num_threads; 3409 cip->total_threads = vc->num_threads; 3410 cip->subcore_threads[0] = vc->num_threads; 3411 cip->vc[0] = vc; 3412 } 3413 3414 static bool subcore_config_ok(int n_subcores, int n_threads) 3415 { 3416 /* 3417 * POWER9 "SMT4" cores are permanently in what is effectively a 4-way 3418 * split-core mode, with one thread per subcore. 3419 */ 3420 if (cpu_has_feature(CPU_FTR_ARCH_300)) 3421 return n_subcores <= 4 && n_threads == 1; 3422 3423 /* On POWER8, can only dynamically split if unsplit to begin with */ 3424 if (n_subcores > 1 && threads_per_subcore < MAX_SMT_THREADS) 3425 return false; 3426 if (n_subcores > MAX_SUBCORES) 3427 return false; 3428 if (n_subcores > 1) { 3429 if (!(dynamic_mt_modes & 2)) 3430 n_subcores = 4; 3431 if (n_subcores > 2 && !(dynamic_mt_modes & 4)) 3432 return false; 3433 } 3434 3435 return n_subcores * roundup_pow_of_two(n_threads) <= MAX_SMT_THREADS; 3436 } 3437 3438 static void init_vcore_to_run(struct kvmppc_vcore *vc) 3439 { 3440 vc->entry_exit_map = 0; 3441 vc->in_guest = 0; 3442 vc->napping_threads = 0; 3443 vc->conferring_threads = 0; 3444 vc->tb_offset_applied = 0; 3445 } 3446 3447 static bool can_dynamic_split(struct kvmppc_vcore *vc, struct core_info *cip) 3448 { 3449 int n_threads = vc->num_threads; 3450 int sub; 3451 3452 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) 3453 return false; 3454 3455 /* In one_vm_per_core mode, require all vcores to be from the same vm */ 3456 if (one_vm_per_core && vc->kvm != cip->vc[0]->kvm) 3457 return false; 3458 3459 if (n_threads < cip->max_subcore_threads) 3460 n_threads = cip->max_subcore_threads; 3461 if (!subcore_config_ok(cip->n_subcores + 1, n_threads)) 3462 return false; 3463 cip->max_subcore_threads = n_threads; 3464 3465 sub = cip->n_subcores; 3466 ++cip->n_subcores; 3467 cip->total_threads += vc->num_threads; 3468 cip->subcore_threads[sub] = vc->num_threads; 3469 cip->vc[sub] = vc; 3470 init_vcore_to_run(vc); 3471 list_del_init(&vc->preempt_list); 3472 3473 return true; 3474 } 3475 3476 /* 3477 * Work out whether it is possible to piggyback the execution of 3478 * vcore *pvc onto the execution of the other vcores described in *cip. 3479 */ 3480 static bool can_piggyback(struct kvmppc_vcore *pvc, struct core_info *cip, 3481 int target_threads) 3482 { 3483 if (cip->total_threads + pvc->num_threads > target_threads) 3484 return false; 3485 3486 return can_dynamic_split(pvc, cip); 3487 } 3488 3489 static void prepare_threads(struct kvmppc_vcore *vc) 3490 { 3491 int i; 3492 struct kvm_vcpu *vcpu; 3493 3494 for_each_runnable_thread(i, vcpu, vc) { 3495 if (signal_pending(vcpu->arch.run_task)) 3496 vcpu->arch.ret = -EINTR; 3497 else if (vcpu->arch.vpa.update_pending || 3498 vcpu->arch.slb_shadow.update_pending || 3499 vcpu->arch.dtl.update_pending) 3500 vcpu->arch.ret = RESUME_GUEST; 3501 else 3502 continue; 3503 kvmppc_remove_runnable(vc, vcpu, mftb()); 3504 wake_up(&vcpu->arch.cpu_run); 3505 } 3506 } 3507 3508 static void collect_piggybacks(struct core_info *cip, int target_threads) 3509 { 3510 struct preempted_vcore_list *lp = this_cpu_ptr(&preempted_vcores); 3511 struct kvmppc_vcore *pvc, *vcnext; 3512 3513 spin_lock(&lp->lock); 3514 list_for_each_entry_safe(pvc, vcnext, &lp->list, preempt_list) { 3515 if (!spin_trylock(&pvc->lock)) 3516 continue; 3517 prepare_threads(pvc); 3518 if (!pvc->n_runnable || !pvc->kvm->arch.mmu_ready) { 3519 list_del_init(&pvc->preempt_list); 3520 if (pvc->runner == NULL) { 3521 pvc->vcore_state = VCORE_INACTIVE; 3522 kvmppc_core_end_stolen(pvc, mftb()); 3523 } 3524 spin_unlock(&pvc->lock); 3525 continue; 3526 } 3527 if (!can_piggyback(pvc, cip, target_threads)) { 3528 spin_unlock(&pvc->lock); 3529 continue; 3530 } 3531 kvmppc_core_end_stolen(pvc, mftb()); 3532 pvc->vcore_state = VCORE_PIGGYBACK; 3533 if (cip->total_threads >= target_threads) 3534 break; 3535 } 3536 spin_unlock(&lp->lock); 3537 } 3538 3539 static bool recheck_signals_and_mmu(struct core_info *cip) 3540 { 3541 int sub, i; 3542 struct kvm_vcpu *vcpu; 3543 struct kvmppc_vcore *vc; 3544 3545 for (sub = 0; sub < cip->n_subcores; ++sub) { 3546 vc = cip->vc[sub]; 3547 if (!vc->kvm->arch.mmu_ready) 3548 return true; 3549 for_each_runnable_thread(i, vcpu, vc) 3550 if (signal_pending(vcpu->arch.run_task)) 3551 return true; 3552 } 3553 return false; 3554 } 3555 3556 static void post_guest_process(struct kvmppc_vcore *vc, bool is_master) 3557 { 3558 int still_running = 0, i; 3559 u64 now; 3560 long ret; 3561 struct kvm_vcpu *vcpu; 3562 3563 spin_lock(&vc->lock); 3564 now = get_tb(); 3565 for_each_runnable_thread(i, vcpu, vc) { 3566 /* 3567 * It's safe to unlock the vcore in the loop here, because 3568 * for_each_runnable_thread() is safe against removal of 3569 * the vcpu, and the vcore state is VCORE_EXITING here, 3570 * so any vcpus becoming runnable will have their arch.trap 3571 * set to zero and can't actually run in the guest. 3572 */ 3573 spin_unlock(&vc->lock); 3574 /* cancel pending dec exception if dec is positive */ 3575 if (now < kvmppc_dec_expires_host_tb(vcpu) && 3576 kvmppc_core_pending_dec(vcpu)) 3577 kvmppc_core_dequeue_dec(vcpu); 3578 3579 trace_kvm_guest_exit(vcpu); 3580 3581 ret = RESUME_GUEST; 3582 if (vcpu->arch.trap) 3583 ret = kvmppc_handle_exit_hv(vcpu, 3584 vcpu->arch.run_task); 3585 3586 vcpu->arch.ret = ret; 3587 vcpu->arch.trap = 0; 3588 3589 spin_lock(&vc->lock); 3590 if (is_kvmppc_resume_guest(vcpu->arch.ret)) { 3591 if (vcpu->arch.pending_exceptions) 3592 kvmppc_core_prepare_to_enter(vcpu); 3593 if (vcpu->arch.ceded) 3594 kvmppc_set_timer(vcpu); 3595 else 3596 ++still_running; 3597 } else { 3598 kvmppc_remove_runnable(vc, vcpu, mftb()); 3599 wake_up(&vcpu->arch.cpu_run); 3600 } 3601 } 3602 if (!is_master) { 3603 if (still_running > 0) { 3604 kvmppc_vcore_preempt(vc); 3605 } else if (vc->runner) { 3606 vc->vcore_state = VCORE_PREEMPT; 3607 kvmppc_core_start_stolen(vc, mftb()); 3608 } else { 3609 vc->vcore_state = VCORE_INACTIVE; 3610 } 3611 if (vc->n_runnable > 0 && vc->runner == NULL) { 3612 /* make sure there's a candidate runner awake */ 3613 i = -1; 3614 vcpu = next_runnable_thread(vc, &i); 3615 wake_up(&vcpu->arch.cpu_run); 3616 } 3617 } 3618 spin_unlock(&vc->lock); 3619 } 3620 3621 /* 3622 * Clear core from the list of active host cores as we are about to 3623 * enter the guest. Only do this if it is the primary thread of the 3624 * core (not if a subcore) that is entering the guest. 3625 */ 3626 static inline int kvmppc_clear_host_core(unsigned int cpu) 3627 { 3628 int core; 3629 3630 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu)) 3631 return 0; 3632 /* 3633 * Memory barrier can be omitted here as we will do a smp_wmb() 3634 * later in kvmppc_start_thread and we need ensure that state is 3635 * visible to other CPUs only after we enter guest. 3636 */ 3637 core = cpu >> threads_shift; 3638 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 0; 3639 return 0; 3640 } 3641 3642 /* 3643 * Advertise this core as an active host core since we exited the guest 3644 * Only need to do this if it is the primary thread of the core that is 3645 * exiting. 3646 */ 3647 static inline int kvmppc_set_host_core(unsigned int cpu) 3648 { 3649 int core; 3650 3651 if (!kvmppc_host_rm_ops_hv || cpu_thread_in_core(cpu)) 3652 return 0; 3653 3654 /* 3655 * Memory barrier can be omitted here because we do a spin_unlock 3656 * immediately after this which provides the memory barrier. 3657 */ 3658 core = cpu >> threads_shift; 3659 kvmppc_host_rm_ops_hv->rm_core[core].rm_state.in_host = 1; 3660 return 0; 3661 } 3662 3663 static void set_irq_happened(int trap) 3664 { 3665 switch (trap) { 3666 case BOOK3S_INTERRUPT_EXTERNAL: 3667 local_paca->irq_happened |= PACA_IRQ_EE; 3668 break; 3669 case BOOK3S_INTERRUPT_H_DOORBELL: 3670 local_paca->irq_happened |= PACA_IRQ_DBELL; 3671 break; 3672 case BOOK3S_INTERRUPT_HMI: 3673 local_paca->irq_happened |= PACA_IRQ_HMI; 3674 break; 3675 case BOOK3S_INTERRUPT_SYSTEM_RESET: 3676 replay_system_reset(); 3677 break; 3678 } 3679 } 3680 3681 /* 3682 * Run a set of guest threads on a physical core. 3683 * Called with vc->lock held. 3684 */ 3685 static noinline void kvmppc_run_core(struct kvmppc_vcore *vc) 3686 { 3687 struct kvm_vcpu *vcpu; 3688 int i; 3689 int srcu_idx; 3690 struct core_info core_info; 3691 struct kvmppc_vcore *pvc; 3692 struct kvm_split_mode split_info, *sip; 3693 int split, subcore_size, active; 3694 int sub; 3695 bool thr0_done; 3696 unsigned long cmd_bit, stat_bit; 3697 int pcpu, thr; 3698 int target_threads; 3699 int controlled_threads; 3700 int trap; 3701 bool is_power8; 3702 3703 if (WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300))) 3704 return; 3705 3706 /* 3707 * Remove from the list any threads that have a signal pending 3708 * or need a VPA update done 3709 */ 3710 prepare_threads(vc); 3711 3712 /* if the runner is no longer runnable, let the caller pick a new one */ 3713 if (vc->runner->arch.state != KVMPPC_VCPU_RUNNABLE) 3714 return; 3715 3716 /* 3717 * Initialize *vc. 3718 */ 3719 init_vcore_to_run(vc); 3720 vc->preempt_tb = TB_NIL; 3721 3722 /* 3723 * Number of threads that we will be controlling: the same as 3724 * the number of threads per subcore, except on POWER9, 3725 * where it's 1 because the threads are (mostly) independent. 3726 */ 3727 controlled_threads = threads_per_vcore(vc->kvm); 3728 3729 /* 3730 * Make sure we are running on primary threads, and that secondary 3731 * threads are offline. Also check if the number of threads in this 3732 * guest are greater than the current system threads per guest. 3733 */ 3734 if ((controlled_threads > 1) && 3735 ((vc->num_threads > threads_per_subcore) || !on_primary_thread())) { 3736 for_each_runnable_thread(i, vcpu, vc) { 3737 vcpu->arch.ret = -EBUSY; 3738 kvmppc_remove_runnable(vc, vcpu, mftb()); 3739 wake_up(&vcpu->arch.cpu_run); 3740 } 3741 goto out; 3742 } 3743 3744 /* 3745 * See if we could run any other vcores on the physical core 3746 * along with this one. 3747 */ 3748 init_core_info(&core_info, vc); 3749 pcpu = smp_processor_id(); 3750 target_threads = controlled_threads; 3751 if (target_smt_mode && target_smt_mode < target_threads) 3752 target_threads = target_smt_mode; 3753 if (vc->num_threads < target_threads) 3754 collect_piggybacks(&core_info, target_threads); 3755 3756 /* 3757 * Hard-disable interrupts, and check resched flag and signals. 3758 * If we need to reschedule or deliver a signal, clean up 3759 * and return without going into the guest(s). 3760 * If the mmu_ready flag has been cleared, don't go into the 3761 * guest because that means a HPT resize operation is in progress. 3762 */ 3763 local_irq_disable(); 3764 hard_irq_disable(); 3765 if (lazy_irq_pending() || need_resched() || 3766 recheck_signals_and_mmu(&core_info)) { 3767 local_irq_enable(); 3768 vc->vcore_state = VCORE_INACTIVE; 3769 /* Unlock all except the primary vcore */ 3770 for (sub = 1; sub < core_info.n_subcores; ++sub) { 3771 pvc = core_info.vc[sub]; 3772 /* Put back on to the preempted vcores list */ 3773 kvmppc_vcore_preempt(pvc); 3774 spin_unlock(&pvc->lock); 3775 } 3776 for (i = 0; i < controlled_threads; ++i) 3777 kvmppc_release_hwthread(pcpu + i); 3778 return; 3779 } 3780 3781 kvmppc_clear_host_core(pcpu); 3782 3783 /* Decide on micro-threading (split-core) mode */ 3784 subcore_size = threads_per_subcore; 3785 cmd_bit = stat_bit = 0; 3786 split = core_info.n_subcores; 3787 sip = NULL; 3788 is_power8 = cpu_has_feature(CPU_FTR_ARCH_207S); 3789 3790 if (split > 1) { 3791 sip = &split_info; 3792 memset(&split_info, 0, sizeof(split_info)); 3793 for (sub = 0; sub < core_info.n_subcores; ++sub) 3794 split_info.vc[sub] = core_info.vc[sub]; 3795 3796 if (is_power8) { 3797 if (split == 2 && (dynamic_mt_modes & 2)) { 3798 cmd_bit = HID0_POWER8_1TO2LPAR; 3799 stat_bit = HID0_POWER8_2LPARMODE; 3800 } else { 3801 split = 4; 3802 cmd_bit = HID0_POWER8_1TO4LPAR; 3803 stat_bit = HID0_POWER8_4LPARMODE; 3804 } 3805 subcore_size = MAX_SMT_THREADS / split; 3806 split_info.rpr = mfspr(SPRN_RPR); 3807 split_info.pmmar = mfspr(SPRN_PMMAR); 3808 split_info.ldbar = mfspr(SPRN_LDBAR); 3809 split_info.subcore_size = subcore_size; 3810 } else { 3811 split_info.subcore_size = 1; 3812 } 3813 3814 /* order writes to split_info before kvm_split_mode pointer */ 3815 smp_wmb(); 3816 } 3817 3818 for (thr = 0; thr < controlled_threads; ++thr) { 3819 struct paca_struct *paca = paca_ptrs[pcpu + thr]; 3820 3821 paca->kvm_hstate.napping = 0; 3822 paca->kvm_hstate.kvm_split_mode = sip; 3823 } 3824 3825 /* Initiate micro-threading (split-core) on POWER8 if required */ 3826 if (cmd_bit) { 3827 unsigned long hid0 = mfspr(SPRN_HID0); 3828 3829 hid0 |= cmd_bit | HID0_POWER8_DYNLPARDIS; 3830 mb(); 3831 mtspr(SPRN_HID0, hid0); 3832 isync(); 3833 for (;;) { 3834 hid0 = mfspr(SPRN_HID0); 3835 if (hid0 & stat_bit) 3836 break; 3837 cpu_relax(); 3838 } 3839 } 3840 3841 /* 3842 * On POWER8, set RWMR register. 3843 * Since it only affects PURR and SPURR, it doesn't affect 3844 * the host, so we don't save/restore the host value. 3845 */ 3846 if (is_power8) { 3847 unsigned long rwmr_val = RWMR_RPA_P8_8THREAD; 3848 int n_online = atomic_read(&vc->online_count); 3849 3850 /* 3851 * Use the 8-thread value if we're doing split-core 3852 * or if the vcore's online count looks bogus. 3853 */ 3854 if (split == 1 && threads_per_subcore == MAX_SMT_THREADS && 3855 n_online >= 1 && n_online <= MAX_SMT_THREADS) 3856 rwmr_val = p8_rwmr_values[n_online]; 3857 mtspr(SPRN_RWMR, rwmr_val); 3858 } 3859 3860 /* Start all the threads */ 3861 active = 0; 3862 for (sub = 0; sub < core_info.n_subcores; ++sub) { 3863 thr = is_power8 ? subcore_thread_map[sub] : sub; 3864 thr0_done = false; 3865 active |= 1 << thr; 3866 pvc = core_info.vc[sub]; 3867 pvc->pcpu = pcpu + thr; 3868 for_each_runnable_thread(i, vcpu, pvc) { 3869 /* 3870 * XXX: is kvmppc_start_thread called too late here? 3871 * It updates vcpu->cpu and vcpu->arch.thread_cpu 3872 * which are used by kvmppc_fast_vcpu_kick_hv(), but 3873 * kick is called after new exceptions become available 3874 * and exceptions are checked earlier than here, by 3875 * kvmppc_core_prepare_to_enter. 3876 */ 3877 kvmppc_start_thread(vcpu, pvc); 3878 kvmppc_update_vpa_dispatch(vcpu, pvc); 3879 trace_kvm_guest_enter(vcpu); 3880 if (!vcpu->arch.ptid) 3881 thr0_done = true; 3882 active |= 1 << (thr + vcpu->arch.ptid); 3883 } 3884 /* 3885 * We need to start the first thread of each subcore 3886 * even if it doesn't have a vcpu. 3887 */ 3888 if (!thr0_done) 3889 kvmppc_start_thread(NULL, pvc); 3890 } 3891 3892 /* 3893 * Ensure that split_info.do_nap is set after setting 3894 * the vcore pointer in the PACA of the secondaries. 3895 */ 3896 smp_mb(); 3897 3898 /* 3899 * When doing micro-threading, poke the inactive threads as well. 3900 * This gets them to the nap instruction after kvm_do_nap, 3901 * which reduces the time taken to unsplit later. 3902 */ 3903 if (cmd_bit) { 3904 split_info.do_nap = 1; /* ask secondaries to nap when done */ 3905 for (thr = 1; thr < threads_per_subcore; ++thr) 3906 if (!(active & (1 << thr))) 3907 kvmppc_ipi_thread(pcpu + thr); 3908 } 3909 3910 vc->vcore_state = VCORE_RUNNING; 3911 preempt_disable(); 3912 3913 trace_kvmppc_run_core(vc, 0); 3914 3915 for (sub = 0; sub < core_info.n_subcores; ++sub) 3916 spin_unlock(&core_info.vc[sub]->lock); 3917 3918 guest_timing_enter_irqoff(); 3919 3920 srcu_idx = srcu_read_lock(&vc->kvm->srcu); 3921 3922 guest_state_enter_irqoff(); 3923 this_cpu_disable_ftrace(); 3924 3925 trap = __kvmppc_vcore_entry(); 3926 3927 this_cpu_enable_ftrace(); 3928 guest_state_exit_irqoff(); 3929 3930 srcu_read_unlock(&vc->kvm->srcu, srcu_idx); 3931 3932 set_irq_happened(trap); 3933 3934 spin_lock(&vc->lock); 3935 /* prevent other vcpu threads from doing kvmppc_start_thread() now */ 3936 vc->vcore_state = VCORE_EXITING; 3937 3938 /* wait for secondary threads to finish writing their state to memory */ 3939 kvmppc_wait_for_nap(controlled_threads); 3940 3941 /* Return to whole-core mode if we split the core earlier */ 3942 if (cmd_bit) { 3943 unsigned long hid0 = mfspr(SPRN_HID0); 3944 unsigned long loops = 0; 3945 3946 hid0 &= ~HID0_POWER8_DYNLPARDIS; 3947 stat_bit = HID0_POWER8_2LPARMODE | HID0_POWER8_4LPARMODE; 3948 mb(); 3949 mtspr(SPRN_HID0, hid0); 3950 isync(); 3951 for (;;) { 3952 hid0 = mfspr(SPRN_HID0); 3953 if (!(hid0 & stat_bit)) 3954 break; 3955 cpu_relax(); 3956 ++loops; 3957 } 3958 split_info.do_nap = 0; 3959 } 3960 3961 kvmppc_set_host_core(pcpu); 3962 3963 if (!vtime_accounting_enabled_this_cpu()) { 3964 local_irq_enable(); 3965 /* 3966 * Service IRQs here before guest_timing_exit_irqoff() so any 3967 * ticks that occurred while running the guest are accounted to 3968 * the guest. If vtime accounting is enabled, accounting uses 3969 * TB rather than ticks, so it can be done without enabling 3970 * interrupts here, which has the problem that it accounts 3971 * interrupt processing overhead to the host. 3972 */ 3973 local_irq_disable(); 3974 } 3975 guest_timing_exit_irqoff(); 3976 3977 local_irq_enable(); 3978 3979 /* Let secondaries go back to the offline loop */ 3980 for (i = 0; i < controlled_threads; ++i) { 3981 kvmppc_release_hwthread(pcpu + i); 3982 if (sip && sip->napped[i]) 3983 kvmppc_ipi_thread(pcpu + i); 3984 } 3985 3986 spin_unlock(&vc->lock); 3987 3988 /* make sure updates to secondary vcpu structs are visible now */ 3989 smp_mb(); 3990 3991 preempt_enable(); 3992 3993 for (sub = 0; sub < core_info.n_subcores; ++sub) { 3994 pvc = core_info.vc[sub]; 3995 post_guest_process(pvc, pvc == vc); 3996 } 3997 3998 spin_lock(&vc->lock); 3999 4000 out: 4001 vc->vcore_state = VCORE_INACTIVE; 4002 trace_kvmppc_run_core(vc, 1); 4003 } 4004 4005 static inline bool hcall_is_xics(unsigned long req) 4006 { 4007 return req == H_EOI || req == H_CPPR || req == H_IPI || 4008 req == H_IPOLL || req == H_XIRR || req == H_XIRR_X; 4009 } 4010 4011 static void vcpu_vpa_increment_dispatch(struct kvm_vcpu *vcpu) 4012 { 4013 struct lppaca *lp = vcpu->arch.vpa.pinned_addr; 4014 if (lp) { 4015 u32 yield_count = be32_to_cpu(lp->yield_count) + 1; 4016 lp->yield_count = cpu_to_be32(yield_count); 4017 vcpu->arch.vpa.dirty = 1; 4018 } 4019 } 4020 4021 /* call our hypervisor to load up HV regs and go */ 4022 static int kvmhv_vcpu_entry_p9_nested(struct kvm_vcpu *vcpu, u64 time_limit, unsigned long lpcr, u64 *tb) 4023 { 4024 struct kvmppc_vcore *vc = vcpu->arch.vcore; 4025 unsigned long host_psscr; 4026 unsigned long msr; 4027 struct hv_guest_state hvregs; 4028 struct p9_host_os_sprs host_os_sprs; 4029 s64 dec; 4030 int trap; 4031 4032 msr = mfmsr(); 4033 4034 save_p9_host_os_sprs(&host_os_sprs); 4035 4036 /* 4037 * We need to save and restore the guest visible part of the 4038 * psscr (i.e. using SPRN_PSSCR_PR) since the hypervisor 4039 * doesn't do this for us. Note only required if pseries since 4040 * this is done in kvmhv_vcpu_entry_p9() below otherwise. 4041 */ 4042 host_psscr = mfspr(SPRN_PSSCR_PR); 4043 4044 kvmppc_msr_hard_disable_set_facilities(vcpu, msr); 4045 if (lazy_irq_pending()) 4046 return 0; 4047 4048 if (unlikely(load_vcpu_state(vcpu, &host_os_sprs))) 4049 msr = mfmsr(); /* TM restore can update msr */ 4050 4051 if (vcpu->arch.psscr != host_psscr) 4052 mtspr(SPRN_PSSCR_PR, vcpu->arch.psscr); 4053 4054 kvmhv_save_hv_regs(vcpu, &hvregs); 4055 hvregs.lpcr = lpcr; 4056 hvregs.amor = ~0; 4057 vcpu->arch.regs.msr = vcpu->arch.shregs.msr; 4058 hvregs.version = HV_GUEST_STATE_VERSION; 4059 if (vcpu->arch.nested) { 4060 hvregs.lpid = vcpu->arch.nested->shadow_lpid; 4061 hvregs.vcpu_token = vcpu->arch.nested_vcpu_id; 4062 } else { 4063 hvregs.lpid = vcpu->kvm->arch.lpid; 4064 hvregs.vcpu_token = vcpu->vcpu_id; 4065 } 4066 hvregs.hdec_expiry = time_limit; 4067 4068 /* 4069 * When setting DEC, we must always deal with irq_work_raise 4070 * via NMI vs setting DEC. The problem occurs right as we 4071 * switch into guest mode if a NMI hits and sets pending work 4072 * and sets DEC, then that will apply to the guest and not 4073 * bring us back to the host. 4074 * 4075 * irq_work_raise could check a flag (or possibly LPCR[HDICE] 4076 * for example) and set HDEC to 1? That wouldn't solve the 4077 * nested hv case which needs to abort the hcall or zero the 4078 * time limit. 4079 * 4080 * XXX: Another day's problem. 4081 */ 4082 mtspr(SPRN_DEC, kvmppc_dec_expires_host_tb(vcpu) - *tb); 4083 4084 mtspr(SPRN_DAR, vcpu->arch.shregs.dar); 4085 mtspr(SPRN_DSISR, vcpu->arch.shregs.dsisr); 4086 switch_pmu_to_guest(vcpu, &host_os_sprs); 4087 accumulate_time(vcpu, &vcpu->arch.in_guest); 4088 trap = plpar_hcall_norets(H_ENTER_NESTED, __pa(&hvregs), 4089 __pa(&vcpu->arch.regs)); 4090 accumulate_time(vcpu, &vcpu->arch.guest_exit); 4091 kvmhv_restore_hv_return_state(vcpu, &hvregs); 4092 switch_pmu_to_host(vcpu, &host_os_sprs); 4093 vcpu->arch.shregs.msr = vcpu->arch.regs.msr; 4094 vcpu->arch.shregs.dar = mfspr(SPRN_DAR); 4095 vcpu->arch.shregs.dsisr = mfspr(SPRN_DSISR); 4096 vcpu->arch.psscr = mfspr(SPRN_PSSCR_PR); 4097 4098 store_vcpu_state(vcpu); 4099 4100 dec = mfspr(SPRN_DEC); 4101 if (!(lpcr & LPCR_LD)) /* Sign extend if not using large decrementer */ 4102 dec = (s32) dec; 4103 *tb = mftb(); 4104 vcpu->arch.dec_expires = dec + (*tb + vc->tb_offset); 4105 4106 timer_rearm_host_dec(*tb); 4107 4108 restore_p9_host_os_sprs(vcpu, &host_os_sprs); 4109 if (vcpu->arch.psscr != host_psscr) 4110 mtspr(SPRN_PSSCR_PR, host_psscr); 4111 4112 return trap; 4113 } 4114 4115 /* 4116 * Guest entry for POWER9 and later CPUs. 4117 */ 4118 static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit, 4119 unsigned long lpcr, u64 *tb) 4120 { 4121 struct kvm *kvm = vcpu->kvm; 4122 struct kvm_nested_guest *nested = vcpu->arch.nested; 4123 u64 next_timer; 4124 int trap; 4125 4126 next_timer = timer_get_next_tb(); 4127 if (*tb >= next_timer) 4128 return BOOK3S_INTERRUPT_HV_DECREMENTER; 4129 if (next_timer < time_limit) 4130 time_limit = next_timer; 4131 else if (*tb >= time_limit) /* nested time limit */ 4132 return BOOK3S_INTERRUPT_NESTED_HV_DECREMENTER; 4133 4134 vcpu->arch.ceded = 0; 4135 4136 vcpu_vpa_increment_dispatch(vcpu); 4137 4138 if (kvmhv_on_pseries()) { 4139 trap = kvmhv_vcpu_entry_p9_nested(vcpu, time_limit, lpcr, tb); 4140 4141 /* H_CEDE has to be handled now, not later */ 4142 if (trap == BOOK3S_INTERRUPT_SYSCALL && !nested && 4143 kvmppc_get_gpr(vcpu, 3) == H_CEDE) { 4144 kvmppc_cede(vcpu); 4145 kvmppc_set_gpr(vcpu, 3, 0); 4146 trap = 0; 4147 } 4148 4149 } else if (nested) { 4150 __this_cpu_write(cpu_in_guest, kvm); 4151 trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr, tb); 4152 __this_cpu_write(cpu_in_guest, NULL); 4153 4154 } else { 4155 kvmppc_xive_push_vcpu(vcpu); 4156 4157 __this_cpu_write(cpu_in_guest, kvm); 4158 trap = kvmhv_vcpu_entry_p9(vcpu, time_limit, lpcr, tb); 4159 __this_cpu_write(cpu_in_guest, NULL); 4160 4161 if (trap == BOOK3S_INTERRUPT_SYSCALL && 4162 !(vcpu->arch.shregs.msr & MSR_PR)) { 4163 unsigned long req = kvmppc_get_gpr(vcpu, 3); 4164 4165 /* 4166 * XIVE rearm and XICS hcalls must be handled 4167 * before xive context is pulled (is this 4168 * true?) 4169 */ 4170 if (req == H_CEDE) { 4171 /* H_CEDE has to be handled now */ 4172 kvmppc_cede(vcpu); 4173 if (!kvmppc_xive_rearm_escalation(vcpu)) { 4174 /* 4175 * Pending escalation so abort 4176 * the cede. 4177 */ 4178 vcpu->arch.ceded = 0; 4179 } 4180 kvmppc_set_gpr(vcpu, 3, 0); 4181 trap = 0; 4182 4183 } else if (req == H_ENTER_NESTED) { 4184 /* 4185 * L2 should not run with the L1 4186 * context so rearm and pull it. 4187 */ 4188 if (!kvmppc_xive_rearm_escalation(vcpu)) { 4189 /* 4190 * Pending escalation so abort 4191 * H_ENTER_NESTED. 4192 */ 4193 kvmppc_set_gpr(vcpu, 3, 0); 4194 trap = 0; 4195 } 4196 4197 } else if (hcall_is_xics(req)) { 4198 int ret; 4199 4200 ret = kvmppc_xive_xics_hcall(vcpu, req); 4201 if (ret != H_TOO_HARD) { 4202 kvmppc_set_gpr(vcpu, 3, ret); 4203 trap = 0; 4204 } 4205 } 4206 } 4207 kvmppc_xive_pull_vcpu(vcpu); 4208 4209 if (kvm_is_radix(kvm)) 4210 vcpu->arch.slb_max = 0; 4211 } 4212 4213 vcpu_vpa_increment_dispatch(vcpu); 4214 4215 return trap; 4216 } 4217 4218 /* 4219 * Wait for some other vcpu thread to execute us, and 4220 * wake us up when we need to handle something in the host. 4221 */ 4222 static void kvmppc_wait_for_exec(struct kvmppc_vcore *vc, 4223 struct kvm_vcpu *vcpu, int wait_state) 4224 { 4225 DEFINE_WAIT(wait); 4226 4227 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state); 4228 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) { 4229 spin_unlock(&vc->lock); 4230 schedule(); 4231 spin_lock(&vc->lock); 4232 } 4233 finish_wait(&vcpu->arch.cpu_run, &wait); 4234 } 4235 4236 static void grow_halt_poll_ns(struct kvmppc_vcore *vc) 4237 { 4238 if (!halt_poll_ns_grow) 4239 return; 4240 4241 vc->halt_poll_ns *= halt_poll_ns_grow; 4242 if (vc->halt_poll_ns < halt_poll_ns_grow_start) 4243 vc->halt_poll_ns = halt_poll_ns_grow_start; 4244 } 4245 4246 static void shrink_halt_poll_ns(struct kvmppc_vcore *vc) 4247 { 4248 if (halt_poll_ns_shrink == 0) 4249 vc->halt_poll_ns = 0; 4250 else 4251 vc->halt_poll_ns /= halt_poll_ns_shrink; 4252 } 4253 4254 #ifdef CONFIG_KVM_XICS 4255 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu) 4256 { 4257 if (!xics_on_xive()) 4258 return false; 4259 return vcpu->arch.irq_pending || vcpu->arch.xive_saved_state.pipr < 4260 vcpu->arch.xive_saved_state.cppr; 4261 } 4262 #else 4263 static inline bool xive_interrupt_pending(struct kvm_vcpu *vcpu) 4264 { 4265 return false; 4266 } 4267 #endif /* CONFIG_KVM_XICS */ 4268 4269 static bool kvmppc_vcpu_woken(struct kvm_vcpu *vcpu) 4270 { 4271 if (vcpu->arch.pending_exceptions || vcpu->arch.prodded || 4272 kvmppc_doorbell_pending(vcpu) || xive_interrupt_pending(vcpu)) 4273 return true; 4274 4275 return false; 4276 } 4277 4278 static bool kvmppc_vcpu_check_block(struct kvm_vcpu *vcpu) 4279 { 4280 if (!vcpu->arch.ceded || kvmppc_vcpu_woken(vcpu)) 4281 return true; 4282 return false; 4283 } 4284 4285 /* 4286 * Check to see if any of the runnable vcpus on the vcore have pending 4287 * exceptions or are no longer ceded 4288 */ 4289 static int kvmppc_vcore_check_block(struct kvmppc_vcore *vc) 4290 { 4291 struct kvm_vcpu *vcpu; 4292 int i; 4293 4294 for_each_runnable_thread(i, vcpu, vc) { 4295 if (kvmppc_vcpu_check_block(vcpu)) 4296 return 1; 4297 } 4298 4299 return 0; 4300 } 4301 4302 /* 4303 * All the vcpus in this vcore are idle, so wait for a decrementer 4304 * or external interrupt to one of the vcpus. vc->lock is held. 4305 */ 4306 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc) 4307 { 4308 ktime_t cur, start_poll, start_wait; 4309 int do_sleep = 1; 4310 u64 block_ns; 4311 4312 WARN_ON_ONCE(cpu_has_feature(CPU_FTR_ARCH_300)); 4313 4314 /* Poll for pending exceptions and ceded state */ 4315 cur = start_poll = ktime_get(); 4316 if (vc->halt_poll_ns) { 4317 ktime_t stop = ktime_add_ns(start_poll, vc->halt_poll_ns); 4318 ++vc->runner->stat.generic.halt_attempted_poll; 4319 4320 vc->vcore_state = VCORE_POLLING; 4321 spin_unlock(&vc->lock); 4322 4323 do { 4324 if (kvmppc_vcore_check_block(vc)) { 4325 do_sleep = 0; 4326 break; 4327 } 4328 cur = ktime_get(); 4329 } while (kvm_vcpu_can_poll(cur, stop)); 4330 4331 spin_lock(&vc->lock); 4332 vc->vcore_state = VCORE_INACTIVE; 4333 4334 if (!do_sleep) { 4335 ++vc->runner->stat.generic.halt_successful_poll; 4336 goto out; 4337 } 4338 } 4339 4340 prepare_to_rcuwait(&vc->wait); 4341 set_current_state(TASK_INTERRUPTIBLE); 4342 if (kvmppc_vcore_check_block(vc)) { 4343 finish_rcuwait(&vc->wait); 4344 do_sleep = 0; 4345 /* If we polled, count this as a successful poll */ 4346 if (vc->halt_poll_ns) 4347 ++vc->runner->stat.generic.halt_successful_poll; 4348 goto out; 4349 } 4350 4351 start_wait = ktime_get(); 4352 4353 vc->vcore_state = VCORE_SLEEPING; 4354 trace_kvmppc_vcore_blocked(vc->runner, 0); 4355 spin_unlock(&vc->lock); 4356 schedule(); 4357 finish_rcuwait(&vc->wait); 4358 spin_lock(&vc->lock); 4359 vc->vcore_state = VCORE_INACTIVE; 4360 trace_kvmppc_vcore_blocked(vc->runner, 1); 4361 ++vc->runner->stat.halt_successful_wait; 4362 4363 cur = ktime_get(); 4364 4365 out: 4366 block_ns = ktime_to_ns(cur) - ktime_to_ns(start_poll); 4367 4368 /* Attribute wait time */ 4369 if (do_sleep) { 4370 vc->runner->stat.generic.halt_wait_ns += 4371 ktime_to_ns(cur) - ktime_to_ns(start_wait); 4372 KVM_STATS_LOG_HIST_UPDATE( 4373 vc->runner->stat.generic.halt_wait_hist, 4374 ktime_to_ns(cur) - ktime_to_ns(start_wait)); 4375 /* Attribute failed poll time */ 4376 if (vc->halt_poll_ns) { 4377 vc->runner->stat.generic.halt_poll_fail_ns += 4378 ktime_to_ns(start_wait) - 4379 ktime_to_ns(start_poll); 4380 KVM_STATS_LOG_HIST_UPDATE( 4381 vc->runner->stat.generic.halt_poll_fail_hist, 4382 ktime_to_ns(start_wait) - 4383 ktime_to_ns(start_poll)); 4384 } 4385 } else { 4386 /* Attribute successful poll time */ 4387 if (vc->halt_poll_ns) { 4388 vc->runner->stat.generic.halt_poll_success_ns += 4389 ktime_to_ns(cur) - 4390 ktime_to_ns(start_poll); 4391 KVM_STATS_LOG_HIST_UPDATE( 4392 vc->runner->stat.generic.halt_poll_success_hist, 4393 ktime_to_ns(cur) - ktime_to_ns(start_poll)); 4394 } 4395 } 4396 4397 /* Adjust poll time */ 4398 if (halt_poll_ns) { 4399 if (block_ns <= vc->halt_poll_ns) 4400 ; 4401 /* We slept and blocked for longer than the max halt time */ 4402 else if (vc->halt_poll_ns && block_ns > halt_poll_ns) 4403 shrink_halt_poll_ns(vc); 4404 /* We slept and our poll time is too small */ 4405 else if (vc->halt_poll_ns < halt_poll_ns && 4406 block_ns < halt_poll_ns) 4407 grow_halt_poll_ns(vc); 4408 if (vc->halt_poll_ns > halt_poll_ns) 4409 vc->halt_poll_ns = halt_poll_ns; 4410 } else 4411 vc->halt_poll_ns = 0; 4412 4413 trace_kvmppc_vcore_wakeup(do_sleep, block_ns); 4414 } 4415 4416 /* 4417 * This never fails for a radix guest, as none of the operations it does 4418 * for a radix guest can fail or have a way to report failure. 4419 */ 4420 static int kvmhv_setup_mmu(struct kvm_vcpu *vcpu) 4421 { 4422 int r = 0; 4423 struct kvm *kvm = vcpu->kvm; 4424 4425 mutex_lock(&kvm->arch.mmu_setup_lock); 4426 if (!kvm->arch.mmu_ready) { 4427 if (!kvm_is_radix(kvm)) 4428 r = kvmppc_hv_setup_htab_rma(vcpu); 4429 if (!r) { 4430 if (cpu_has_feature(CPU_FTR_ARCH_300)) 4431 kvmppc_setup_partition_table(kvm); 4432 kvm->arch.mmu_ready = 1; 4433 } 4434 } 4435 mutex_unlock(&kvm->arch.mmu_setup_lock); 4436 return r; 4437 } 4438 4439 static int kvmppc_run_vcpu(struct kvm_vcpu *vcpu) 4440 { 4441 struct kvm_run *run = vcpu->run; 4442 int n_ceded, i, r; 4443 struct kvmppc_vcore *vc; 4444 struct kvm_vcpu *v; 4445 4446 trace_kvmppc_run_vcpu_enter(vcpu); 4447 4448 run->exit_reason = 0; 4449 vcpu->arch.ret = RESUME_GUEST; 4450 vcpu->arch.trap = 0; 4451 kvmppc_update_vpas(vcpu); 4452 4453 /* 4454 * Synchronize with other threads in this virtual core 4455 */ 4456 vc = vcpu->arch.vcore; 4457 spin_lock(&vc->lock); 4458 vcpu->arch.ceded = 0; 4459 vcpu->arch.run_task = current; 4460 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb()); 4461 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE; 4462 vcpu->arch.busy_preempt = TB_NIL; 4463 WRITE_ONCE(vc->runnable_threads[vcpu->arch.ptid], vcpu); 4464 ++vc->n_runnable; 4465 4466 /* 4467 * This happens the first time this is called for a vcpu. 4468 * If the vcore is already running, we may be able to start 4469 * this thread straight away and have it join in. 4470 */ 4471 if (!signal_pending(current)) { 4472 if ((vc->vcore_state == VCORE_PIGGYBACK || 4473 vc->vcore_state == VCORE_RUNNING) && 4474 !VCORE_IS_EXITING(vc)) { 4475 kvmppc_update_vpa_dispatch(vcpu, vc); 4476 kvmppc_start_thread(vcpu, vc); 4477 trace_kvm_guest_enter(vcpu); 4478 } else if (vc->vcore_state == VCORE_SLEEPING) { 4479 rcuwait_wake_up(&vc->wait); 4480 } 4481 4482 } 4483 4484 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE && 4485 !signal_pending(current)) { 4486 /* See if the MMU is ready to go */ 4487 if (!vcpu->kvm->arch.mmu_ready) { 4488 spin_unlock(&vc->lock); 4489 r = kvmhv_setup_mmu(vcpu); 4490 spin_lock(&vc->lock); 4491 if (r) { 4492 run->exit_reason = KVM_EXIT_FAIL_ENTRY; 4493 run->fail_entry. 4494 hardware_entry_failure_reason = 0; 4495 vcpu->arch.ret = r; 4496 break; 4497 } 4498 } 4499 4500 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL) 4501 kvmppc_vcore_end_preempt(vc); 4502 4503 if (vc->vcore_state != VCORE_INACTIVE) { 4504 kvmppc_wait_for_exec(vc, vcpu, TASK_INTERRUPTIBLE); 4505 continue; 4506 } 4507 for_each_runnable_thread(i, v, vc) { 4508 kvmppc_core_prepare_to_enter(v); 4509 if (signal_pending(v->arch.run_task)) { 4510 kvmppc_remove_runnable(vc, v, mftb()); 4511 v->stat.signal_exits++; 4512 v->run->exit_reason = KVM_EXIT_INTR; 4513 v->arch.ret = -EINTR; 4514 wake_up(&v->arch.cpu_run); 4515 } 4516 } 4517 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE) 4518 break; 4519 n_ceded = 0; 4520 for_each_runnable_thread(i, v, vc) { 4521 if (!kvmppc_vcpu_woken(v)) 4522 n_ceded += v->arch.ceded; 4523 else 4524 v->arch.ceded = 0; 4525 } 4526 vc->runner = vcpu; 4527 if (n_ceded == vc->n_runnable) { 4528 kvmppc_vcore_blocked(vc); 4529 } else if (need_resched()) { 4530 kvmppc_vcore_preempt(vc); 4531 /* Let something else run */ 4532 cond_resched_lock(&vc->lock); 4533 if (vc->vcore_state == VCORE_PREEMPT) 4534 kvmppc_vcore_end_preempt(vc); 4535 } else { 4536 kvmppc_run_core(vc); 4537 } 4538 vc->runner = NULL; 4539 } 4540 4541 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE && 4542 (vc->vcore_state == VCORE_RUNNING || 4543 vc->vcore_state == VCORE_EXITING || 4544 vc->vcore_state == VCORE_PIGGYBACK)) 4545 kvmppc_wait_for_exec(vc, vcpu, TASK_UNINTERRUPTIBLE); 4546 4547 if (vc->vcore_state == VCORE_PREEMPT && vc->runner == NULL) 4548 kvmppc_vcore_end_preempt(vc); 4549 4550 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) { 4551 kvmppc_remove_runnable(vc, vcpu, mftb()); 4552 vcpu->stat.signal_exits++; 4553 run->exit_reason = KVM_EXIT_INTR; 4554 vcpu->arch.ret = -EINTR; 4555 } 4556 4557 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) { 4558 /* Wake up some vcpu to run the core */ 4559 i = -1; 4560 v = next_runnable_thread(vc, &i); 4561 wake_up(&v->arch.cpu_run); 4562 } 4563 4564 trace_kvmppc_run_vcpu_exit(vcpu); 4565 spin_unlock(&vc->lock); 4566 return vcpu->arch.ret; 4567 } 4568 4569 int kvmhv_run_single_vcpu(struct kvm_vcpu *vcpu, u64 time_limit, 4570 unsigned long lpcr) 4571 { 4572 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu); 4573 struct kvm_run *run = vcpu->run; 4574 int trap, r, pcpu; 4575 int srcu_idx; 4576 struct kvmppc_vcore *vc; 4577 struct kvm *kvm = vcpu->kvm; 4578 struct kvm_nested_guest *nested = vcpu->arch.nested; 4579 unsigned long flags; 4580 u64 tb; 4581 4582 trace_kvmppc_run_vcpu_enter(vcpu); 4583 4584 run->exit_reason = 0; 4585 vcpu->arch.ret = RESUME_GUEST; 4586 vcpu->arch.trap = 0; 4587 4588 vc = vcpu->arch.vcore; 4589 vcpu->arch.ceded = 0; 4590 vcpu->arch.run_task = current; 4591 vcpu->arch.last_inst = KVM_INST_FETCH_FAILED; 4592 4593 /* See if the MMU is ready to go */ 4594 if (unlikely(!kvm->arch.mmu_ready)) { 4595 r = kvmhv_setup_mmu(vcpu); 4596 if (r) { 4597 run->exit_reason = KVM_EXIT_FAIL_ENTRY; 4598 run->fail_entry.hardware_entry_failure_reason = 0; 4599 vcpu->arch.ret = r; 4600 return r; 4601 } 4602 } 4603 4604 if (need_resched()) 4605 cond_resched(); 4606 4607 kvmppc_update_vpas(vcpu); 4608 4609 preempt_disable(); 4610 pcpu = smp_processor_id(); 4611 if (kvm_is_radix(kvm)) 4612 kvmppc_prepare_radix_vcpu(vcpu, pcpu); 4613 4614 /* flags save not required, but irq_pmu has no disable/enable API */ 4615 powerpc_local_irq_pmu_save(flags); 4616 4617 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE; 4618 4619 if (signal_pending(current)) 4620 goto sigpend; 4621 if (need_resched() || !kvm->arch.mmu_ready) 4622 goto out; 4623 4624 vcpu->cpu = pcpu; 4625 vcpu->arch.thread_cpu = pcpu; 4626 vc->pcpu = pcpu; 4627 local_paca->kvm_hstate.kvm_vcpu = vcpu; 4628 local_paca->kvm_hstate.ptid = 0; 4629 local_paca->kvm_hstate.fake_suspend = 0; 4630 4631 /* 4632 * Orders set cpu/thread_cpu vs testing for pending interrupts and 4633 * doorbells below. The other side is when these fields are set vs 4634 * kvmppc_fast_vcpu_kick_hv reading the cpu/thread_cpu fields to 4635 * kick a vCPU to notice the pending interrupt. 4636 */ 4637 smp_mb(); 4638 4639 if (!nested) { 4640 kvmppc_core_prepare_to_enter(vcpu); 4641 if (vcpu->arch.shregs.msr & MSR_EE) { 4642 if (xive_interrupt_pending(vcpu)) 4643 kvmppc_inject_interrupt_hv(vcpu, 4644 BOOK3S_INTERRUPT_EXTERNAL, 0); 4645 } else if (test_bit(BOOK3S_IRQPRIO_EXTERNAL, 4646 &vcpu->arch.pending_exceptions)) { 4647 lpcr |= LPCR_MER; 4648 } 4649 } else if (vcpu->arch.pending_exceptions || 4650 vcpu->arch.doorbell_request || 4651 xive_interrupt_pending(vcpu)) { 4652 vcpu->arch.ret = RESUME_HOST; 4653 goto out; 4654 } 4655 4656 if (vcpu->arch.timer_running) { 4657 hrtimer_try_to_cancel(&vcpu->arch.dec_timer); 4658 vcpu->arch.timer_running = 0; 4659 } 4660 4661 tb = mftb(); 4662 4663 kvmppc_update_vpa_dispatch_p9(vcpu, vc, tb + vc->tb_offset); 4664 4665 trace_kvm_guest_enter(vcpu); 4666 4667 guest_timing_enter_irqoff(); 4668 4669 srcu_idx = srcu_read_lock(&kvm->srcu); 4670 4671 guest_state_enter_irqoff(); 4672 this_cpu_disable_ftrace(); 4673 4674 trap = kvmhv_p9_guest_entry(vcpu, time_limit, lpcr, &tb); 4675 vcpu->arch.trap = trap; 4676 4677 this_cpu_enable_ftrace(); 4678 guest_state_exit_irqoff(); 4679 4680 srcu_read_unlock(&kvm->srcu, srcu_idx); 4681 4682 set_irq_happened(trap); 4683 4684 vcpu->cpu = -1; 4685 vcpu->arch.thread_cpu = -1; 4686 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST; 4687 4688 if (!vtime_accounting_enabled_this_cpu()) { 4689 powerpc_local_irq_pmu_restore(flags); 4690 /* 4691 * Service IRQs here before guest_timing_exit_irqoff() so any 4692 * ticks that occurred while running the guest are accounted to 4693 * the guest. If vtime accounting is enabled, accounting uses 4694 * TB rather than ticks, so it can be done without enabling 4695 * interrupts here, which has the problem that it accounts 4696 * interrupt processing overhead to the host. 4697 */ 4698 powerpc_local_irq_pmu_save(flags); 4699 } 4700 guest_timing_exit_irqoff(); 4701 4702 powerpc_local_irq_pmu_restore(flags); 4703 4704 preempt_enable(); 4705 4706 /* 4707 * cancel pending decrementer exception if DEC is now positive, or if 4708 * entering a nested guest in which case the decrementer is now owned 4709 * by L2 and the L1 decrementer is provided in hdec_expires 4710 */ 4711 if (kvmppc_core_pending_dec(vcpu) && 4712 ((tb < kvmppc_dec_expires_host_tb(vcpu)) || 4713 (trap == BOOK3S_INTERRUPT_SYSCALL && 4714 kvmppc_get_gpr(vcpu, 3) == H_ENTER_NESTED))) 4715 kvmppc_core_dequeue_dec(vcpu); 4716 4717 trace_kvm_guest_exit(vcpu); 4718 r = RESUME_GUEST; 4719 if (trap) { 4720 if (!nested) 4721 r = kvmppc_handle_exit_hv(vcpu, current); 4722 else 4723 r = kvmppc_handle_nested_exit(vcpu); 4724 } 4725 vcpu->arch.ret = r; 4726 4727 if (is_kvmppc_resume_guest(r) && !kvmppc_vcpu_check_block(vcpu)) { 4728 kvmppc_set_timer(vcpu); 4729 4730 prepare_to_rcuwait(wait); 4731 for (;;) { 4732 set_current_state(TASK_INTERRUPTIBLE); 4733 if (signal_pending(current)) { 4734 vcpu->stat.signal_exits++; 4735 run->exit_reason = KVM_EXIT_INTR; 4736 vcpu->arch.ret = -EINTR; 4737 break; 4738 } 4739 4740 if (kvmppc_vcpu_check_block(vcpu)) 4741 break; 4742 4743 trace_kvmppc_vcore_blocked(vcpu, 0); 4744 schedule(); 4745 trace_kvmppc_vcore_blocked(vcpu, 1); 4746 } 4747 finish_rcuwait(wait); 4748 } 4749 vcpu->arch.ceded = 0; 4750 4751 done: 4752 trace_kvmppc_run_vcpu_exit(vcpu); 4753 4754 return vcpu->arch.ret; 4755 4756 sigpend: 4757 vcpu->stat.signal_exits++; 4758 run->exit_reason = KVM_EXIT_INTR; 4759 vcpu->arch.ret = -EINTR; 4760 out: 4761 vcpu->cpu = -1; 4762 vcpu->arch.thread_cpu = -1; 4763 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST; 4764 powerpc_local_irq_pmu_restore(flags); 4765 preempt_enable(); 4766 goto done; 4767 } 4768 4769 static int kvmppc_vcpu_run_hv(struct kvm_vcpu *vcpu) 4770 { 4771 struct kvm_run *run = vcpu->run; 4772 int r; 4773 int srcu_idx; 4774 struct kvm *kvm; 4775 unsigned long msr; 4776 4777 start_timing(vcpu, &vcpu->arch.vcpu_entry); 4778 4779 if (!vcpu->arch.sane) { 4780 run->exit_reason = KVM_EXIT_INTERNAL_ERROR; 4781 return -EINVAL; 4782 } 4783 4784 /* No need to go into the guest when all we'll do is come back out */ 4785 if (signal_pending(current)) { 4786 run->exit_reason = KVM_EXIT_INTR; 4787 return -EINTR; 4788 } 4789 4790 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 4791 /* 4792 * Don't allow entry with a suspended transaction, because 4793 * the guest entry/exit code will lose it. 4794 */ 4795 if (cpu_has_feature(CPU_FTR_TM) && current->thread.regs && 4796 (current->thread.regs->msr & MSR_TM)) { 4797 if (MSR_TM_ACTIVE(current->thread.regs->msr)) { 4798 run->exit_reason = KVM_EXIT_FAIL_ENTRY; 4799 run->fail_entry.hardware_entry_failure_reason = 0; 4800 return -EINVAL; 4801 } 4802 } 4803 #endif 4804 4805 /* 4806 * Force online to 1 for the sake of old userspace which doesn't 4807 * set it. 4808 */ 4809 if (!vcpu->arch.online) { 4810 atomic_inc(&vcpu->arch.vcore->online_count); 4811 vcpu->arch.online = 1; 4812 } 4813 4814 kvmppc_core_prepare_to_enter(vcpu); 4815 4816 kvm = vcpu->kvm; 4817 atomic_inc(&kvm->arch.vcpus_running); 4818 /* Order vcpus_running vs. mmu_ready, see kvmppc_alloc_reset_hpt */ 4819 smp_mb(); 4820 4821 msr = 0; 4822 if (IS_ENABLED(CONFIG_PPC_FPU)) 4823 msr |= MSR_FP; 4824 if (cpu_has_feature(CPU_FTR_ALTIVEC)) 4825 msr |= MSR_VEC; 4826 if (cpu_has_feature(CPU_FTR_VSX)) 4827 msr |= MSR_VSX; 4828 if ((cpu_has_feature(CPU_FTR_TM) || 4829 cpu_has_feature(CPU_FTR_P9_TM_HV_ASSIST)) && 4830 (vcpu->arch.hfscr & HFSCR_TM)) 4831 msr |= MSR_TM; 4832 msr = msr_check_and_set(msr); 4833 4834 kvmppc_save_user_regs(); 4835 4836 kvmppc_save_current_sprs(); 4837 4838 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 4839 vcpu->arch.waitp = &vcpu->arch.vcore->wait; 4840 vcpu->arch.pgdir = kvm->mm->pgd; 4841 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST; 4842 4843 do { 4844 accumulate_time(vcpu, &vcpu->arch.guest_entry); 4845 if (cpu_has_feature(CPU_FTR_ARCH_300)) 4846 r = kvmhv_run_single_vcpu(vcpu, ~(u64)0, 4847 vcpu->arch.vcore->lpcr); 4848 else 4849 r = kvmppc_run_vcpu(vcpu); 4850 4851 if (run->exit_reason == KVM_EXIT_PAPR_HCALL) { 4852 accumulate_time(vcpu, &vcpu->arch.hcall); 4853 4854 if (WARN_ON_ONCE(vcpu->arch.shregs.msr & MSR_PR)) { 4855 /* 4856 * These should have been caught reflected 4857 * into the guest by now. Final sanity check: 4858 * don't allow userspace to execute hcalls in 4859 * the hypervisor. 4860 */ 4861 r = RESUME_GUEST; 4862 continue; 4863 } 4864 trace_kvm_hcall_enter(vcpu); 4865 r = kvmppc_pseries_do_hcall(vcpu); 4866 trace_kvm_hcall_exit(vcpu, r); 4867 kvmppc_core_prepare_to_enter(vcpu); 4868 } else if (r == RESUME_PAGE_FAULT) { 4869 accumulate_time(vcpu, &vcpu->arch.pg_fault); 4870 srcu_idx = srcu_read_lock(&kvm->srcu); 4871 r = kvmppc_book3s_hv_page_fault(vcpu, 4872 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr); 4873 srcu_read_unlock(&kvm->srcu, srcu_idx); 4874 } else if (r == RESUME_PASSTHROUGH) { 4875 if (WARN_ON(xics_on_xive())) 4876 r = H_SUCCESS; 4877 else 4878 r = kvmppc_xics_rm_complete(vcpu, 0); 4879 } 4880 } while (is_kvmppc_resume_guest(r)); 4881 accumulate_time(vcpu, &vcpu->arch.vcpu_exit); 4882 4883 vcpu->arch.state = KVMPPC_VCPU_NOTREADY; 4884 atomic_dec(&kvm->arch.vcpus_running); 4885 4886 srr_regs_clobbered(); 4887 4888 end_timing(vcpu); 4889 4890 return r; 4891 } 4892 4893 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps, 4894 int shift, int sllp) 4895 { 4896 (*sps)->page_shift = shift; 4897 (*sps)->slb_enc = sllp; 4898 (*sps)->enc[0].page_shift = shift; 4899 (*sps)->enc[0].pte_enc = kvmppc_pgsize_lp_encoding(shift, shift); 4900 /* 4901 * Add 16MB MPSS support (may get filtered out by userspace) 4902 */ 4903 if (shift != 24) { 4904 int penc = kvmppc_pgsize_lp_encoding(shift, 24); 4905 if (penc != -1) { 4906 (*sps)->enc[1].page_shift = 24; 4907 (*sps)->enc[1].pte_enc = penc; 4908 } 4909 } 4910 (*sps)++; 4911 } 4912 4913 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm, 4914 struct kvm_ppc_smmu_info *info) 4915 { 4916 struct kvm_ppc_one_seg_page_size *sps; 4917 4918 /* 4919 * POWER7, POWER8 and POWER9 all support 32 storage keys for data. 4920 * POWER7 doesn't support keys for instruction accesses, 4921 * POWER8 and POWER9 do. 4922 */ 4923 info->data_keys = 32; 4924 info->instr_keys = cpu_has_feature(CPU_FTR_ARCH_207S) ? 32 : 0; 4925 4926 /* POWER7, 8 and 9 all have 1T segments and 32-entry SLB */ 4927 info->flags = KVM_PPC_PAGE_SIZES_REAL | KVM_PPC_1T_SEGMENTS; 4928 info->slb_size = 32; 4929 4930 /* We only support these sizes for now, and no muti-size segments */ 4931 sps = &info->sps[0]; 4932 kvmppc_add_seg_page_size(&sps, 12, 0); 4933 kvmppc_add_seg_page_size(&sps, 16, SLB_VSID_L | SLB_VSID_LP_01); 4934 kvmppc_add_seg_page_size(&sps, 24, SLB_VSID_L); 4935 4936 /* If running as a nested hypervisor, we don't support HPT guests */ 4937 if (kvmhv_on_pseries()) 4938 info->flags |= KVM_PPC_NO_HASH; 4939 4940 return 0; 4941 } 4942 4943 /* 4944 * Get (and clear) the dirty memory log for a memory slot. 4945 */ 4946 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm, 4947 struct kvm_dirty_log *log) 4948 { 4949 struct kvm_memslots *slots; 4950 struct kvm_memory_slot *memslot; 4951 int r; 4952 unsigned long n, i; 4953 unsigned long *buf, *p; 4954 struct kvm_vcpu *vcpu; 4955 4956 mutex_lock(&kvm->slots_lock); 4957 4958 r = -EINVAL; 4959 if (log->slot >= KVM_USER_MEM_SLOTS) 4960 goto out; 4961 4962 slots = kvm_memslots(kvm); 4963 memslot = id_to_memslot(slots, log->slot); 4964 r = -ENOENT; 4965 if (!memslot || !memslot->dirty_bitmap) 4966 goto out; 4967 4968 /* 4969 * Use second half of bitmap area because both HPT and radix 4970 * accumulate bits in the first half. 4971 */ 4972 n = kvm_dirty_bitmap_bytes(memslot); 4973 buf = memslot->dirty_bitmap + n / sizeof(long); 4974 memset(buf, 0, n); 4975 4976 if (kvm_is_radix(kvm)) 4977 r = kvmppc_hv_get_dirty_log_radix(kvm, memslot, buf); 4978 else 4979 r = kvmppc_hv_get_dirty_log_hpt(kvm, memslot, buf); 4980 if (r) 4981 goto out; 4982 4983 /* 4984 * We accumulate dirty bits in the first half of the 4985 * memslot's dirty_bitmap area, for when pages are paged 4986 * out or modified by the host directly. Pick up these 4987 * bits and add them to the map. 4988 */ 4989 p = memslot->dirty_bitmap; 4990 for (i = 0; i < n / sizeof(long); ++i) 4991 buf[i] |= xchg(&p[i], 0); 4992 4993 /* Harvest dirty bits from VPA and DTL updates */ 4994 /* Note: we never modify the SLB shadow buffer areas */ 4995 kvm_for_each_vcpu(i, vcpu, kvm) { 4996 spin_lock(&vcpu->arch.vpa_update_lock); 4997 kvmppc_harvest_vpa_dirty(&vcpu->arch.vpa, memslot, buf); 4998 kvmppc_harvest_vpa_dirty(&vcpu->arch.dtl, memslot, buf); 4999 spin_unlock(&vcpu->arch.vpa_update_lock); 5000 } 5001 5002 r = -EFAULT; 5003 if (copy_to_user(log->dirty_bitmap, buf, n)) 5004 goto out; 5005 5006 r = 0; 5007 out: 5008 mutex_unlock(&kvm->slots_lock); 5009 return r; 5010 } 5011 5012 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *slot) 5013 { 5014 vfree(slot->arch.rmap); 5015 slot->arch.rmap = NULL; 5016 } 5017 5018 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm, 5019 const struct kvm_memory_slot *old, 5020 struct kvm_memory_slot *new, 5021 enum kvm_mr_change change) 5022 { 5023 if (change == KVM_MR_CREATE) { 5024 unsigned long size = array_size(new->npages, sizeof(*new->arch.rmap)); 5025 5026 if ((size >> PAGE_SHIFT) > totalram_pages()) 5027 return -ENOMEM; 5028 5029 new->arch.rmap = vzalloc(size); 5030 if (!new->arch.rmap) 5031 return -ENOMEM; 5032 } else if (change != KVM_MR_DELETE) { 5033 new->arch.rmap = old->arch.rmap; 5034 } 5035 5036 return 0; 5037 } 5038 5039 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm, 5040 struct kvm_memory_slot *old, 5041 const struct kvm_memory_slot *new, 5042 enum kvm_mr_change change) 5043 { 5044 /* 5045 * If we are creating or modifying a memslot, it might make 5046 * some address that was previously cached as emulated 5047 * MMIO be no longer emulated MMIO, so invalidate 5048 * all the caches of emulated MMIO translations. 5049 */ 5050 if (change != KVM_MR_DELETE) 5051 atomic64_inc(&kvm->arch.mmio_update); 5052 5053 /* 5054 * For change == KVM_MR_MOVE or KVM_MR_DELETE, higher levels 5055 * have already called kvm_arch_flush_shadow_memslot() to 5056 * flush shadow mappings. For KVM_MR_CREATE we have no 5057 * previous mappings. So the only case to handle is 5058 * KVM_MR_FLAGS_ONLY when the KVM_MEM_LOG_DIRTY_PAGES bit 5059 * has been changed. 5060 * For radix guests, we flush on setting KVM_MEM_LOG_DIRTY_PAGES 5061 * to get rid of any THP PTEs in the partition-scoped page tables 5062 * so we can track dirtiness at the page level; we flush when 5063 * clearing KVM_MEM_LOG_DIRTY_PAGES so that we can go back to 5064 * using THP PTEs. 5065 */ 5066 if (change == KVM_MR_FLAGS_ONLY && kvm_is_radix(kvm) && 5067 ((new->flags ^ old->flags) & KVM_MEM_LOG_DIRTY_PAGES)) 5068 kvmppc_radix_flush_memslot(kvm, old); 5069 /* 5070 * If UV hasn't yet called H_SVM_INIT_START, don't register memslots. 5071 */ 5072 if (!kvm->arch.secure_guest) 5073 return; 5074 5075 switch (change) { 5076 case KVM_MR_CREATE: 5077 /* 5078 * @TODO kvmppc_uvmem_memslot_create() can fail and 5079 * return error. Fix this. 5080 */ 5081 kvmppc_uvmem_memslot_create(kvm, new); 5082 break; 5083 case KVM_MR_DELETE: 5084 kvmppc_uvmem_memslot_delete(kvm, old); 5085 break; 5086 default: 5087 /* TODO: Handle KVM_MR_MOVE */ 5088 break; 5089 } 5090 } 5091 5092 /* 5093 * Update LPCR values in kvm->arch and in vcores. 5094 * Caller must hold kvm->arch.mmu_setup_lock (for mutual exclusion 5095 * of kvm->arch.lpcr update). 5096 */ 5097 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask) 5098 { 5099 long int i; 5100 u32 cores_done = 0; 5101 5102 if ((kvm->arch.lpcr & mask) == lpcr) 5103 return; 5104 5105 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr; 5106 5107 for (i = 0; i < KVM_MAX_VCORES; ++i) { 5108 struct kvmppc_vcore *vc = kvm->arch.vcores[i]; 5109 if (!vc) 5110 continue; 5111 5112 spin_lock(&vc->lock); 5113 vc->lpcr = (vc->lpcr & ~mask) | lpcr; 5114 verify_lpcr(kvm, vc->lpcr); 5115 spin_unlock(&vc->lock); 5116 if (++cores_done >= kvm->arch.online_vcores) 5117 break; 5118 } 5119 } 5120 5121 void kvmppc_setup_partition_table(struct kvm *kvm) 5122 { 5123 unsigned long dw0, dw1; 5124 5125 if (!kvm_is_radix(kvm)) { 5126 /* PS field - page size for VRMA */ 5127 dw0 = ((kvm->arch.vrma_slb_v & SLB_VSID_L) >> 1) | 5128 ((kvm->arch.vrma_slb_v & SLB_VSID_LP) << 1); 5129 /* HTABSIZE and HTABORG fields */ 5130 dw0 |= kvm->arch.sdr1; 5131 5132 /* Second dword as set by userspace */ 5133 dw1 = kvm->arch.process_table; 5134 } else { 5135 dw0 = PATB_HR | radix__get_tree_size() | 5136 __pa(kvm->arch.pgtable) | RADIX_PGD_INDEX_SIZE; 5137 dw1 = PATB_GR | kvm->arch.process_table; 5138 } 5139 kvmhv_set_ptbl_entry(kvm->arch.lpid, dw0, dw1); 5140 } 5141 5142 /* 5143 * Set up HPT (hashed page table) and RMA (real-mode area). 5144 * Must be called with kvm->arch.mmu_setup_lock held. 5145 */ 5146 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu) 5147 { 5148 int err = 0; 5149 struct kvm *kvm = vcpu->kvm; 5150 unsigned long hva; 5151 struct kvm_memory_slot *memslot; 5152 struct vm_area_struct *vma; 5153 unsigned long lpcr = 0, senc; 5154 unsigned long psize, porder; 5155 int srcu_idx; 5156 5157 /* Allocate hashed page table (if not done already) and reset it */ 5158 if (!kvm->arch.hpt.virt) { 5159 int order = KVM_DEFAULT_HPT_ORDER; 5160 struct kvm_hpt_info info; 5161 5162 err = kvmppc_allocate_hpt(&info, order); 5163 /* If we get here, it means userspace didn't specify a 5164 * size explicitly. So, try successively smaller 5165 * sizes if the default failed. */ 5166 while ((err == -ENOMEM) && --order >= PPC_MIN_HPT_ORDER) 5167 err = kvmppc_allocate_hpt(&info, order); 5168 5169 if (err < 0) { 5170 pr_err("KVM: Couldn't alloc HPT\n"); 5171 goto out; 5172 } 5173 5174 kvmppc_set_hpt(kvm, &info); 5175 } 5176 5177 /* Look up the memslot for guest physical address 0 */ 5178 srcu_idx = srcu_read_lock(&kvm->srcu); 5179 memslot = gfn_to_memslot(kvm, 0); 5180 5181 /* We must have some memory at 0 by now */ 5182 err = -EINVAL; 5183 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) 5184 goto out_srcu; 5185 5186 /* Look up the VMA for the start of this memory slot */ 5187 hva = memslot->userspace_addr; 5188 mmap_read_lock(kvm->mm); 5189 vma = vma_lookup(kvm->mm, hva); 5190 if (!vma || (vma->vm_flags & VM_IO)) 5191 goto up_out; 5192 5193 psize = vma_kernel_pagesize(vma); 5194 5195 mmap_read_unlock(kvm->mm); 5196 5197 /* We can handle 4k, 64k or 16M pages in the VRMA */ 5198 if (psize >= 0x1000000) 5199 psize = 0x1000000; 5200 else if (psize >= 0x10000) 5201 psize = 0x10000; 5202 else 5203 psize = 0x1000; 5204 porder = __ilog2(psize); 5205 5206 senc = slb_pgsize_encoding(psize); 5207 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T | 5208 (VRMA_VSID << SLB_VSID_SHIFT_1T); 5209 /* Create HPTEs in the hash page table for the VRMA */ 5210 kvmppc_map_vrma(vcpu, memslot, porder); 5211 5212 /* Update VRMASD field in the LPCR */ 5213 if (!cpu_has_feature(CPU_FTR_ARCH_300)) { 5214 /* the -4 is to account for senc values starting at 0x10 */ 5215 lpcr = senc << (LPCR_VRMASD_SH - 4); 5216 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD); 5217 } 5218 5219 /* Order updates to kvm->arch.lpcr etc. vs. mmu_ready */ 5220 smp_wmb(); 5221 err = 0; 5222 out_srcu: 5223 srcu_read_unlock(&kvm->srcu, srcu_idx); 5224 out: 5225 return err; 5226 5227 up_out: 5228 mmap_read_unlock(kvm->mm); 5229 goto out_srcu; 5230 } 5231 5232 /* 5233 * Must be called with kvm->arch.mmu_setup_lock held and 5234 * mmu_ready = 0 and no vcpus running. 5235 */ 5236 int kvmppc_switch_mmu_to_hpt(struct kvm *kvm) 5237 { 5238 unsigned long lpcr, lpcr_mask; 5239 5240 if (nesting_enabled(kvm)) 5241 kvmhv_release_all_nested(kvm); 5242 kvmppc_rmap_reset(kvm); 5243 kvm->arch.process_table = 0; 5244 /* Mutual exclusion with kvm_unmap_gfn_range etc. */ 5245 spin_lock(&kvm->mmu_lock); 5246 kvm->arch.radix = 0; 5247 spin_unlock(&kvm->mmu_lock); 5248 kvmppc_free_radix(kvm); 5249 5250 lpcr = LPCR_VPM1; 5251 lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR; 5252 if (cpu_has_feature(CPU_FTR_ARCH_31)) 5253 lpcr_mask |= LPCR_HAIL; 5254 kvmppc_update_lpcr(kvm, lpcr, lpcr_mask); 5255 5256 return 0; 5257 } 5258 5259 /* 5260 * Must be called with kvm->arch.mmu_setup_lock held and 5261 * mmu_ready = 0 and no vcpus running. 5262 */ 5263 int kvmppc_switch_mmu_to_radix(struct kvm *kvm) 5264 { 5265 unsigned long lpcr, lpcr_mask; 5266 int err; 5267 5268 err = kvmppc_init_vm_radix(kvm); 5269 if (err) 5270 return err; 5271 kvmppc_rmap_reset(kvm); 5272 /* Mutual exclusion with kvm_unmap_gfn_range etc. */ 5273 spin_lock(&kvm->mmu_lock); 5274 kvm->arch.radix = 1; 5275 spin_unlock(&kvm->mmu_lock); 5276 kvmppc_free_hpt(&kvm->arch.hpt); 5277 5278 lpcr = LPCR_UPRT | LPCR_GTSE | LPCR_HR; 5279 lpcr_mask = LPCR_VPM1 | LPCR_UPRT | LPCR_GTSE | LPCR_HR; 5280 if (cpu_has_feature(CPU_FTR_ARCH_31)) { 5281 lpcr_mask |= LPCR_HAIL; 5282 if (cpu_has_feature(CPU_FTR_HVMODE) && 5283 (kvm->arch.host_lpcr & LPCR_HAIL)) 5284 lpcr |= LPCR_HAIL; 5285 } 5286 kvmppc_update_lpcr(kvm, lpcr, lpcr_mask); 5287 5288 return 0; 5289 } 5290 5291 #ifdef CONFIG_KVM_XICS 5292 /* 5293 * Allocate a per-core structure for managing state about which cores are 5294 * running in the host versus the guest and for exchanging data between 5295 * real mode KVM and CPU running in the host. 5296 * This is only done for the first VM. 5297 * The allocated structure stays even if all VMs have stopped. 5298 * It is only freed when the kvm-hv module is unloaded. 5299 * It's OK for this routine to fail, we just don't support host 5300 * core operations like redirecting H_IPI wakeups. 5301 */ 5302 void kvmppc_alloc_host_rm_ops(void) 5303 { 5304 struct kvmppc_host_rm_ops *ops; 5305 unsigned long l_ops; 5306 int cpu, core; 5307 int size; 5308 5309 if (cpu_has_feature(CPU_FTR_ARCH_300)) 5310 return; 5311 5312 /* Not the first time here ? */ 5313 if (kvmppc_host_rm_ops_hv != NULL) 5314 return; 5315 5316 ops = kzalloc(sizeof(struct kvmppc_host_rm_ops), GFP_KERNEL); 5317 if (!ops) 5318 return; 5319 5320 size = cpu_nr_cores() * sizeof(struct kvmppc_host_rm_core); 5321 ops->rm_core = kzalloc(size, GFP_KERNEL); 5322 5323 if (!ops->rm_core) { 5324 kfree(ops); 5325 return; 5326 } 5327 5328 cpus_read_lock(); 5329 5330 for (cpu = 0; cpu < nr_cpu_ids; cpu += threads_per_core) { 5331 if (!cpu_online(cpu)) 5332 continue; 5333 5334 core = cpu >> threads_shift; 5335 ops->rm_core[core].rm_state.in_host = 1; 5336 } 5337 5338 ops->vcpu_kick = kvmppc_fast_vcpu_kick_hv; 5339 5340 /* 5341 * Make the contents of the kvmppc_host_rm_ops structure visible 5342 * to other CPUs before we assign it to the global variable. 5343 * Do an atomic assignment (no locks used here), but if someone 5344 * beats us to it, just free our copy and return. 5345 */ 5346 smp_wmb(); 5347 l_ops = (unsigned long) ops; 5348 5349 if (cmpxchg64((unsigned long *)&kvmppc_host_rm_ops_hv, 0, l_ops)) { 5350 cpus_read_unlock(); 5351 kfree(ops->rm_core); 5352 kfree(ops); 5353 return; 5354 } 5355 5356 cpuhp_setup_state_nocalls_cpuslocked(CPUHP_KVM_PPC_BOOK3S_PREPARE, 5357 "ppc/kvm_book3s:prepare", 5358 kvmppc_set_host_core, 5359 kvmppc_clear_host_core); 5360 cpus_read_unlock(); 5361 } 5362 5363 void kvmppc_free_host_rm_ops(void) 5364 { 5365 if (kvmppc_host_rm_ops_hv) { 5366 cpuhp_remove_state_nocalls(CPUHP_KVM_PPC_BOOK3S_PREPARE); 5367 kfree(kvmppc_host_rm_ops_hv->rm_core); 5368 kfree(kvmppc_host_rm_ops_hv); 5369 kvmppc_host_rm_ops_hv = NULL; 5370 } 5371 } 5372 #endif 5373 5374 static int kvmppc_core_init_vm_hv(struct kvm *kvm) 5375 { 5376 unsigned long lpcr, lpid; 5377 int ret; 5378 5379 mutex_init(&kvm->arch.uvmem_lock); 5380 INIT_LIST_HEAD(&kvm->arch.uvmem_pfns); 5381 mutex_init(&kvm->arch.mmu_setup_lock); 5382 5383 /* Allocate the guest's logical partition ID */ 5384 5385 lpid = kvmppc_alloc_lpid(); 5386 if ((long)lpid < 0) 5387 return -ENOMEM; 5388 kvm->arch.lpid = lpid; 5389 5390 kvmppc_alloc_host_rm_ops(); 5391 5392 kvmhv_vm_nested_init(kvm); 5393 5394 /* 5395 * Since we don't flush the TLB when tearing down a VM, 5396 * and this lpid might have previously been used, 5397 * make sure we flush on each core before running the new VM. 5398 * On POWER9, the tlbie in mmu_partition_table_set_entry() 5399 * does this flush for us. 5400 */ 5401 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 5402 cpumask_setall(&kvm->arch.need_tlb_flush); 5403 5404 /* Start out with the default set of hcalls enabled */ 5405 memcpy(kvm->arch.enabled_hcalls, default_enabled_hcalls, 5406 sizeof(kvm->arch.enabled_hcalls)); 5407 5408 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 5409 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1); 5410 5411 /* Init LPCR for virtual RMA mode */ 5412 if (cpu_has_feature(CPU_FTR_HVMODE)) { 5413 kvm->arch.host_lpid = mfspr(SPRN_LPID); 5414 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR); 5415 lpcr &= LPCR_PECE | LPCR_LPES; 5416 } else { 5417 /* 5418 * The L2 LPES mode will be set by the L0 according to whether 5419 * or not it needs to take external interrupts in HV mode. 5420 */ 5421 lpcr = 0; 5422 } 5423 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE | 5424 LPCR_VPM0 | LPCR_VPM1; 5425 kvm->arch.vrma_slb_v = SLB_VSID_B_1T | 5426 (VRMA_VSID << SLB_VSID_SHIFT_1T); 5427 /* On POWER8 turn on online bit to enable PURR/SPURR */ 5428 if (cpu_has_feature(CPU_FTR_ARCH_207S)) 5429 lpcr |= LPCR_ONL; 5430 /* 5431 * On POWER9, VPM0 bit is reserved (VPM0=1 behaviour is assumed) 5432 * Set HVICE bit to enable hypervisor virtualization interrupts. 5433 * Set HEIC to prevent OS interrupts to go to hypervisor (should 5434 * be unnecessary but better safe than sorry in case we re-enable 5435 * EE in HV mode with this LPCR still set) 5436 */ 5437 if (cpu_has_feature(CPU_FTR_ARCH_300)) { 5438 lpcr &= ~LPCR_VPM0; 5439 lpcr |= LPCR_HVICE | LPCR_HEIC; 5440 5441 /* 5442 * If xive is enabled, we route 0x500 interrupts directly 5443 * to the guest. 5444 */ 5445 if (xics_on_xive()) 5446 lpcr |= LPCR_LPES; 5447 } 5448 5449 /* 5450 * If the host uses radix, the guest starts out as radix. 5451 */ 5452 if (radix_enabled()) { 5453 kvm->arch.radix = 1; 5454 kvm->arch.mmu_ready = 1; 5455 lpcr &= ~LPCR_VPM1; 5456 lpcr |= LPCR_UPRT | LPCR_GTSE | LPCR_HR; 5457 if (cpu_has_feature(CPU_FTR_HVMODE) && 5458 cpu_has_feature(CPU_FTR_ARCH_31) && 5459 (kvm->arch.host_lpcr & LPCR_HAIL)) 5460 lpcr |= LPCR_HAIL; 5461 ret = kvmppc_init_vm_radix(kvm); 5462 if (ret) { 5463 kvmppc_free_lpid(kvm->arch.lpid); 5464 return ret; 5465 } 5466 kvmppc_setup_partition_table(kvm); 5467 } 5468 5469 verify_lpcr(kvm, lpcr); 5470 kvm->arch.lpcr = lpcr; 5471 5472 /* Initialization for future HPT resizes */ 5473 kvm->arch.resize_hpt = NULL; 5474 5475 /* 5476 * Work out how many sets the TLB has, for the use of 5477 * the TLB invalidation loop in book3s_hv_rmhandlers.S. 5478 */ 5479 if (cpu_has_feature(CPU_FTR_ARCH_31)) { 5480 /* 5481 * P10 will flush all the congruence class with a single tlbiel 5482 */ 5483 kvm->arch.tlb_sets = 1; 5484 } else if (radix_enabled()) 5485 kvm->arch.tlb_sets = POWER9_TLB_SETS_RADIX; /* 128 */ 5486 else if (cpu_has_feature(CPU_FTR_ARCH_300)) 5487 kvm->arch.tlb_sets = POWER9_TLB_SETS_HASH; /* 256 */ 5488 else if (cpu_has_feature(CPU_FTR_ARCH_207S)) 5489 kvm->arch.tlb_sets = POWER8_TLB_SETS; /* 512 */ 5490 else 5491 kvm->arch.tlb_sets = POWER7_TLB_SETS; /* 128 */ 5492 5493 /* 5494 * Track that we now have a HV mode VM active. This blocks secondary 5495 * CPU threads from coming online. 5496 */ 5497 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 5498 kvm_hv_vm_activated(); 5499 5500 /* 5501 * Initialize smt_mode depending on processor. 5502 * POWER8 and earlier have to use "strict" threading, where 5503 * all vCPUs in a vcore have to run on the same (sub)core, 5504 * whereas on POWER9 the threads can each run a different 5505 * guest. 5506 */ 5507 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 5508 kvm->arch.smt_mode = threads_per_subcore; 5509 else 5510 kvm->arch.smt_mode = 1; 5511 kvm->arch.emul_smt_mode = 1; 5512 5513 return 0; 5514 } 5515 5516 static int kvmppc_arch_create_vm_debugfs_hv(struct kvm *kvm) 5517 { 5518 kvmppc_mmu_debugfs_init(kvm); 5519 if (radix_enabled()) 5520 kvmhv_radix_debugfs_init(kvm); 5521 return 0; 5522 } 5523 5524 static void kvmppc_free_vcores(struct kvm *kvm) 5525 { 5526 long int i; 5527 5528 for (i = 0; i < KVM_MAX_VCORES; ++i) 5529 kfree(kvm->arch.vcores[i]); 5530 kvm->arch.online_vcores = 0; 5531 } 5532 5533 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm) 5534 { 5535 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 5536 kvm_hv_vm_deactivated(); 5537 5538 kvmppc_free_vcores(kvm); 5539 5540 5541 if (kvm_is_radix(kvm)) 5542 kvmppc_free_radix(kvm); 5543 else 5544 kvmppc_free_hpt(&kvm->arch.hpt); 5545 5546 /* Perform global invalidation and return lpid to the pool */ 5547 if (cpu_has_feature(CPU_FTR_ARCH_300)) { 5548 if (nesting_enabled(kvm)) 5549 kvmhv_release_all_nested(kvm); 5550 kvm->arch.process_table = 0; 5551 if (kvm->arch.secure_guest) 5552 uv_svm_terminate(kvm->arch.lpid); 5553 kvmhv_set_ptbl_entry(kvm->arch.lpid, 0, 0); 5554 } 5555 5556 kvmppc_free_lpid(kvm->arch.lpid); 5557 5558 kvmppc_free_pimap(kvm); 5559 } 5560 5561 /* We don't need to emulate any privileged instructions or dcbz */ 5562 static int kvmppc_core_emulate_op_hv(struct kvm_vcpu *vcpu, 5563 unsigned int inst, int *advance) 5564 { 5565 return EMULATE_FAIL; 5566 } 5567 5568 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn, 5569 ulong spr_val) 5570 { 5571 return EMULATE_FAIL; 5572 } 5573 5574 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn, 5575 ulong *spr_val) 5576 { 5577 return EMULATE_FAIL; 5578 } 5579 5580 static int kvmppc_core_check_processor_compat_hv(void) 5581 { 5582 if (cpu_has_feature(CPU_FTR_HVMODE) && 5583 cpu_has_feature(CPU_FTR_ARCH_206)) 5584 return 0; 5585 5586 /* POWER9 in radix mode is capable of being a nested hypervisor. */ 5587 if (cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled()) 5588 return 0; 5589 5590 return -EIO; 5591 } 5592 5593 #ifdef CONFIG_KVM_XICS 5594 5595 void kvmppc_free_pimap(struct kvm *kvm) 5596 { 5597 kfree(kvm->arch.pimap); 5598 } 5599 5600 static struct kvmppc_passthru_irqmap *kvmppc_alloc_pimap(void) 5601 { 5602 return kzalloc(sizeof(struct kvmppc_passthru_irqmap), GFP_KERNEL); 5603 } 5604 5605 static int kvmppc_set_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi) 5606 { 5607 struct irq_desc *desc; 5608 struct kvmppc_irq_map *irq_map; 5609 struct kvmppc_passthru_irqmap *pimap; 5610 struct irq_chip *chip; 5611 int i, rc = 0; 5612 struct irq_data *host_data; 5613 5614 if (!kvm_irq_bypass) 5615 return 1; 5616 5617 desc = irq_to_desc(host_irq); 5618 if (!desc) 5619 return -EIO; 5620 5621 mutex_lock(&kvm->lock); 5622 5623 pimap = kvm->arch.pimap; 5624 if (pimap == NULL) { 5625 /* First call, allocate structure to hold IRQ map */ 5626 pimap = kvmppc_alloc_pimap(); 5627 if (pimap == NULL) { 5628 mutex_unlock(&kvm->lock); 5629 return -ENOMEM; 5630 } 5631 kvm->arch.pimap = pimap; 5632 } 5633 5634 /* 5635 * For now, we only support interrupts for which the EOI operation 5636 * is an OPAL call followed by a write to XIRR, since that's 5637 * what our real-mode EOI code does, or a XIVE interrupt 5638 */ 5639 chip = irq_data_get_irq_chip(&desc->irq_data); 5640 if (!chip || !is_pnv_opal_msi(chip)) { 5641 pr_warn("kvmppc_set_passthru_irq_hv: Could not assign IRQ map for (%d,%d)\n", 5642 host_irq, guest_gsi); 5643 mutex_unlock(&kvm->lock); 5644 return -ENOENT; 5645 } 5646 5647 /* 5648 * See if we already have an entry for this guest IRQ number. 5649 * If it's mapped to a hardware IRQ number, that's an error, 5650 * otherwise re-use this entry. 5651 */ 5652 for (i = 0; i < pimap->n_mapped; i++) { 5653 if (guest_gsi == pimap->mapped[i].v_hwirq) { 5654 if (pimap->mapped[i].r_hwirq) { 5655 mutex_unlock(&kvm->lock); 5656 return -EINVAL; 5657 } 5658 break; 5659 } 5660 } 5661 5662 if (i == KVMPPC_PIRQ_MAPPED) { 5663 mutex_unlock(&kvm->lock); 5664 return -EAGAIN; /* table is full */ 5665 } 5666 5667 irq_map = &pimap->mapped[i]; 5668 5669 irq_map->v_hwirq = guest_gsi; 5670 irq_map->desc = desc; 5671 5672 /* 5673 * Order the above two stores before the next to serialize with 5674 * the KVM real mode handler. 5675 */ 5676 smp_wmb(); 5677 5678 /* 5679 * The 'host_irq' number is mapped in the PCI-MSI domain but 5680 * the underlying calls, which will EOI the interrupt in real 5681 * mode, need an HW IRQ number mapped in the XICS IRQ domain. 5682 */ 5683 host_data = irq_domain_get_irq_data(irq_get_default_host(), host_irq); 5684 irq_map->r_hwirq = (unsigned int)irqd_to_hwirq(host_data); 5685 5686 if (i == pimap->n_mapped) 5687 pimap->n_mapped++; 5688 5689 if (xics_on_xive()) 5690 rc = kvmppc_xive_set_mapped(kvm, guest_gsi, host_irq); 5691 else 5692 kvmppc_xics_set_mapped(kvm, guest_gsi, irq_map->r_hwirq); 5693 if (rc) 5694 irq_map->r_hwirq = 0; 5695 5696 mutex_unlock(&kvm->lock); 5697 5698 return 0; 5699 } 5700 5701 static int kvmppc_clr_passthru_irq(struct kvm *kvm, int host_irq, int guest_gsi) 5702 { 5703 struct irq_desc *desc; 5704 struct kvmppc_passthru_irqmap *pimap; 5705 int i, rc = 0; 5706 5707 if (!kvm_irq_bypass) 5708 return 0; 5709 5710 desc = irq_to_desc(host_irq); 5711 if (!desc) 5712 return -EIO; 5713 5714 mutex_lock(&kvm->lock); 5715 if (!kvm->arch.pimap) 5716 goto unlock; 5717 5718 pimap = kvm->arch.pimap; 5719 5720 for (i = 0; i < pimap->n_mapped; i++) { 5721 if (guest_gsi == pimap->mapped[i].v_hwirq) 5722 break; 5723 } 5724 5725 if (i == pimap->n_mapped) { 5726 mutex_unlock(&kvm->lock); 5727 return -ENODEV; 5728 } 5729 5730 if (xics_on_xive()) 5731 rc = kvmppc_xive_clr_mapped(kvm, guest_gsi, host_irq); 5732 else 5733 kvmppc_xics_clr_mapped(kvm, guest_gsi, pimap->mapped[i].r_hwirq); 5734 5735 /* invalidate the entry (what to do on error from the above ?) */ 5736 pimap->mapped[i].r_hwirq = 0; 5737 5738 /* 5739 * We don't free this structure even when the count goes to 5740 * zero. The structure is freed when we destroy the VM. 5741 */ 5742 unlock: 5743 mutex_unlock(&kvm->lock); 5744 return rc; 5745 } 5746 5747 static int kvmppc_irq_bypass_add_producer_hv(struct irq_bypass_consumer *cons, 5748 struct irq_bypass_producer *prod) 5749 { 5750 int ret = 0; 5751 struct kvm_kernel_irqfd *irqfd = 5752 container_of(cons, struct kvm_kernel_irqfd, consumer); 5753 5754 irqfd->producer = prod; 5755 5756 ret = kvmppc_set_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi); 5757 if (ret) 5758 pr_info("kvmppc_set_passthru_irq (irq %d, gsi %d) fails: %d\n", 5759 prod->irq, irqfd->gsi, ret); 5760 5761 return ret; 5762 } 5763 5764 static void kvmppc_irq_bypass_del_producer_hv(struct irq_bypass_consumer *cons, 5765 struct irq_bypass_producer *prod) 5766 { 5767 int ret; 5768 struct kvm_kernel_irqfd *irqfd = 5769 container_of(cons, struct kvm_kernel_irqfd, consumer); 5770 5771 irqfd->producer = NULL; 5772 5773 /* 5774 * When producer of consumer is unregistered, we change back to 5775 * default external interrupt handling mode - KVM real mode 5776 * will switch back to host. 5777 */ 5778 ret = kvmppc_clr_passthru_irq(irqfd->kvm, prod->irq, irqfd->gsi); 5779 if (ret) 5780 pr_warn("kvmppc_clr_passthru_irq (irq %d, gsi %d) fails: %d\n", 5781 prod->irq, irqfd->gsi, ret); 5782 } 5783 #endif 5784 5785 static long kvm_arch_vm_ioctl_hv(struct file *filp, 5786 unsigned int ioctl, unsigned long arg) 5787 { 5788 struct kvm *kvm __maybe_unused = filp->private_data; 5789 void __user *argp = (void __user *)arg; 5790 long r; 5791 5792 switch (ioctl) { 5793 5794 case KVM_PPC_ALLOCATE_HTAB: { 5795 u32 htab_order; 5796 5797 /* If we're a nested hypervisor, we currently only support radix */ 5798 if (kvmhv_on_pseries()) { 5799 r = -EOPNOTSUPP; 5800 break; 5801 } 5802 5803 r = -EFAULT; 5804 if (get_user(htab_order, (u32 __user *)argp)) 5805 break; 5806 r = kvmppc_alloc_reset_hpt(kvm, htab_order); 5807 if (r) 5808 break; 5809 r = 0; 5810 break; 5811 } 5812 5813 case KVM_PPC_GET_HTAB_FD: { 5814 struct kvm_get_htab_fd ghf; 5815 5816 r = -EFAULT; 5817 if (copy_from_user(&ghf, argp, sizeof(ghf))) 5818 break; 5819 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf); 5820 break; 5821 } 5822 5823 case KVM_PPC_RESIZE_HPT_PREPARE: { 5824 struct kvm_ppc_resize_hpt rhpt; 5825 5826 r = -EFAULT; 5827 if (copy_from_user(&rhpt, argp, sizeof(rhpt))) 5828 break; 5829 5830 r = kvm_vm_ioctl_resize_hpt_prepare(kvm, &rhpt); 5831 break; 5832 } 5833 5834 case KVM_PPC_RESIZE_HPT_COMMIT: { 5835 struct kvm_ppc_resize_hpt rhpt; 5836 5837 r = -EFAULT; 5838 if (copy_from_user(&rhpt, argp, sizeof(rhpt))) 5839 break; 5840 5841 r = kvm_vm_ioctl_resize_hpt_commit(kvm, &rhpt); 5842 break; 5843 } 5844 5845 default: 5846 r = -ENOTTY; 5847 } 5848 5849 return r; 5850 } 5851 5852 /* 5853 * List of hcall numbers to enable by default. 5854 * For compatibility with old userspace, we enable by default 5855 * all hcalls that were implemented before the hcall-enabling 5856 * facility was added. Note this list should not include H_RTAS. 5857 */ 5858 static unsigned int default_hcall_list[] = { 5859 H_REMOVE, 5860 H_ENTER, 5861 H_READ, 5862 H_PROTECT, 5863 H_BULK_REMOVE, 5864 #ifdef CONFIG_SPAPR_TCE_IOMMU 5865 H_GET_TCE, 5866 H_PUT_TCE, 5867 #endif 5868 H_SET_DABR, 5869 H_SET_XDABR, 5870 H_CEDE, 5871 H_PROD, 5872 H_CONFER, 5873 H_REGISTER_VPA, 5874 #ifdef CONFIG_KVM_XICS 5875 H_EOI, 5876 H_CPPR, 5877 H_IPI, 5878 H_IPOLL, 5879 H_XIRR, 5880 H_XIRR_X, 5881 #endif 5882 0 5883 }; 5884 5885 static void init_default_hcalls(void) 5886 { 5887 int i; 5888 unsigned int hcall; 5889 5890 for (i = 0; default_hcall_list[i]; ++i) { 5891 hcall = default_hcall_list[i]; 5892 WARN_ON(!kvmppc_hcall_impl_hv(hcall)); 5893 __set_bit(hcall / 4, default_enabled_hcalls); 5894 } 5895 } 5896 5897 static int kvmhv_configure_mmu(struct kvm *kvm, struct kvm_ppc_mmuv3_cfg *cfg) 5898 { 5899 unsigned long lpcr; 5900 int radix; 5901 int err; 5902 5903 /* If not on a POWER9, reject it */ 5904 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 5905 return -ENODEV; 5906 5907 /* If any unknown flags set, reject it */ 5908 if (cfg->flags & ~(KVM_PPC_MMUV3_RADIX | KVM_PPC_MMUV3_GTSE)) 5909 return -EINVAL; 5910 5911 /* GR (guest radix) bit in process_table field must match */ 5912 radix = !!(cfg->flags & KVM_PPC_MMUV3_RADIX); 5913 if (!!(cfg->process_table & PATB_GR) != radix) 5914 return -EINVAL; 5915 5916 /* Process table size field must be reasonable, i.e. <= 24 */ 5917 if ((cfg->process_table & PRTS_MASK) > 24) 5918 return -EINVAL; 5919 5920 /* We can change a guest to/from radix now, if the host is radix */ 5921 if (radix && !radix_enabled()) 5922 return -EINVAL; 5923 5924 /* If we're a nested hypervisor, we currently only support radix */ 5925 if (kvmhv_on_pseries() && !radix) 5926 return -EINVAL; 5927 5928 mutex_lock(&kvm->arch.mmu_setup_lock); 5929 if (radix != kvm_is_radix(kvm)) { 5930 if (kvm->arch.mmu_ready) { 5931 kvm->arch.mmu_ready = 0; 5932 /* order mmu_ready vs. vcpus_running */ 5933 smp_mb(); 5934 if (atomic_read(&kvm->arch.vcpus_running)) { 5935 kvm->arch.mmu_ready = 1; 5936 err = -EBUSY; 5937 goto out_unlock; 5938 } 5939 } 5940 if (radix) 5941 err = kvmppc_switch_mmu_to_radix(kvm); 5942 else 5943 err = kvmppc_switch_mmu_to_hpt(kvm); 5944 if (err) 5945 goto out_unlock; 5946 } 5947 5948 kvm->arch.process_table = cfg->process_table; 5949 kvmppc_setup_partition_table(kvm); 5950 5951 lpcr = (cfg->flags & KVM_PPC_MMUV3_GTSE) ? LPCR_GTSE : 0; 5952 kvmppc_update_lpcr(kvm, lpcr, LPCR_GTSE); 5953 err = 0; 5954 5955 out_unlock: 5956 mutex_unlock(&kvm->arch.mmu_setup_lock); 5957 return err; 5958 } 5959 5960 static int kvmhv_enable_nested(struct kvm *kvm) 5961 { 5962 if (!nested) 5963 return -EPERM; 5964 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 5965 return -ENODEV; 5966 if (!radix_enabled()) 5967 return -ENODEV; 5968 5969 /* kvm == NULL means the caller is testing if the capability exists */ 5970 if (kvm) 5971 kvm->arch.nested_enable = true; 5972 return 0; 5973 } 5974 5975 static int kvmhv_load_from_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr, 5976 int size) 5977 { 5978 int rc = -EINVAL; 5979 5980 if (kvmhv_vcpu_is_radix(vcpu)) { 5981 rc = kvmhv_copy_from_guest_radix(vcpu, *eaddr, ptr, size); 5982 5983 if (rc > 0) 5984 rc = -EINVAL; 5985 } 5986 5987 /* For now quadrants are the only way to access nested guest memory */ 5988 if (rc && vcpu->arch.nested) 5989 rc = -EAGAIN; 5990 5991 return rc; 5992 } 5993 5994 static int kvmhv_store_to_eaddr(struct kvm_vcpu *vcpu, ulong *eaddr, void *ptr, 5995 int size) 5996 { 5997 int rc = -EINVAL; 5998 5999 if (kvmhv_vcpu_is_radix(vcpu)) { 6000 rc = kvmhv_copy_to_guest_radix(vcpu, *eaddr, ptr, size); 6001 6002 if (rc > 0) 6003 rc = -EINVAL; 6004 } 6005 6006 /* For now quadrants are the only way to access nested guest memory */ 6007 if (rc && vcpu->arch.nested) 6008 rc = -EAGAIN; 6009 6010 return rc; 6011 } 6012 6013 static void unpin_vpa_reset(struct kvm *kvm, struct kvmppc_vpa *vpa) 6014 { 6015 unpin_vpa(kvm, vpa); 6016 vpa->gpa = 0; 6017 vpa->pinned_addr = NULL; 6018 vpa->dirty = false; 6019 vpa->update_pending = 0; 6020 } 6021 6022 /* 6023 * Enable a guest to become a secure VM, or test whether 6024 * that could be enabled. 6025 * Called when the KVM_CAP_PPC_SECURE_GUEST capability is 6026 * tested (kvm == NULL) or enabled (kvm != NULL). 6027 */ 6028 static int kvmhv_enable_svm(struct kvm *kvm) 6029 { 6030 if (!kvmppc_uvmem_available()) 6031 return -EINVAL; 6032 if (kvm) 6033 kvm->arch.svm_enabled = 1; 6034 return 0; 6035 } 6036 6037 /* 6038 * IOCTL handler to turn off secure mode of guest 6039 * 6040 * - Release all device pages 6041 * - Issue ucall to terminate the guest on the UV side 6042 * - Unpin the VPA pages. 6043 * - Reinit the partition scoped page tables 6044 */ 6045 static int kvmhv_svm_off(struct kvm *kvm) 6046 { 6047 struct kvm_vcpu *vcpu; 6048 int mmu_was_ready; 6049 int srcu_idx; 6050 int ret = 0; 6051 unsigned long i; 6052 6053 if (!(kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START)) 6054 return ret; 6055 6056 mutex_lock(&kvm->arch.mmu_setup_lock); 6057 mmu_was_ready = kvm->arch.mmu_ready; 6058 if (kvm->arch.mmu_ready) { 6059 kvm->arch.mmu_ready = 0; 6060 /* order mmu_ready vs. vcpus_running */ 6061 smp_mb(); 6062 if (atomic_read(&kvm->arch.vcpus_running)) { 6063 kvm->arch.mmu_ready = 1; 6064 ret = -EBUSY; 6065 goto out; 6066 } 6067 } 6068 6069 srcu_idx = srcu_read_lock(&kvm->srcu); 6070 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { 6071 struct kvm_memory_slot *memslot; 6072 struct kvm_memslots *slots = __kvm_memslots(kvm, i); 6073 int bkt; 6074 6075 if (!slots) 6076 continue; 6077 6078 kvm_for_each_memslot(memslot, bkt, slots) { 6079 kvmppc_uvmem_drop_pages(memslot, kvm, true); 6080 uv_unregister_mem_slot(kvm->arch.lpid, memslot->id); 6081 } 6082 } 6083 srcu_read_unlock(&kvm->srcu, srcu_idx); 6084 6085 ret = uv_svm_terminate(kvm->arch.lpid); 6086 if (ret != U_SUCCESS) { 6087 ret = -EINVAL; 6088 goto out; 6089 } 6090 6091 /* 6092 * When secure guest is reset, all the guest pages are sent 6093 * to UV via UV_PAGE_IN before the non-boot vcpus get a 6094 * chance to run and unpin their VPA pages. Unpinning of all 6095 * VPA pages is done here explicitly so that VPA pages 6096 * can be migrated to the secure side. 6097 * 6098 * This is required to for the secure SMP guest to reboot 6099 * correctly. 6100 */ 6101 kvm_for_each_vcpu(i, vcpu, kvm) { 6102 spin_lock(&vcpu->arch.vpa_update_lock); 6103 unpin_vpa_reset(kvm, &vcpu->arch.dtl); 6104 unpin_vpa_reset(kvm, &vcpu->arch.slb_shadow); 6105 unpin_vpa_reset(kvm, &vcpu->arch.vpa); 6106 spin_unlock(&vcpu->arch.vpa_update_lock); 6107 } 6108 6109 kvmppc_setup_partition_table(kvm); 6110 kvm->arch.secure_guest = 0; 6111 kvm->arch.mmu_ready = mmu_was_ready; 6112 out: 6113 mutex_unlock(&kvm->arch.mmu_setup_lock); 6114 return ret; 6115 } 6116 6117 static int kvmhv_enable_dawr1(struct kvm *kvm) 6118 { 6119 if (!cpu_has_feature(CPU_FTR_DAWR1)) 6120 return -ENODEV; 6121 6122 /* kvm == NULL means the caller is testing if the capability exists */ 6123 if (kvm) 6124 kvm->arch.dawr1_enabled = true; 6125 return 0; 6126 } 6127 6128 static bool kvmppc_hash_v3_possible(void) 6129 { 6130 if (!cpu_has_feature(CPU_FTR_ARCH_300)) 6131 return false; 6132 6133 if (!cpu_has_feature(CPU_FTR_HVMODE)) 6134 return false; 6135 6136 /* 6137 * POWER9 chips before version 2.02 can't have some threads in 6138 * HPT mode and some in radix mode on the same core. 6139 */ 6140 if (radix_enabled()) { 6141 unsigned int pvr = mfspr(SPRN_PVR); 6142 if ((pvr >> 16) == PVR_POWER9 && 6143 (((pvr & 0xe000) == 0 && (pvr & 0xfff) < 0x202) || 6144 ((pvr & 0xe000) == 0x2000 && (pvr & 0xfff) < 0x101))) 6145 return false; 6146 } 6147 6148 return true; 6149 } 6150 6151 static struct kvmppc_ops kvm_ops_hv = { 6152 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv, 6153 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv, 6154 .get_one_reg = kvmppc_get_one_reg_hv, 6155 .set_one_reg = kvmppc_set_one_reg_hv, 6156 .vcpu_load = kvmppc_core_vcpu_load_hv, 6157 .vcpu_put = kvmppc_core_vcpu_put_hv, 6158 .inject_interrupt = kvmppc_inject_interrupt_hv, 6159 .set_msr = kvmppc_set_msr_hv, 6160 .vcpu_run = kvmppc_vcpu_run_hv, 6161 .vcpu_create = kvmppc_core_vcpu_create_hv, 6162 .vcpu_free = kvmppc_core_vcpu_free_hv, 6163 .check_requests = kvmppc_core_check_requests_hv, 6164 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv, 6165 .flush_memslot = kvmppc_core_flush_memslot_hv, 6166 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv, 6167 .commit_memory_region = kvmppc_core_commit_memory_region_hv, 6168 .unmap_gfn_range = kvm_unmap_gfn_range_hv, 6169 .age_gfn = kvm_age_gfn_hv, 6170 .test_age_gfn = kvm_test_age_gfn_hv, 6171 .set_spte_gfn = kvm_set_spte_gfn_hv, 6172 .free_memslot = kvmppc_core_free_memslot_hv, 6173 .init_vm = kvmppc_core_init_vm_hv, 6174 .destroy_vm = kvmppc_core_destroy_vm_hv, 6175 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv, 6176 .emulate_op = kvmppc_core_emulate_op_hv, 6177 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv, 6178 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv, 6179 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv, 6180 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv, 6181 .hcall_implemented = kvmppc_hcall_impl_hv, 6182 #ifdef CONFIG_KVM_XICS 6183 .irq_bypass_add_producer = kvmppc_irq_bypass_add_producer_hv, 6184 .irq_bypass_del_producer = kvmppc_irq_bypass_del_producer_hv, 6185 #endif 6186 .configure_mmu = kvmhv_configure_mmu, 6187 .get_rmmu_info = kvmhv_get_rmmu_info, 6188 .set_smt_mode = kvmhv_set_smt_mode, 6189 .enable_nested = kvmhv_enable_nested, 6190 .load_from_eaddr = kvmhv_load_from_eaddr, 6191 .store_to_eaddr = kvmhv_store_to_eaddr, 6192 .enable_svm = kvmhv_enable_svm, 6193 .svm_off = kvmhv_svm_off, 6194 .enable_dawr1 = kvmhv_enable_dawr1, 6195 .hash_v3_possible = kvmppc_hash_v3_possible, 6196 .create_vcpu_debugfs = kvmppc_arch_create_vcpu_debugfs_hv, 6197 .create_vm_debugfs = kvmppc_arch_create_vm_debugfs_hv, 6198 }; 6199 6200 static int kvm_init_subcore_bitmap(void) 6201 { 6202 int i, j; 6203 int nr_cores = cpu_nr_cores(); 6204 struct sibling_subcore_state *sibling_subcore_state; 6205 6206 for (i = 0; i < nr_cores; i++) { 6207 int first_cpu = i * threads_per_core; 6208 int node = cpu_to_node(first_cpu); 6209 6210 /* Ignore if it is already allocated. */ 6211 if (paca_ptrs[first_cpu]->sibling_subcore_state) 6212 continue; 6213 6214 sibling_subcore_state = 6215 kzalloc_node(sizeof(struct sibling_subcore_state), 6216 GFP_KERNEL, node); 6217 if (!sibling_subcore_state) 6218 return -ENOMEM; 6219 6220 6221 for (j = 0; j < threads_per_core; j++) { 6222 int cpu = first_cpu + j; 6223 6224 paca_ptrs[cpu]->sibling_subcore_state = 6225 sibling_subcore_state; 6226 } 6227 } 6228 return 0; 6229 } 6230 6231 static int kvmppc_radix_possible(void) 6232 { 6233 return cpu_has_feature(CPU_FTR_ARCH_300) && radix_enabled(); 6234 } 6235 6236 static int kvmppc_book3s_init_hv(void) 6237 { 6238 int r; 6239 6240 if (!tlbie_capable) { 6241 pr_err("KVM-HV: Host does not support TLBIE\n"); 6242 return -ENODEV; 6243 } 6244 6245 /* 6246 * FIXME!! Do we need to check on all cpus ? 6247 */ 6248 r = kvmppc_core_check_processor_compat_hv(); 6249 if (r < 0) 6250 return -ENODEV; 6251 6252 r = kvmhv_nested_init(); 6253 if (r) 6254 return r; 6255 6256 if (!cpu_has_feature(CPU_FTR_ARCH_300)) { 6257 r = kvm_init_subcore_bitmap(); 6258 if (r) 6259 goto err; 6260 } 6261 6262 /* 6263 * We need a way of accessing the XICS interrupt controller, 6264 * either directly, via paca_ptrs[cpu]->kvm_hstate.xics_phys, or 6265 * indirectly, via OPAL. 6266 */ 6267 #ifdef CONFIG_SMP 6268 if (!xics_on_xive() && !kvmhv_on_pseries() && 6269 !local_paca->kvm_hstate.xics_phys) { 6270 struct device_node *np; 6271 6272 np = of_find_compatible_node(NULL, NULL, "ibm,opal-intc"); 6273 if (!np) { 6274 pr_err("KVM-HV: Cannot determine method for accessing XICS\n"); 6275 r = -ENODEV; 6276 goto err; 6277 } 6278 /* presence of intc confirmed - node can be dropped again */ 6279 of_node_put(np); 6280 } 6281 #endif 6282 6283 init_default_hcalls(); 6284 6285 init_vcore_lists(); 6286 6287 r = kvmppc_mmu_hv_init(); 6288 if (r) 6289 goto err; 6290 6291 if (kvmppc_radix_possible()) { 6292 r = kvmppc_radix_init(); 6293 if (r) 6294 goto err; 6295 } 6296 6297 r = kvmppc_uvmem_init(); 6298 if (r < 0) { 6299 pr_err("KVM-HV: kvmppc_uvmem_init failed %d\n", r); 6300 return r; 6301 } 6302 6303 kvm_ops_hv.owner = THIS_MODULE; 6304 kvmppc_hv_ops = &kvm_ops_hv; 6305 6306 return 0; 6307 6308 err: 6309 kvmhv_nested_exit(); 6310 kvmppc_radix_exit(); 6311 6312 return r; 6313 } 6314 6315 static void kvmppc_book3s_exit_hv(void) 6316 { 6317 kvmppc_uvmem_free(); 6318 kvmppc_free_host_rm_ops(); 6319 if (kvmppc_radix_possible()) 6320 kvmppc_radix_exit(); 6321 kvmppc_hv_ops = NULL; 6322 kvmhv_nested_exit(); 6323 } 6324 6325 module_init(kvmppc_book3s_init_hv); 6326 module_exit(kvmppc_book3s_exit_hv); 6327 MODULE_LICENSE("GPL"); 6328 MODULE_ALIAS_MISCDEV(KVM_MINOR); 6329 MODULE_ALIAS("devname:kvm"); 6330