1 /* 2 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com> 3 * Copyright (C) 2009. SUSE Linux Products GmbH. All rights reserved. 4 * 5 * Authors: 6 * Paul Mackerras <paulus@au1.ibm.com> 7 * Alexander Graf <agraf@suse.de> 8 * Kevin Wolf <mail@kevin-wolf.de> 9 * 10 * Description: KVM functions specific to running on Book 3S 11 * processors in hypervisor mode (specifically POWER7 and later). 12 * 13 * This file is derived from arch/powerpc/kvm/book3s.c, 14 * by Alexander Graf <agraf@suse.de>. 15 * 16 * This program is free software; you can redistribute it and/or modify 17 * it under the terms of the GNU General Public License, version 2, as 18 * published by the Free Software Foundation. 19 */ 20 21 #include <linux/kvm_host.h> 22 #include <linux/err.h> 23 #include <linux/slab.h> 24 #include <linux/preempt.h> 25 #include <linux/sched.h> 26 #include <linux/delay.h> 27 #include <linux/export.h> 28 #include <linux/fs.h> 29 #include <linux/anon_inodes.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 36 #include <asm/reg.h> 37 #include <asm/cputable.h> 38 #include <asm/cacheflush.h> 39 #include <asm/tlbflush.h> 40 #include <asm/uaccess.h> 41 #include <asm/io.h> 42 #include <asm/kvm_ppc.h> 43 #include <asm/kvm_book3s.h> 44 #include <asm/mmu_context.h> 45 #include <asm/lppaca.h> 46 #include <asm/processor.h> 47 #include <asm/cputhreads.h> 48 #include <asm/page.h> 49 #include <asm/hvcall.h> 50 #include <asm/switch_to.h> 51 #include <asm/smp.h> 52 #include <linux/gfp.h> 53 #include <linux/vmalloc.h> 54 #include <linux/highmem.h> 55 #include <linux/hugetlb.h> 56 #include <linux/module.h> 57 58 #include "book3s.h" 59 60 /* #define EXIT_DEBUG */ 61 /* #define EXIT_DEBUG_SIMPLE */ 62 /* #define EXIT_DEBUG_INT */ 63 64 /* Used to indicate that a guest page fault needs to be handled */ 65 #define RESUME_PAGE_FAULT (RESUME_GUEST | RESUME_FLAG_ARCH1) 66 67 /* Used as a "null" value for timebase values */ 68 #define TB_NIL (~(u64)0) 69 70 static void kvmppc_end_cede(struct kvm_vcpu *vcpu); 71 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu); 72 73 static void kvmppc_fast_vcpu_kick_hv(struct kvm_vcpu *vcpu) 74 { 75 int me; 76 int cpu = vcpu->cpu; 77 wait_queue_head_t *wqp; 78 79 wqp = kvm_arch_vcpu_wq(vcpu); 80 if (waitqueue_active(wqp)) { 81 wake_up_interruptible(wqp); 82 ++vcpu->stat.halt_wakeup; 83 } 84 85 me = get_cpu(); 86 87 /* CPU points to the first thread of the core */ 88 if (cpu != me && cpu >= 0 && cpu < nr_cpu_ids) { 89 #ifdef CONFIG_PPC_ICP_NATIVE 90 int real_cpu = cpu + vcpu->arch.ptid; 91 if (paca[real_cpu].kvm_hstate.xics_phys) 92 xics_wake_cpu(real_cpu); 93 else 94 #endif 95 if (cpu_online(cpu)) 96 smp_send_reschedule(cpu); 97 } 98 put_cpu(); 99 } 100 101 /* 102 * We use the vcpu_load/put functions to measure stolen time. 103 * Stolen time is counted as time when either the vcpu is able to 104 * run as part of a virtual core, but the task running the vcore 105 * is preempted or sleeping, or when the vcpu needs something done 106 * in the kernel by the task running the vcpu, but that task is 107 * preempted or sleeping. Those two things have to be counted 108 * separately, since one of the vcpu tasks will take on the job 109 * of running the core, and the other vcpu tasks in the vcore will 110 * sleep waiting for it to do that, but that sleep shouldn't count 111 * as stolen time. 112 * 113 * Hence we accumulate stolen time when the vcpu can run as part of 114 * a vcore using vc->stolen_tb, and the stolen time when the vcpu 115 * needs its task to do other things in the kernel (for example, 116 * service a page fault) in busy_stolen. We don't accumulate 117 * stolen time for a vcore when it is inactive, or for a vcpu 118 * when it is in state RUNNING or NOTREADY. NOTREADY is a bit of 119 * a misnomer; it means that the vcpu task is not executing in 120 * the KVM_VCPU_RUN ioctl, i.e. it is in userspace or elsewhere in 121 * the kernel. We don't have any way of dividing up that time 122 * between time that the vcpu is genuinely stopped, time that 123 * the task is actively working on behalf of the vcpu, and time 124 * that the task is preempted, so we don't count any of it as 125 * stolen. 126 * 127 * Updates to busy_stolen are protected by arch.tbacct_lock; 128 * updates to vc->stolen_tb are protected by the arch.tbacct_lock 129 * of the vcpu that has taken responsibility for running the vcore 130 * (i.e. vc->runner). The stolen times are measured in units of 131 * timebase ticks. (Note that the != TB_NIL checks below are 132 * purely defensive; they should never fail.) 133 */ 134 135 static void kvmppc_core_vcpu_load_hv(struct kvm_vcpu *vcpu, int cpu) 136 { 137 struct kvmppc_vcore *vc = vcpu->arch.vcore; 138 unsigned long flags; 139 140 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags); 141 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE && 142 vc->preempt_tb != TB_NIL) { 143 vc->stolen_tb += mftb() - vc->preempt_tb; 144 vc->preempt_tb = TB_NIL; 145 } 146 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST && 147 vcpu->arch.busy_preempt != TB_NIL) { 148 vcpu->arch.busy_stolen += mftb() - vcpu->arch.busy_preempt; 149 vcpu->arch.busy_preempt = TB_NIL; 150 } 151 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags); 152 } 153 154 static void kvmppc_core_vcpu_put_hv(struct kvm_vcpu *vcpu) 155 { 156 struct kvmppc_vcore *vc = vcpu->arch.vcore; 157 unsigned long flags; 158 159 spin_lock_irqsave(&vcpu->arch.tbacct_lock, flags); 160 if (vc->runner == vcpu && vc->vcore_state != VCORE_INACTIVE) 161 vc->preempt_tb = mftb(); 162 if (vcpu->arch.state == KVMPPC_VCPU_BUSY_IN_HOST) 163 vcpu->arch.busy_preempt = mftb(); 164 spin_unlock_irqrestore(&vcpu->arch.tbacct_lock, flags); 165 } 166 167 static void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr) 168 { 169 vcpu->arch.shregs.msr = msr; 170 kvmppc_end_cede(vcpu); 171 } 172 173 void kvmppc_set_pvr_hv(struct kvm_vcpu *vcpu, u32 pvr) 174 { 175 vcpu->arch.pvr = pvr; 176 } 177 178 int kvmppc_set_arch_compat(struct kvm_vcpu *vcpu, u32 arch_compat) 179 { 180 unsigned long pcr = 0; 181 struct kvmppc_vcore *vc = vcpu->arch.vcore; 182 183 if (arch_compat) { 184 if (!cpu_has_feature(CPU_FTR_ARCH_206)) 185 return -EINVAL; /* 970 has no compat mode support */ 186 187 switch (arch_compat) { 188 case PVR_ARCH_205: 189 /* 190 * If an arch bit is set in PCR, all the defined 191 * higher-order arch bits also have to be set. 192 */ 193 pcr = PCR_ARCH_206 | PCR_ARCH_205; 194 break; 195 case PVR_ARCH_206: 196 case PVR_ARCH_206p: 197 pcr = PCR_ARCH_206; 198 break; 199 case PVR_ARCH_207: 200 break; 201 default: 202 return -EINVAL; 203 } 204 205 if (!cpu_has_feature(CPU_FTR_ARCH_207S)) { 206 /* POWER7 can't emulate POWER8 */ 207 if (!(pcr & PCR_ARCH_206)) 208 return -EINVAL; 209 pcr &= ~PCR_ARCH_206; 210 } 211 } 212 213 spin_lock(&vc->lock); 214 vc->arch_compat = arch_compat; 215 vc->pcr = pcr; 216 spin_unlock(&vc->lock); 217 218 return 0; 219 } 220 221 void kvmppc_dump_regs(struct kvm_vcpu *vcpu) 222 { 223 int r; 224 225 pr_err("vcpu %p (%d):\n", vcpu, vcpu->vcpu_id); 226 pr_err("pc = %.16lx msr = %.16llx trap = %x\n", 227 vcpu->arch.pc, vcpu->arch.shregs.msr, vcpu->arch.trap); 228 for (r = 0; r < 16; ++r) 229 pr_err("r%2d = %.16lx r%d = %.16lx\n", 230 r, kvmppc_get_gpr(vcpu, r), 231 r+16, kvmppc_get_gpr(vcpu, r+16)); 232 pr_err("ctr = %.16lx lr = %.16lx\n", 233 vcpu->arch.ctr, vcpu->arch.lr); 234 pr_err("srr0 = %.16llx srr1 = %.16llx\n", 235 vcpu->arch.shregs.srr0, vcpu->arch.shregs.srr1); 236 pr_err("sprg0 = %.16llx sprg1 = %.16llx\n", 237 vcpu->arch.shregs.sprg0, vcpu->arch.shregs.sprg1); 238 pr_err("sprg2 = %.16llx sprg3 = %.16llx\n", 239 vcpu->arch.shregs.sprg2, vcpu->arch.shregs.sprg3); 240 pr_err("cr = %.8x xer = %.16lx dsisr = %.8x\n", 241 vcpu->arch.cr, vcpu->arch.xer, vcpu->arch.shregs.dsisr); 242 pr_err("dar = %.16llx\n", vcpu->arch.shregs.dar); 243 pr_err("fault dar = %.16lx dsisr = %.8x\n", 244 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr); 245 pr_err("SLB (%d entries):\n", vcpu->arch.slb_max); 246 for (r = 0; r < vcpu->arch.slb_max; ++r) 247 pr_err(" ESID = %.16llx VSID = %.16llx\n", 248 vcpu->arch.slb[r].orige, vcpu->arch.slb[r].origv); 249 pr_err("lpcr = %.16lx sdr1 = %.16lx last_inst = %.8x\n", 250 vcpu->arch.vcore->lpcr, vcpu->kvm->arch.sdr1, 251 vcpu->arch.last_inst); 252 } 253 254 struct kvm_vcpu *kvmppc_find_vcpu(struct kvm *kvm, int id) 255 { 256 int r; 257 struct kvm_vcpu *v, *ret = NULL; 258 259 mutex_lock(&kvm->lock); 260 kvm_for_each_vcpu(r, v, kvm) { 261 if (v->vcpu_id == id) { 262 ret = v; 263 break; 264 } 265 } 266 mutex_unlock(&kvm->lock); 267 return ret; 268 } 269 270 static void init_vpa(struct kvm_vcpu *vcpu, struct lppaca *vpa) 271 { 272 vpa->__old_status |= LPPACA_OLD_SHARED_PROC; 273 vpa->yield_count = 1; 274 } 275 276 static int set_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *v, 277 unsigned long addr, unsigned long len) 278 { 279 /* check address is cacheline aligned */ 280 if (addr & (L1_CACHE_BYTES - 1)) 281 return -EINVAL; 282 spin_lock(&vcpu->arch.vpa_update_lock); 283 if (v->next_gpa != addr || v->len != len) { 284 v->next_gpa = addr; 285 v->len = addr ? len : 0; 286 v->update_pending = 1; 287 } 288 spin_unlock(&vcpu->arch.vpa_update_lock); 289 return 0; 290 } 291 292 /* Length for a per-processor buffer is passed in at offset 4 in the buffer */ 293 struct reg_vpa { 294 u32 dummy; 295 union { 296 u16 hword; 297 u32 word; 298 } length; 299 }; 300 301 static int vpa_is_registered(struct kvmppc_vpa *vpap) 302 { 303 if (vpap->update_pending) 304 return vpap->next_gpa != 0; 305 return vpap->pinned_addr != NULL; 306 } 307 308 static unsigned long do_h_register_vpa(struct kvm_vcpu *vcpu, 309 unsigned long flags, 310 unsigned long vcpuid, unsigned long vpa) 311 { 312 struct kvm *kvm = vcpu->kvm; 313 unsigned long len, nb; 314 void *va; 315 struct kvm_vcpu *tvcpu; 316 int err; 317 int subfunc; 318 struct kvmppc_vpa *vpap; 319 320 tvcpu = kvmppc_find_vcpu(kvm, vcpuid); 321 if (!tvcpu) 322 return H_PARAMETER; 323 324 subfunc = (flags >> H_VPA_FUNC_SHIFT) & H_VPA_FUNC_MASK; 325 if (subfunc == H_VPA_REG_VPA || subfunc == H_VPA_REG_DTL || 326 subfunc == H_VPA_REG_SLB) { 327 /* Registering new area - address must be cache-line aligned */ 328 if ((vpa & (L1_CACHE_BYTES - 1)) || !vpa) 329 return H_PARAMETER; 330 331 /* convert logical addr to kernel addr and read length */ 332 va = kvmppc_pin_guest_page(kvm, vpa, &nb); 333 if (va == NULL) 334 return H_PARAMETER; 335 if (subfunc == H_VPA_REG_VPA) 336 len = ((struct reg_vpa *)va)->length.hword; 337 else 338 len = ((struct reg_vpa *)va)->length.word; 339 kvmppc_unpin_guest_page(kvm, va, vpa, false); 340 341 /* Check length */ 342 if (len > nb || len < sizeof(struct reg_vpa)) 343 return H_PARAMETER; 344 } else { 345 vpa = 0; 346 len = 0; 347 } 348 349 err = H_PARAMETER; 350 vpap = NULL; 351 spin_lock(&tvcpu->arch.vpa_update_lock); 352 353 switch (subfunc) { 354 case H_VPA_REG_VPA: /* register VPA */ 355 if (len < sizeof(struct lppaca)) 356 break; 357 vpap = &tvcpu->arch.vpa; 358 err = 0; 359 break; 360 361 case H_VPA_REG_DTL: /* register DTL */ 362 if (len < sizeof(struct dtl_entry)) 363 break; 364 len -= len % sizeof(struct dtl_entry); 365 366 /* Check that they have previously registered a VPA */ 367 err = H_RESOURCE; 368 if (!vpa_is_registered(&tvcpu->arch.vpa)) 369 break; 370 371 vpap = &tvcpu->arch.dtl; 372 err = 0; 373 break; 374 375 case H_VPA_REG_SLB: /* register SLB shadow buffer */ 376 /* Check that they have previously registered a VPA */ 377 err = H_RESOURCE; 378 if (!vpa_is_registered(&tvcpu->arch.vpa)) 379 break; 380 381 vpap = &tvcpu->arch.slb_shadow; 382 err = 0; 383 break; 384 385 case H_VPA_DEREG_VPA: /* deregister VPA */ 386 /* Check they don't still have a DTL or SLB buf registered */ 387 err = H_RESOURCE; 388 if (vpa_is_registered(&tvcpu->arch.dtl) || 389 vpa_is_registered(&tvcpu->arch.slb_shadow)) 390 break; 391 392 vpap = &tvcpu->arch.vpa; 393 err = 0; 394 break; 395 396 case H_VPA_DEREG_DTL: /* deregister DTL */ 397 vpap = &tvcpu->arch.dtl; 398 err = 0; 399 break; 400 401 case H_VPA_DEREG_SLB: /* deregister SLB shadow buffer */ 402 vpap = &tvcpu->arch.slb_shadow; 403 err = 0; 404 break; 405 } 406 407 if (vpap) { 408 vpap->next_gpa = vpa; 409 vpap->len = len; 410 vpap->update_pending = 1; 411 } 412 413 spin_unlock(&tvcpu->arch.vpa_update_lock); 414 415 return err; 416 } 417 418 static void kvmppc_update_vpa(struct kvm_vcpu *vcpu, struct kvmppc_vpa *vpap) 419 { 420 struct kvm *kvm = vcpu->kvm; 421 void *va; 422 unsigned long nb; 423 unsigned long gpa; 424 425 /* 426 * We need to pin the page pointed to by vpap->next_gpa, 427 * but we can't call kvmppc_pin_guest_page under the lock 428 * as it does get_user_pages() and down_read(). So we 429 * have to drop the lock, pin the page, then get the lock 430 * again and check that a new area didn't get registered 431 * in the meantime. 432 */ 433 for (;;) { 434 gpa = vpap->next_gpa; 435 spin_unlock(&vcpu->arch.vpa_update_lock); 436 va = NULL; 437 nb = 0; 438 if (gpa) 439 va = kvmppc_pin_guest_page(kvm, gpa, &nb); 440 spin_lock(&vcpu->arch.vpa_update_lock); 441 if (gpa == vpap->next_gpa) 442 break; 443 /* sigh... unpin that one and try again */ 444 if (va) 445 kvmppc_unpin_guest_page(kvm, va, gpa, false); 446 } 447 448 vpap->update_pending = 0; 449 if (va && nb < vpap->len) { 450 /* 451 * If it's now too short, it must be that userspace 452 * has changed the mappings underlying guest memory, 453 * so unregister the region. 454 */ 455 kvmppc_unpin_guest_page(kvm, va, gpa, false); 456 va = NULL; 457 } 458 if (vpap->pinned_addr) 459 kvmppc_unpin_guest_page(kvm, vpap->pinned_addr, vpap->gpa, 460 vpap->dirty); 461 vpap->gpa = gpa; 462 vpap->pinned_addr = va; 463 vpap->dirty = false; 464 if (va) 465 vpap->pinned_end = va + vpap->len; 466 } 467 468 static void kvmppc_update_vpas(struct kvm_vcpu *vcpu) 469 { 470 if (!(vcpu->arch.vpa.update_pending || 471 vcpu->arch.slb_shadow.update_pending || 472 vcpu->arch.dtl.update_pending)) 473 return; 474 475 spin_lock(&vcpu->arch.vpa_update_lock); 476 if (vcpu->arch.vpa.update_pending) { 477 kvmppc_update_vpa(vcpu, &vcpu->arch.vpa); 478 if (vcpu->arch.vpa.pinned_addr) 479 init_vpa(vcpu, vcpu->arch.vpa.pinned_addr); 480 } 481 if (vcpu->arch.dtl.update_pending) { 482 kvmppc_update_vpa(vcpu, &vcpu->arch.dtl); 483 vcpu->arch.dtl_ptr = vcpu->arch.dtl.pinned_addr; 484 vcpu->arch.dtl_index = 0; 485 } 486 if (vcpu->arch.slb_shadow.update_pending) 487 kvmppc_update_vpa(vcpu, &vcpu->arch.slb_shadow); 488 spin_unlock(&vcpu->arch.vpa_update_lock); 489 } 490 491 /* 492 * Return the accumulated stolen time for the vcore up until `now'. 493 * The caller should hold the vcore lock. 494 */ 495 static u64 vcore_stolen_time(struct kvmppc_vcore *vc, u64 now) 496 { 497 u64 p; 498 499 /* 500 * If we are the task running the vcore, then since we hold 501 * the vcore lock, we can't be preempted, so stolen_tb/preempt_tb 502 * can't be updated, so we don't need the tbacct_lock. 503 * If the vcore is inactive, it can't become active (since we 504 * hold the vcore lock), so the vcpu load/put functions won't 505 * update stolen_tb/preempt_tb, and we don't need tbacct_lock. 506 */ 507 if (vc->vcore_state != VCORE_INACTIVE && 508 vc->runner->arch.run_task != current) { 509 spin_lock_irq(&vc->runner->arch.tbacct_lock); 510 p = vc->stolen_tb; 511 if (vc->preempt_tb != TB_NIL) 512 p += now - vc->preempt_tb; 513 spin_unlock_irq(&vc->runner->arch.tbacct_lock); 514 } else { 515 p = vc->stolen_tb; 516 } 517 return p; 518 } 519 520 static void kvmppc_create_dtl_entry(struct kvm_vcpu *vcpu, 521 struct kvmppc_vcore *vc) 522 { 523 struct dtl_entry *dt; 524 struct lppaca *vpa; 525 unsigned long stolen; 526 unsigned long core_stolen; 527 u64 now; 528 529 dt = vcpu->arch.dtl_ptr; 530 vpa = vcpu->arch.vpa.pinned_addr; 531 now = mftb(); 532 core_stolen = vcore_stolen_time(vc, now); 533 stolen = core_stolen - vcpu->arch.stolen_logged; 534 vcpu->arch.stolen_logged = core_stolen; 535 spin_lock_irq(&vcpu->arch.tbacct_lock); 536 stolen += vcpu->arch.busy_stolen; 537 vcpu->arch.busy_stolen = 0; 538 spin_unlock_irq(&vcpu->arch.tbacct_lock); 539 if (!dt || !vpa) 540 return; 541 memset(dt, 0, sizeof(struct dtl_entry)); 542 dt->dispatch_reason = 7; 543 dt->processor_id = vc->pcpu + vcpu->arch.ptid; 544 dt->timebase = now + vc->tb_offset; 545 dt->enqueue_to_dispatch_time = stolen; 546 dt->srr0 = kvmppc_get_pc(vcpu); 547 dt->srr1 = vcpu->arch.shregs.msr; 548 ++dt; 549 if (dt == vcpu->arch.dtl.pinned_end) 550 dt = vcpu->arch.dtl.pinned_addr; 551 vcpu->arch.dtl_ptr = dt; 552 /* order writing *dt vs. writing vpa->dtl_idx */ 553 smp_wmb(); 554 vpa->dtl_idx = ++vcpu->arch.dtl_index; 555 vcpu->arch.dtl.dirty = true; 556 } 557 558 int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu) 559 { 560 unsigned long req = kvmppc_get_gpr(vcpu, 3); 561 unsigned long target, ret = H_SUCCESS; 562 struct kvm_vcpu *tvcpu; 563 int idx, rc; 564 565 switch (req) { 566 case H_ENTER: 567 idx = srcu_read_lock(&vcpu->kvm->srcu); 568 ret = kvmppc_virtmode_h_enter(vcpu, kvmppc_get_gpr(vcpu, 4), 569 kvmppc_get_gpr(vcpu, 5), 570 kvmppc_get_gpr(vcpu, 6), 571 kvmppc_get_gpr(vcpu, 7)); 572 srcu_read_unlock(&vcpu->kvm->srcu, idx); 573 break; 574 case H_CEDE: 575 break; 576 case H_PROD: 577 target = kvmppc_get_gpr(vcpu, 4); 578 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target); 579 if (!tvcpu) { 580 ret = H_PARAMETER; 581 break; 582 } 583 tvcpu->arch.prodded = 1; 584 smp_mb(); 585 if (vcpu->arch.ceded) { 586 if (waitqueue_active(&vcpu->wq)) { 587 wake_up_interruptible(&vcpu->wq); 588 vcpu->stat.halt_wakeup++; 589 } 590 } 591 break; 592 case H_CONFER: 593 target = kvmppc_get_gpr(vcpu, 4); 594 if (target == -1) 595 break; 596 tvcpu = kvmppc_find_vcpu(vcpu->kvm, target); 597 if (!tvcpu) { 598 ret = H_PARAMETER; 599 break; 600 } 601 kvm_vcpu_yield_to(tvcpu); 602 break; 603 case H_REGISTER_VPA: 604 ret = do_h_register_vpa(vcpu, kvmppc_get_gpr(vcpu, 4), 605 kvmppc_get_gpr(vcpu, 5), 606 kvmppc_get_gpr(vcpu, 6)); 607 break; 608 case H_RTAS: 609 if (list_empty(&vcpu->kvm->arch.rtas_tokens)) 610 return RESUME_HOST; 611 612 idx = srcu_read_lock(&vcpu->kvm->srcu); 613 rc = kvmppc_rtas_hcall(vcpu); 614 srcu_read_unlock(&vcpu->kvm->srcu, idx); 615 616 if (rc == -ENOENT) 617 return RESUME_HOST; 618 else if (rc == 0) 619 break; 620 621 /* Send the error out to userspace via KVM_RUN */ 622 return rc; 623 624 case H_XIRR: 625 case H_CPPR: 626 case H_EOI: 627 case H_IPI: 628 case H_IPOLL: 629 case H_XIRR_X: 630 if (kvmppc_xics_enabled(vcpu)) { 631 ret = kvmppc_xics_hcall(vcpu, req); 632 break; 633 } /* fallthrough */ 634 default: 635 return RESUME_HOST; 636 } 637 kvmppc_set_gpr(vcpu, 3, ret); 638 vcpu->arch.hcall_needed = 0; 639 return RESUME_GUEST; 640 } 641 642 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu, 643 struct task_struct *tsk) 644 { 645 int r = RESUME_HOST; 646 647 vcpu->stat.sum_exits++; 648 649 run->exit_reason = KVM_EXIT_UNKNOWN; 650 run->ready_for_interrupt_injection = 1; 651 switch (vcpu->arch.trap) { 652 /* We're good on these - the host merely wanted to get our attention */ 653 case BOOK3S_INTERRUPT_HV_DECREMENTER: 654 vcpu->stat.dec_exits++; 655 r = RESUME_GUEST; 656 break; 657 case BOOK3S_INTERRUPT_EXTERNAL: 658 case BOOK3S_INTERRUPT_H_DOORBELL: 659 vcpu->stat.ext_intr_exits++; 660 r = RESUME_GUEST; 661 break; 662 case BOOK3S_INTERRUPT_PERFMON: 663 r = RESUME_GUEST; 664 break; 665 case BOOK3S_INTERRUPT_MACHINE_CHECK: 666 /* 667 * Deliver a machine check interrupt to the guest. 668 * We have to do this, even if the host has handled the 669 * machine check, because machine checks use SRR0/1 and 670 * the interrupt might have trashed guest state in them. 671 */ 672 kvmppc_book3s_queue_irqprio(vcpu, 673 BOOK3S_INTERRUPT_MACHINE_CHECK); 674 r = RESUME_GUEST; 675 break; 676 case BOOK3S_INTERRUPT_PROGRAM: 677 { 678 ulong flags; 679 /* 680 * Normally program interrupts are delivered directly 681 * to the guest by the hardware, but we can get here 682 * as a result of a hypervisor emulation interrupt 683 * (e40) getting turned into a 700 by BML RTAS. 684 */ 685 flags = vcpu->arch.shregs.msr & 0x1f0000ull; 686 kvmppc_core_queue_program(vcpu, flags); 687 r = RESUME_GUEST; 688 break; 689 } 690 case BOOK3S_INTERRUPT_SYSCALL: 691 { 692 /* hcall - punt to userspace */ 693 int i; 694 695 /* hypercall with MSR_PR has already been handled in rmode, 696 * and never reaches here. 697 */ 698 699 run->papr_hcall.nr = kvmppc_get_gpr(vcpu, 3); 700 for (i = 0; i < 9; ++i) 701 run->papr_hcall.args[i] = kvmppc_get_gpr(vcpu, 4 + i); 702 run->exit_reason = KVM_EXIT_PAPR_HCALL; 703 vcpu->arch.hcall_needed = 1; 704 r = RESUME_HOST; 705 break; 706 } 707 /* 708 * We get these next two if the guest accesses a page which it thinks 709 * it has mapped but which is not actually present, either because 710 * it is for an emulated I/O device or because the corresonding 711 * host page has been paged out. Any other HDSI/HISI interrupts 712 * have been handled already. 713 */ 714 case BOOK3S_INTERRUPT_H_DATA_STORAGE: 715 r = RESUME_PAGE_FAULT; 716 break; 717 case BOOK3S_INTERRUPT_H_INST_STORAGE: 718 vcpu->arch.fault_dar = kvmppc_get_pc(vcpu); 719 vcpu->arch.fault_dsisr = 0; 720 r = RESUME_PAGE_FAULT; 721 break; 722 /* 723 * This occurs if the guest executes an illegal instruction. 724 * We just generate a program interrupt to the guest, since 725 * we don't emulate any guest instructions at this stage. 726 */ 727 case BOOK3S_INTERRUPT_H_EMUL_ASSIST: 728 kvmppc_core_queue_program(vcpu, SRR1_PROGILL); 729 r = RESUME_GUEST; 730 break; 731 /* 732 * This occurs if the guest (kernel or userspace), does something that 733 * is prohibited by HFSCR. We just generate a program interrupt to 734 * the guest. 735 */ 736 case BOOK3S_INTERRUPT_H_FAC_UNAVAIL: 737 kvmppc_core_queue_program(vcpu, SRR1_PROGILL); 738 r = RESUME_GUEST; 739 break; 740 default: 741 kvmppc_dump_regs(vcpu); 742 printk(KERN_EMERG "trap=0x%x | pc=0x%lx | msr=0x%llx\n", 743 vcpu->arch.trap, kvmppc_get_pc(vcpu), 744 vcpu->arch.shregs.msr); 745 run->hw.hardware_exit_reason = vcpu->arch.trap; 746 r = RESUME_HOST; 747 break; 748 } 749 750 return r; 751 } 752 753 static int kvm_arch_vcpu_ioctl_get_sregs_hv(struct kvm_vcpu *vcpu, 754 struct kvm_sregs *sregs) 755 { 756 int i; 757 758 memset(sregs, 0, sizeof(struct kvm_sregs)); 759 sregs->pvr = vcpu->arch.pvr; 760 for (i = 0; i < vcpu->arch.slb_max; i++) { 761 sregs->u.s.ppc64.slb[i].slbe = vcpu->arch.slb[i].orige; 762 sregs->u.s.ppc64.slb[i].slbv = vcpu->arch.slb[i].origv; 763 } 764 765 return 0; 766 } 767 768 static int kvm_arch_vcpu_ioctl_set_sregs_hv(struct kvm_vcpu *vcpu, 769 struct kvm_sregs *sregs) 770 { 771 int i, j; 772 773 kvmppc_set_pvr_hv(vcpu, sregs->pvr); 774 775 j = 0; 776 for (i = 0; i < vcpu->arch.slb_nr; i++) { 777 if (sregs->u.s.ppc64.slb[i].slbe & SLB_ESID_V) { 778 vcpu->arch.slb[j].orige = sregs->u.s.ppc64.slb[i].slbe; 779 vcpu->arch.slb[j].origv = sregs->u.s.ppc64.slb[i].slbv; 780 ++j; 781 } 782 } 783 vcpu->arch.slb_max = j; 784 785 return 0; 786 } 787 788 static void kvmppc_set_lpcr(struct kvm_vcpu *vcpu, u64 new_lpcr) 789 { 790 struct kvmppc_vcore *vc = vcpu->arch.vcore; 791 u64 mask; 792 793 spin_lock(&vc->lock); 794 /* 795 * If ILE (interrupt little-endian) has changed, update the 796 * MSR_LE bit in the intr_msr for each vcpu in this vcore. 797 */ 798 if ((new_lpcr & LPCR_ILE) != (vc->lpcr & LPCR_ILE)) { 799 struct kvm *kvm = vcpu->kvm; 800 struct kvm_vcpu *vcpu; 801 int i; 802 803 mutex_lock(&kvm->lock); 804 kvm_for_each_vcpu(i, vcpu, kvm) { 805 if (vcpu->arch.vcore != vc) 806 continue; 807 if (new_lpcr & LPCR_ILE) 808 vcpu->arch.intr_msr |= MSR_LE; 809 else 810 vcpu->arch.intr_msr &= ~MSR_LE; 811 } 812 mutex_unlock(&kvm->lock); 813 } 814 815 /* 816 * Userspace can only modify DPFD (default prefetch depth), 817 * ILE (interrupt little-endian) and TC (translation control). 818 * On POWER8 userspace can also modify AIL (alt. interrupt loc.) 819 */ 820 mask = LPCR_DPFD | LPCR_ILE | LPCR_TC; 821 if (cpu_has_feature(CPU_FTR_ARCH_207S)) 822 mask |= LPCR_AIL; 823 vc->lpcr = (vc->lpcr & ~mask) | (new_lpcr & mask); 824 spin_unlock(&vc->lock); 825 } 826 827 static int kvmppc_get_one_reg_hv(struct kvm_vcpu *vcpu, u64 id, 828 union kvmppc_one_reg *val) 829 { 830 int r = 0; 831 long int i; 832 833 switch (id) { 834 case KVM_REG_PPC_HIOR: 835 *val = get_reg_val(id, 0); 836 break; 837 case KVM_REG_PPC_DABR: 838 *val = get_reg_val(id, vcpu->arch.dabr); 839 break; 840 case KVM_REG_PPC_DABRX: 841 *val = get_reg_val(id, vcpu->arch.dabrx); 842 break; 843 case KVM_REG_PPC_DSCR: 844 *val = get_reg_val(id, vcpu->arch.dscr); 845 break; 846 case KVM_REG_PPC_PURR: 847 *val = get_reg_val(id, vcpu->arch.purr); 848 break; 849 case KVM_REG_PPC_SPURR: 850 *val = get_reg_val(id, vcpu->arch.spurr); 851 break; 852 case KVM_REG_PPC_AMR: 853 *val = get_reg_val(id, vcpu->arch.amr); 854 break; 855 case KVM_REG_PPC_UAMOR: 856 *val = get_reg_val(id, vcpu->arch.uamor); 857 break; 858 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS: 859 i = id - KVM_REG_PPC_MMCR0; 860 *val = get_reg_val(id, vcpu->arch.mmcr[i]); 861 break; 862 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8: 863 i = id - KVM_REG_PPC_PMC1; 864 *val = get_reg_val(id, vcpu->arch.pmc[i]); 865 break; 866 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2: 867 i = id - KVM_REG_PPC_SPMC1; 868 *val = get_reg_val(id, vcpu->arch.spmc[i]); 869 break; 870 case KVM_REG_PPC_SIAR: 871 *val = get_reg_val(id, vcpu->arch.siar); 872 break; 873 case KVM_REG_PPC_SDAR: 874 *val = get_reg_val(id, vcpu->arch.sdar); 875 break; 876 case KVM_REG_PPC_SIER: 877 *val = get_reg_val(id, vcpu->arch.sier); 878 break; 879 case KVM_REG_PPC_IAMR: 880 *val = get_reg_val(id, vcpu->arch.iamr); 881 break; 882 case KVM_REG_PPC_PSPB: 883 *val = get_reg_val(id, vcpu->arch.pspb); 884 break; 885 case KVM_REG_PPC_DPDES: 886 *val = get_reg_val(id, vcpu->arch.vcore->dpdes); 887 break; 888 case KVM_REG_PPC_DAWR: 889 *val = get_reg_val(id, vcpu->arch.dawr); 890 break; 891 case KVM_REG_PPC_DAWRX: 892 *val = get_reg_val(id, vcpu->arch.dawrx); 893 break; 894 case KVM_REG_PPC_CIABR: 895 *val = get_reg_val(id, vcpu->arch.ciabr); 896 break; 897 case KVM_REG_PPC_IC: 898 *val = get_reg_val(id, vcpu->arch.ic); 899 break; 900 case KVM_REG_PPC_VTB: 901 *val = get_reg_val(id, vcpu->arch.vtb); 902 break; 903 case KVM_REG_PPC_CSIGR: 904 *val = get_reg_val(id, vcpu->arch.csigr); 905 break; 906 case KVM_REG_PPC_TACR: 907 *val = get_reg_val(id, vcpu->arch.tacr); 908 break; 909 case KVM_REG_PPC_TCSCR: 910 *val = get_reg_val(id, vcpu->arch.tcscr); 911 break; 912 case KVM_REG_PPC_PID: 913 *val = get_reg_val(id, vcpu->arch.pid); 914 break; 915 case KVM_REG_PPC_ACOP: 916 *val = get_reg_val(id, vcpu->arch.acop); 917 break; 918 case KVM_REG_PPC_WORT: 919 *val = get_reg_val(id, vcpu->arch.wort); 920 break; 921 case KVM_REG_PPC_VPA_ADDR: 922 spin_lock(&vcpu->arch.vpa_update_lock); 923 *val = get_reg_val(id, vcpu->arch.vpa.next_gpa); 924 spin_unlock(&vcpu->arch.vpa_update_lock); 925 break; 926 case KVM_REG_PPC_VPA_SLB: 927 spin_lock(&vcpu->arch.vpa_update_lock); 928 val->vpaval.addr = vcpu->arch.slb_shadow.next_gpa; 929 val->vpaval.length = vcpu->arch.slb_shadow.len; 930 spin_unlock(&vcpu->arch.vpa_update_lock); 931 break; 932 case KVM_REG_PPC_VPA_DTL: 933 spin_lock(&vcpu->arch.vpa_update_lock); 934 val->vpaval.addr = vcpu->arch.dtl.next_gpa; 935 val->vpaval.length = vcpu->arch.dtl.len; 936 spin_unlock(&vcpu->arch.vpa_update_lock); 937 break; 938 case KVM_REG_PPC_TB_OFFSET: 939 *val = get_reg_val(id, vcpu->arch.vcore->tb_offset); 940 break; 941 case KVM_REG_PPC_LPCR: 942 *val = get_reg_val(id, vcpu->arch.vcore->lpcr); 943 break; 944 case KVM_REG_PPC_PPR: 945 *val = get_reg_val(id, vcpu->arch.ppr); 946 break; 947 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 948 case KVM_REG_PPC_TFHAR: 949 *val = get_reg_val(id, vcpu->arch.tfhar); 950 break; 951 case KVM_REG_PPC_TFIAR: 952 *val = get_reg_val(id, vcpu->arch.tfiar); 953 break; 954 case KVM_REG_PPC_TEXASR: 955 *val = get_reg_val(id, vcpu->arch.texasr); 956 break; 957 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31: 958 i = id - KVM_REG_PPC_TM_GPR0; 959 *val = get_reg_val(id, vcpu->arch.gpr_tm[i]); 960 break; 961 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63: 962 { 963 int j; 964 i = id - KVM_REG_PPC_TM_VSR0; 965 if (i < 32) 966 for (j = 0; j < TS_FPRWIDTH; j++) 967 val->vsxval[j] = vcpu->arch.fp_tm.fpr[i][j]; 968 else { 969 if (cpu_has_feature(CPU_FTR_ALTIVEC)) 970 val->vval = vcpu->arch.vr_tm.vr[i-32]; 971 else 972 r = -ENXIO; 973 } 974 break; 975 } 976 case KVM_REG_PPC_TM_CR: 977 *val = get_reg_val(id, vcpu->arch.cr_tm); 978 break; 979 case KVM_REG_PPC_TM_LR: 980 *val = get_reg_val(id, vcpu->arch.lr_tm); 981 break; 982 case KVM_REG_PPC_TM_CTR: 983 *val = get_reg_val(id, vcpu->arch.ctr_tm); 984 break; 985 case KVM_REG_PPC_TM_FPSCR: 986 *val = get_reg_val(id, vcpu->arch.fp_tm.fpscr); 987 break; 988 case KVM_REG_PPC_TM_AMR: 989 *val = get_reg_val(id, vcpu->arch.amr_tm); 990 break; 991 case KVM_REG_PPC_TM_PPR: 992 *val = get_reg_val(id, vcpu->arch.ppr_tm); 993 break; 994 case KVM_REG_PPC_TM_VRSAVE: 995 *val = get_reg_val(id, vcpu->arch.vrsave_tm); 996 break; 997 case KVM_REG_PPC_TM_VSCR: 998 if (cpu_has_feature(CPU_FTR_ALTIVEC)) 999 *val = get_reg_val(id, vcpu->arch.vr_tm.vscr.u[3]); 1000 else 1001 r = -ENXIO; 1002 break; 1003 case KVM_REG_PPC_TM_DSCR: 1004 *val = get_reg_val(id, vcpu->arch.dscr_tm); 1005 break; 1006 case KVM_REG_PPC_TM_TAR: 1007 *val = get_reg_val(id, vcpu->arch.tar_tm); 1008 break; 1009 #endif 1010 case KVM_REG_PPC_ARCH_COMPAT: 1011 *val = get_reg_val(id, vcpu->arch.vcore->arch_compat); 1012 break; 1013 default: 1014 r = -EINVAL; 1015 break; 1016 } 1017 1018 return r; 1019 } 1020 1021 static int kvmppc_set_one_reg_hv(struct kvm_vcpu *vcpu, u64 id, 1022 union kvmppc_one_reg *val) 1023 { 1024 int r = 0; 1025 long int i; 1026 unsigned long addr, len; 1027 1028 switch (id) { 1029 case KVM_REG_PPC_HIOR: 1030 /* Only allow this to be set to zero */ 1031 if (set_reg_val(id, *val)) 1032 r = -EINVAL; 1033 break; 1034 case KVM_REG_PPC_DABR: 1035 vcpu->arch.dabr = set_reg_val(id, *val); 1036 break; 1037 case KVM_REG_PPC_DABRX: 1038 vcpu->arch.dabrx = set_reg_val(id, *val) & ~DABRX_HYP; 1039 break; 1040 case KVM_REG_PPC_DSCR: 1041 vcpu->arch.dscr = set_reg_val(id, *val); 1042 break; 1043 case KVM_REG_PPC_PURR: 1044 vcpu->arch.purr = set_reg_val(id, *val); 1045 break; 1046 case KVM_REG_PPC_SPURR: 1047 vcpu->arch.spurr = set_reg_val(id, *val); 1048 break; 1049 case KVM_REG_PPC_AMR: 1050 vcpu->arch.amr = set_reg_val(id, *val); 1051 break; 1052 case KVM_REG_PPC_UAMOR: 1053 vcpu->arch.uamor = set_reg_val(id, *val); 1054 break; 1055 case KVM_REG_PPC_MMCR0 ... KVM_REG_PPC_MMCRS: 1056 i = id - KVM_REG_PPC_MMCR0; 1057 vcpu->arch.mmcr[i] = set_reg_val(id, *val); 1058 break; 1059 case KVM_REG_PPC_PMC1 ... KVM_REG_PPC_PMC8: 1060 i = id - KVM_REG_PPC_PMC1; 1061 vcpu->arch.pmc[i] = set_reg_val(id, *val); 1062 break; 1063 case KVM_REG_PPC_SPMC1 ... KVM_REG_PPC_SPMC2: 1064 i = id - KVM_REG_PPC_SPMC1; 1065 vcpu->arch.spmc[i] = set_reg_val(id, *val); 1066 break; 1067 case KVM_REG_PPC_SIAR: 1068 vcpu->arch.siar = set_reg_val(id, *val); 1069 break; 1070 case KVM_REG_PPC_SDAR: 1071 vcpu->arch.sdar = set_reg_val(id, *val); 1072 break; 1073 case KVM_REG_PPC_SIER: 1074 vcpu->arch.sier = set_reg_val(id, *val); 1075 break; 1076 case KVM_REG_PPC_IAMR: 1077 vcpu->arch.iamr = set_reg_val(id, *val); 1078 break; 1079 case KVM_REG_PPC_PSPB: 1080 vcpu->arch.pspb = set_reg_val(id, *val); 1081 break; 1082 case KVM_REG_PPC_DPDES: 1083 vcpu->arch.vcore->dpdes = set_reg_val(id, *val); 1084 break; 1085 case KVM_REG_PPC_DAWR: 1086 vcpu->arch.dawr = set_reg_val(id, *val); 1087 break; 1088 case KVM_REG_PPC_DAWRX: 1089 vcpu->arch.dawrx = set_reg_val(id, *val) & ~DAWRX_HYP; 1090 break; 1091 case KVM_REG_PPC_CIABR: 1092 vcpu->arch.ciabr = set_reg_val(id, *val); 1093 /* Don't allow setting breakpoints in hypervisor code */ 1094 if ((vcpu->arch.ciabr & CIABR_PRIV) == CIABR_PRIV_HYPER) 1095 vcpu->arch.ciabr &= ~CIABR_PRIV; /* disable */ 1096 break; 1097 case KVM_REG_PPC_IC: 1098 vcpu->arch.ic = set_reg_val(id, *val); 1099 break; 1100 case KVM_REG_PPC_VTB: 1101 vcpu->arch.vtb = set_reg_val(id, *val); 1102 break; 1103 case KVM_REG_PPC_CSIGR: 1104 vcpu->arch.csigr = set_reg_val(id, *val); 1105 break; 1106 case KVM_REG_PPC_TACR: 1107 vcpu->arch.tacr = set_reg_val(id, *val); 1108 break; 1109 case KVM_REG_PPC_TCSCR: 1110 vcpu->arch.tcscr = set_reg_val(id, *val); 1111 break; 1112 case KVM_REG_PPC_PID: 1113 vcpu->arch.pid = set_reg_val(id, *val); 1114 break; 1115 case KVM_REG_PPC_ACOP: 1116 vcpu->arch.acop = set_reg_val(id, *val); 1117 break; 1118 case KVM_REG_PPC_WORT: 1119 vcpu->arch.wort = set_reg_val(id, *val); 1120 break; 1121 case KVM_REG_PPC_VPA_ADDR: 1122 addr = set_reg_val(id, *val); 1123 r = -EINVAL; 1124 if (!addr && (vcpu->arch.slb_shadow.next_gpa || 1125 vcpu->arch.dtl.next_gpa)) 1126 break; 1127 r = set_vpa(vcpu, &vcpu->arch.vpa, addr, sizeof(struct lppaca)); 1128 break; 1129 case KVM_REG_PPC_VPA_SLB: 1130 addr = val->vpaval.addr; 1131 len = val->vpaval.length; 1132 r = -EINVAL; 1133 if (addr && !vcpu->arch.vpa.next_gpa) 1134 break; 1135 r = set_vpa(vcpu, &vcpu->arch.slb_shadow, addr, len); 1136 break; 1137 case KVM_REG_PPC_VPA_DTL: 1138 addr = val->vpaval.addr; 1139 len = val->vpaval.length; 1140 r = -EINVAL; 1141 if (addr && (len < sizeof(struct dtl_entry) || 1142 !vcpu->arch.vpa.next_gpa)) 1143 break; 1144 len -= len % sizeof(struct dtl_entry); 1145 r = set_vpa(vcpu, &vcpu->arch.dtl, addr, len); 1146 break; 1147 case KVM_REG_PPC_TB_OFFSET: 1148 /* round up to multiple of 2^24 */ 1149 vcpu->arch.vcore->tb_offset = 1150 ALIGN(set_reg_val(id, *val), 1UL << 24); 1151 break; 1152 case KVM_REG_PPC_LPCR: 1153 kvmppc_set_lpcr(vcpu, set_reg_val(id, *val)); 1154 break; 1155 case KVM_REG_PPC_PPR: 1156 vcpu->arch.ppr = set_reg_val(id, *val); 1157 break; 1158 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 1159 case KVM_REG_PPC_TFHAR: 1160 vcpu->arch.tfhar = set_reg_val(id, *val); 1161 break; 1162 case KVM_REG_PPC_TFIAR: 1163 vcpu->arch.tfiar = set_reg_val(id, *val); 1164 break; 1165 case KVM_REG_PPC_TEXASR: 1166 vcpu->arch.texasr = set_reg_val(id, *val); 1167 break; 1168 case KVM_REG_PPC_TM_GPR0 ... KVM_REG_PPC_TM_GPR31: 1169 i = id - KVM_REG_PPC_TM_GPR0; 1170 vcpu->arch.gpr_tm[i] = set_reg_val(id, *val); 1171 break; 1172 case KVM_REG_PPC_TM_VSR0 ... KVM_REG_PPC_TM_VSR63: 1173 { 1174 int j; 1175 i = id - KVM_REG_PPC_TM_VSR0; 1176 if (i < 32) 1177 for (j = 0; j < TS_FPRWIDTH; j++) 1178 vcpu->arch.fp_tm.fpr[i][j] = val->vsxval[j]; 1179 else 1180 if (cpu_has_feature(CPU_FTR_ALTIVEC)) 1181 vcpu->arch.vr_tm.vr[i-32] = val->vval; 1182 else 1183 r = -ENXIO; 1184 break; 1185 } 1186 case KVM_REG_PPC_TM_CR: 1187 vcpu->arch.cr_tm = set_reg_val(id, *val); 1188 break; 1189 case KVM_REG_PPC_TM_LR: 1190 vcpu->arch.lr_tm = set_reg_val(id, *val); 1191 break; 1192 case KVM_REG_PPC_TM_CTR: 1193 vcpu->arch.ctr_tm = set_reg_val(id, *val); 1194 break; 1195 case KVM_REG_PPC_TM_FPSCR: 1196 vcpu->arch.fp_tm.fpscr = set_reg_val(id, *val); 1197 break; 1198 case KVM_REG_PPC_TM_AMR: 1199 vcpu->arch.amr_tm = set_reg_val(id, *val); 1200 break; 1201 case KVM_REG_PPC_TM_PPR: 1202 vcpu->arch.ppr_tm = set_reg_val(id, *val); 1203 break; 1204 case KVM_REG_PPC_TM_VRSAVE: 1205 vcpu->arch.vrsave_tm = set_reg_val(id, *val); 1206 break; 1207 case KVM_REG_PPC_TM_VSCR: 1208 if (cpu_has_feature(CPU_FTR_ALTIVEC)) 1209 vcpu->arch.vr.vscr.u[3] = set_reg_val(id, *val); 1210 else 1211 r = - ENXIO; 1212 break; 1213 case KVM_REG_PPC_TM_DSCR: 1214 vcpu->arch.dscr_tm = set_reg_val(id, *val); 1215 break; 1216 case KVM_REG_PPC_TM_TAR: 1217 vcpu->arch.tar_tm = set_reg_val(id, *val); 1218 break; 1219 #endif 1220 case KVM_REG_PPC_ARCH_COMPAT: 1221 r = kvmppc_set_arch_compat(vcpu, set_reg_val(id, *val)); 1222 break; 1223 default: 1224 r = -EINVAL; 1225 break; 1226 } 1227 1228 return r; 1229 } 1230 1231 static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm, 1232 unsigned int id) 1233 { 1234 struct kvm_vcpu *vcpu; 1235 int err = -EINVAL; 1236 int core; 1237 struct kvmppc_vcore *vcore; 1238 1239 core = id / threads_per_core; 1240 if (core >= KVM_MAX_VCORES) 1241 goto out; 1242 1243 err = -ENOMEM; 1244 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL); 1245 if (!vcpu) 1246 goto out; 1247 1248 err = kvm_vcpu_init(vcpu, kvm, id); 1249 if (err) 1250 goto free_vcpu; 1251 1252 vcpu->arch.shared = &vcpu->arch.shregs; 1253 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE 1254 /* 1255 * The shared struct is never shared on HV, 1256 * so we can always use host endianness 1257 */ 1258 #ifdef __BIG_ENDIAN__ 1259 vcpu->arch.shared_big_endian = true; 1260 #else 1261 vcpu->arch.shared_big_endian = false; 1262 #endif 1263 #endif 1264 vcpu->arch.mmcr[0] = MMCR0_FC; 1265 vcpu->arch.ctrl = CTRL_RUNLATCH; 1266 /* default to host PVR, since we can't spoof it */ 1267 kvmppc_set_pvr_hv(vcpu, mfspr(SPRN_PVR)); 1268 spin_lock_init(&vcpu->arch.vpa_update_lock); 1269 spin_lock_init(&vcpu->arch.tbacct_lock); 1270 vcpu->arch.busy_preempt = TB_NIL; 1271 vcpu->arch.intr_msr = MSR_SF | MSR_ME; 1272 1273 kvmppc_mmu_book3s_hv_init(vcpu); 1274 1275 vcpu->arch.state = KVMPPC_VCPU_NOTREADY; 1276 1277 init_waitqueue_head(&vcpu->arch.cpu_run); 1278 1279 mutex_lock(&kvm->lock); 1280 vcore = kvm->arch.vcores[core]; 1281 if (!vcore) { 1282 vcore = kzalloc(sizeof(struct kvmppc_vcore), GFP_KERNEL); 1283 if (vcore) { 1284 INIT_LIST_HEAD(&vcore->runnable_threads); 1285 spin_lock_init(&vcore->lock); 1286 init_waitqueue_head(&vcore->wq); 1287 vcore->preempt_tb = TB_NIL; 1288 vcore->lpcr = kvm->arch.lpcr; 1289 vcore->first_vcpuid = core * threads_per_core; 1290 vcore->kvm = kvm; 1291 } 1292 kvm->arch.vcores[core] = vcore; 1293 kvm->arch.online_vcores++; 1294 } 1295 mutex_unlock(&kvm->lock); 1296 1297 if (!vcore) 1298 goto free_vcpu; 1299 1300 spin_lock(&vcore->lock); 1301 ++vcore->num_threads; 1302 spin_unlock(&vcore->lock); 1303 vcpu->arch.vcore = vcore; 1304 vcpu->arch.ptid = vcpu->vcpu_id - vcore->first_vcpuid; 1305 1306 vcpu->arch.cpu_type = KVM_CPU_3S_64; 1307 kvmppc_sanity_check(vcpu); 1308 1309 return vcpu; 1310 1311 free_vcpu: 1312 kmem_cache_free(kvm_vcpu_cache, vcpu); 1313 out: 1314 return ERR_PTR(err); 1315 } 1316 1317 static void unpin_vpa(struct kvm *kvm, struct kvmppc_vpa *vpa) 1318 { 1319 if (vpa->pinned_addr) 1320 kvmppc_unpin_guest_page(kvm, vpa->pinned_addr, vpa->gpa, 1321 vpa->dirty); 1322 } 1323 1324 static void kvmppc_core_vcpu_free_hv(struct kvm_vcpu *vcpu) 1325 { 1326 spin_lock(&vcpu->arch.vpa_update_lock); 1327 unpin_vpa(vcpu->kvm, &vcpu->arch.dtl); 1328 unpin_vpa(vcpu->kvm, &vcpu->arch.slb_shadow); 1329 unpin_vpa(vcpu->kvm, &vcpu->arch.vpa); 1330 spin_unlock(&vcpu->arch.vpa_update_lock); 1331 kvm_vcpu_uninit(vcpu); 1332 kmem_cache_free(kvm_vcpu_cache, vcpu); 1333 } 1334 1335 static int kvmppc_core_check_requests_hv(struct kvm_vcpu *vcpu) 1336 { 1337 /* Indicate we want to get back into the guest */ 1338 return 1; 1339 } 1340 1341 static void kvmppc_set_timer(struct kvm_vcpu *vcpu) 1342 { 1343 unsigned long dec_nsec, now; 1344 1345 now = get_tb(); 1346 if (now > vcpu->arch.dec_expires) { 1347 /* decrementer has already gone negative */ 1348 kvmppc_core_queue_dec(vcpu); 1349 kvmppc_core_prepare_to_enter(vcpu); 1350 return; 1351 } 1352 dec_nsec = (vcpu->arch.dec_expires - now) * NSEC_PER_SEC 1353 / tb_ticks_per_sec; 1354 hrtimer_start(&vcpu->arch.dec_timer, ktime_set(0, dec_nsec), 1355 HRTIMER_MODE_REL); 1356 vcpu->arch.timer_running = 1; 1357 } 1358 1359 static void kvmppc_end_cede(struct kvm_vcpu *vcpu) 1360 { 1361 vcpu->arch.ceded = 0; 1362 if (vcpu->arch.timer_running) { 1363 hrtimer_try_to_cancel(&vcpu->arch.dec_timer); 1364 vcpu->arch.timer_running = 0; 1365 } 1366 } 1367 1368 extern void __kvmppc_vcore_entry(void); 1369 1370 static void kvmppc_remove_runnable(struct kvmppc_vcore *vc, 1371 struct kvm_vcpu *vcpu) 1372 { 1373 u64 now; 1374 1375 if (vcpu->arch.state != KVMPPC_VCPU_RUNNABLE) 1376 return; 1377 spin_lock_irq(&vcpu->arch.tbacct_lock); 1378 now = mftb(); 1379 vcpu->arch.busy_stolen += vcore_stolen_time(vc, now) - 1380 vcpu->arch.stolen_logged; 1381 vcpu->arch.busy_preempt = now; 1382 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST; 1383 spin_unlock_irq(&vcpu->arch.tbacct_lock); 1384 --vc->n_runnable; 1385 list_del(&vcpu->arch.run_list); 1386 } 1387 1388 static int kvmppc_grab_hwthread(int cpu) 1389 { 1390 struct paca_struct *tpaca; 1391 long timeout = 1000; 1392 1393 tpaca = &paca[cpu]; 1394 1395 /* Ensure the thread won't go into the kernel if it wakes */ 1396 tpaca->kvm_hstate.hwthread_req = 1; 1397 tpaca->kvm_hstate.kvm_vcpu = NULL; 1398 1399 /* 1400 * If the thread is already executing in the kernel (e.g. handling 1401 * a stray interrupt), wait for it to get back to nap mode. 1402 * The smp_mb() is to ensure that our setting of hwthread_req 1403 * is visible before we look at hwthread_state, so if this 1404 * races with the code at system_reset_pSeries and the thread 1405 * misses our setting of hwthread_req, we are sure to see its 1406 * setting of hwthread_state, and vice versa. 1407 */ 1408 smp_mb(); 1409 while (tpaca->kvm_hstate.hwthread_state == KVM_HWTHREAD_IN_KERNEL) { 1410 if (--timeout <= 0) { 1411 pr_err("KVM: couldn't grab cpu %d\n", cpu); 1412 return -EBUSY; 1413 } 1414 udelay(1); 1415 } 1416 return 0; 1417 } 1418 1419 static void kvmppc_release_hwthread(int cpu) 1420 { 1421 struct paca_struct *tpaca; 1422 1423 tpaca = &paca[cpu]; 1424 tpaca->kvm_hstate.hwthread_req = 0; 1425 tpaca->kvm_hstate.kvm_vcpu = NULL; 1426 } 1427 1428 static void kvmppc_start_thread(struct kvm_vcpu *vcpu) 1429 { 1430 int cpu; 1431 struct paca_struct *tpaca; 1432 struct kvmppc_vcore *vc = vcpu->arch.vcore; 1433 1434 if (vcpu->arch.timer_running) { 1435 hrtimer_try_to_cancel(&vcpu->arch.dec_timer); 1436 vcpu->arch.timer_running = 0; 1437 } 1438 cpu = vc->pcpu + vcpu->arch.ptid; 1439 tpaca = &paca[cpu]; 1440 tpaca->kvm_hstate.kvm_vcpu = vcpu; 1441 tpaca->kvm_hstate.kvm_vcore = vc; 1442 tpaca->kvm_hstate.ptid = vcpu->arch.ptid; 1443 vcpu->cpu = vc->pcpu; 1444 smp_wmb(); 1445 #if defined(CONFIG_PPC_ICP_NATIVE) && defined(CONFIG_SMP) 1446 if (cpu != smp_processor_id()) { 1447 xics_wake_cpu(cpu); 1448 if (vcpu->arch.ptid) 1449 ++vc->n_woken; 1450 } 1451 #endif 1452 } 1453 1454 static void kvmppc_wait_for_nap(struct kvmppc_vcore *vc) 1455 { 1456 int i; 1457 1458 HMT_low(); 1459 i = 0; 1460 while (vc->nap_count < vc->n_woken) { 1461 if (++i >= 1000000) { 1462 pr_err("kvmppc_wait_for_nap timeout %d %d\n", 1463 vc->nap_count, vc->n_woken); 1464 break; 1465 } 1466 cpu_relax(); 1467 } 1468 HMT_medium(); 1469 } 1470 1471 /* 1472 * Check that we are on thread 0 and that any other threads in 1473 * this core are off-line. Then grab the threads so they can't 1474 * enter the kernel. 1475 */ 1476 static int on_primary_thread(void) 1477 { 1478 int cpu = smp_processor_id(); 1479 int thr = cpu_thread_in_core(cpu); 1480 1481 if (thr) 1482 return 0; 1483 while (++thr < threads_per_core) 1484 if (cpu_online(cpu + thr)) 1485 return 0; 1486 1487 /* Grab all hw threads so they can't go into the kernel */ 1488 for (thr = 1; thr < threads_per_core; ++thr) { 1489 if (kvmppc_grab_hwthread(cpu + thr)) { 1490 /* Couldn't grab one; let the others go */ 1491 do { 1492 kvmppc_release_hwthread(cpu + thr); 1493 } while (--thr > 0); 1494 return 0; 1495 } 1496 } 1497 return 1; 1498 } 1499 1500 /* 1501 * Run a set of guest threads on a physical core. 1502 * Called with vc->lock held. 1503 */ 1504 static void kvmppc_run_core(struct kvmppc_vcore *vc) 1505 { 1506 struct kvm_vcpu *vcpu, *vnext; 1507 long ret; 1508 u64 now; 1509 int i, need_vpa_update; 1510 int srcu_idx; 1511 struct kvm_vcpu *vcpus_to_update[threads_per_core]; 1512 1513 /* don't start if any threads have a signal pending */ 1514 need_vpa_update = 0; 1515 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) { 1516 if (signal_pending(vcpu->arch.run_task)) 1517 return; 1518 if (vcpu->arch.vpa.update_pending || 1519 vcpu->arch.slb_shadow.update_pending || 1520 vcpu->arch.dtl.update_pending) 1521 vcpus_to_update[need_vpa_update++] = vcpu; 1522 } 1523 1524 /* 1525 * Initialize *vc, in particular vc->vcore_state, so we can 1526 * drop the vcore lock if necessary. 1527 */ 1528 vc->n_woken = 0; 1529 vc->nap_count = 0; 1530 vc->entry_exit_count = 0; 1531 vc->vcore_state = VCORE_STARTING; 1532 vc->in_guest = 0; 1533 vc->napping_threads = 0; 1534 1535 /* 1536 * Updating any of the vpas requires calling kvmppc_pin_guest_page, 1537 * which can't be called with any spinlocks held. 1538 */ 1539 if (need_vpa_update) { 1540 spin_unlock(&vc->lock); 1541 for (i = 0; i < need_vpa_update; ++i) 1542 kvmppc_update_vpas(vcpus_to_update[i]); 1543 spin_lock(&vc->lock); 1544 } 1545 1546 /* 1547 * Make sure we are running on thread 0, and that 1548 * secondary threads are offline. 1549 */ 1550 if (threads_per_core > 1 && !on_primary_thread()) { 1551 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) 1552 vcpu->arch.ret = -EBUSY; 1553 goto out; 1554 } 1555 1556 vc->pcpu = smp_processor_id(); 1557 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) { 1558 kvmppc_start_thread(vcpu); 1559 kvmppc_create_dtl_entry(vcpu, vc); 1560 } 1561 1562 /* Set this explicitly in case thread 0 doesn't have a vcpu */ 1563 get_paca()->kvm_hstate.kvm_vcore = vc; 1564 get_paca()->kvm_hstate.ptid = 0; 1565 1566 vc->vcore_state = VCORE_RUNNING; 1567 preempt_disable(); 1568 spin_unlock(&vc->lock); 1569 1570 kvm_guest_enter(); 1571 1572 srcu_idx = srcu_read_lock(&vc->kvm->srcu); 1573 1574 __kvmppc_vcore_entry(); 1575 1576 spin_lock(&vc->lock); 1577 /* disable sending of IPIs on virtual external irqs */ 1578 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) 1579 vcpu->cpu = -1; 1580 /* wait for secondary threads to finish writing their state to memory */ 1581 if (vc->nap_count < vc->n_woken) 1582 kvmppc_wait_for_nap(vc); 1583 for (i = 0; i < threads_per_core; ++i) 1584 kvmppc_release_hwthread(vc->pcpu + i); 1585 /* prevent other vcpu threads from doing kvmppc_start_thread() now */ 1586 vc->vcore_state = VCORE_EXITING; 1587 spin_unlock(&vc->lock); 1588 1589 srcu_read_unlock(&vc->kvm->srcu, srcu_idx); 1590 1591 /* make sure updates to secondary vcpu structs are visible now */ 1592 smp_mb(); 1593 kvm_guest_exit(); 1594 1595 preempt_enable(); 1596 cond_resched(); 1597 1598 spin_lock(&vc->lock); 1599 now = get_tb(); 1600 list_for_each_entry(vcpu, &vc->runnable_threads, arch.run_list) { 1601 /* cancel pending dec exception if dec is positive */ 1602 if (now < vcpu->arch.dec_expires && 1603 kvmppc_core_pending_dec(vcpu)) 1604 kvmppc_core_dequeue_dec(vcpu); 1605 1606 ret = RESUME_GUEST; 1607 if (vcpu->arch.trap) 1608 ret = kvmppc_handle_exit_hv(vcpu->arch.kvm_run, vcpu, 1609 vcpu->arch.run_task); 1610 1611 vcpu->arch.ret = ret; 1612 vcpu->arch.trap = 0; 1613 1614 if (vcpu->arch.ceded) { 1615 if (!is_kvmppc_resume_guest(ret)) 1616 kvmppc_end_cede(vcpu); 1617 else 1618 kvmppc_set_timer(vcpu); 1619 } 1620 } 1621 1622 out: 1623 vc->vcore_state = VCORE_INACTIVE; 1624 list_for_each_entry_safe(vcpu, vnext, &vc->runnable_threads, 1625 arch.run_list) { 1626 if (!is_kvmppc_resume_guest(vcpu->arch.ret)) { 1627 kvmppc_remove_runnable(vc, vcpu); 1628 wake_up(&vcpu->arch.cpu_run); 1629 } 1630 } 1631 } 1632 1633 /* 1634 * Wait for some other vcpu thread to execute us, and 1635 * wake us up when we need to handle something in the host. 1636 */ 1637 static void kvmppc_wait_for_exec(struct kvm_vcpu *vcpu, int wait_state) 1638 { 1639 DEFINE_WAIT(wait); 1640 1641 prepare_to_wait(&vcpu->arch.cpu_run, &wait, wait_state); 1642 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) 1643 schedule(); 1644 finish_wait(&vcpu->arch.cpu_run, &wait); 1645 } 1646 1647 /* 1648 * All the vcpus in this vcore are idle, so wait for a decrementer 1649 * or external interrupt to one of the vcpus. vc->lock is held. 1650 */ 1651 static void kvmppc_vcore_blocked(struct kvmppc_vcore *vc) 1652 { 1653 DEFINE_WAIT(wait); 1654 1655 prepare_to_wait(&vc->wq, &wait, TASK_INTERRUPTIBLE); 1656 vc->vcore_state = VCORE_SLEEPING; 1657 spin_unlock(&vc->lock); 1658 schedule(); 1659 finish_wait(&vc->wq, &wait); 1660 spin_lock(&vc->lock); 1661 vc->vcore_state = VCORE_INACTIVE; 1662 } 1663 1664 static int kvmppc_run_vcpu(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu) 1665 { 1666 int n_ceded; 1667 struct kvmppc_vcore *vc; 1668 struct kvm_vcpu *v, *vn; 1669 1670 kvm_run->exit_reason = 0; 1671 vcpu->arch.ret = RESUME_GUEST; 1672 vcpu->arch.trap = 0; 1673 kvmppc_update_vpas(vcpu); 1674 1675 /* 1676 * Synchronize with other threads in this virtual core 1677 */ 1678 vc = vcpu->arch.vcore; 1679 spin_lock(&vc->lock); 1680 vcpu->arch.ceded = 0; 1681 vcpu->arch.run_task = current; 1682 vcpu->arch.kvm_run = kvm_run; 1683 vcpu->arch.stolen_logged = vcore_stolen_time(vc, mftb()); 1684 vcpu->arch.state = KVMPPC_VCPU_RUNNABLE; 1685 vcpu->arch.busy_preempt = TB_NIL; 1686 list_add_tail(&vcpu->arch.run_list, &vc->runnable_threads); 1687 ++vc->n_runnable; 1688 1689 /* 1690 * This happens the first time this is called for a vcpu. 1691 * If the vcore is already running, we may be able to start 1692 * this thread straight away and have it join in. 1693 */ 1694 if (!signal_pending(current)) { 1695 if (vc->vcore_state == VCORE_RUNNING && 1696 VCORE_EXIT_COUNT(vc) == 0) { 1697 kvmppc_create_dtl_entry(vcpu, vc); 1698 kvmppc_start_thread(vcpu); 1699 } else if (vc->vcore_state == VCORE_SLEEPING) { 1700 wake_up(&vc->wq); 1701 } 1702 1703 } 1704 1705 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE && 1706 !signal_pending(current)) { 1707 if (vc->vcore_state != VCORE_INACTIVE) { 1708 spin_unlock(&vc->lock); 1709 kvmppc_wait_for_exec(vcpu, TASK_INTERRUPTIBLE); 1710 spin_lock(&vc->lock); 1711 continue; 1712 } 1713 list_for_each_entry_safe(v, vn, &vc->runnable_threads, 1714 arch.run_list) { 1715 kvmppc_core_prepare_to_enter(v); 1716 if (signal_pending(v->arch.run_task)) { 1717 kvmppc_remove_runnable(vc, v); 1718 v->stat.signal_exits++; 1719 v->arch.kvm_run->exit_reason = KVM_EXIT_INTR; 1720 v->arch.ret = -EINTR; 1721 wake_up(&v->arch.cpu_run); 1722 } 1723 } 1724 if (!vc->n_runnable || vcpu->arch.state != KVMPPC_VCPU_RUNNABLE) 1725 break; 1726 vc->runner = vcpu; 1727 n_ceded = 0; 1728 list_for_each_entry(v, &vc->runnable_threads, arch.run_list) { 1729 if (!v->arch.pending_exceptions) 1730 n_ceded += v->arch.ceded; 1731 else 1732 v->arch.ceded = 0; 1733 } 1734 if (n_ceded == vc->n_runnable) 1735 kvmppc_vcore_blocked(vc); 1736 else 1737 kvmppc_run_core(vc); 1738 vc->runner = NULL; 1739 } 1740 1741 while (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE && 1742 (vc->vcore_state == VCORE_RUNNING || 1743 vc->vcore_state == VCORE_EXITING)) { 1744 spin_unlock(&vc->lock); 1745 kvmppc_wait_for_exec(vcpu, TASK_UNINTERRUPTIBLE); 1746 spin_lock(&vc->lock); 1747 } 1748 1749 if (vcpu->arch.state == KVMPPC_VCPU_RUNNABLE) { 1750 kvmppc_remove_runnable(vc, vcpu); 1751 vcpu->stat.signal_exits++; 1752 kvm_run->exit_reason = KVM_EXIT_INTR; 1753 vcpu->arch.ret = -EINTR; 1754 } 1755 1756 if (vc->n_runnable && vc->vcore_state == VCORE_INACTIVE) { 1757 /* Wake up some vcpu to run the core */ 1758 v = list_first_entry(&vc->runnable_threads, 1759 struct kvm_vcpu, arch.run_list); 1760 wake_up(&v->arch.cpu_run); 1761 } 1762 1763 spin_unlock(&vc->lock); 1764 return vcpu->arch.ret; 1765 } 1766 1767 static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu) 1768 { 1769 int r; 1770 int srcu_idx; 1771 1772 if (!vcpu->arch.sane) { 1773 run->exit_reason = KVM_EXIT_INTERNAL_ERROR; 1774 return -EINVAL; 1775 } 1776 1777 kvmppc_core_prepare_to_enter(vcpu); 1778 1779 /* No need to go into the guest when all we'll do is come back out */ 1780 if (signal_pending(current)) { 1781 run->exit_reason = KVM_EXIT_INTR; 1782 return -EINTR; 1783 } 1784 1785 atomic_inc(&vcpu->kvm->arch.vcpus_running); 1786 /* Order vcpus_running vs. rma_setup_done, see kvmppc_alloc_reset_hpt */ 1787 smp_mb(); 1788 1789 /* On the first time here, set up HTAB and VRMA or RMA */ 1790 if (!vcpu->kvm->arch.rma_setup_done) { 1791 r = kvmppc_hv_setup_htab_rma(vcpu); 1792 if (r) 1793 goto out; 1794 } 1795 1796 flush_fp_to_thread(current); 1797 flush_altivec_to_thread(current); 1798 flush_vsx_to_thread(current); 1799 vcpu->arch.wqp = &vcpu->arch.vcore->wq; 1800 vcpu->arch.pgdir = current->mm->pgd; 1801 vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST; 1802 1803 do { 1804 r = kvmppc_run_vcpu(run, vcpu); 1805 1806 if (run->exit_reason == KVM_EXIT_PAPR_HCALL && 1807 !(vcpu->arch.shregs.msr & MSR_PR)) { 1808 r = kvmppc_pseries_do_hcall(vcpu); 1809 kvmppc_core_prepare_to_enter(vcpu); 1810 } else if (r == RESUME_PAGE_FAULT) { 1811 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); 1812 r = kvmppc_book3s_hv_page_fault(run, vcpu, 1813 vcpu->arch.fault_dar, vcpu->arch.fault_dsisr); 1814 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx); 1815 } 1816 } while (is_kvmppc_resume_guest(r)); 1817 1818 out: 1819 vcpu->arch.state = KVMPPC_VCPU_NOTREADY; 1820 atomic_dec(&vcpu->kvm->arch.vcpus_running); 1821 return r; 1822 } 1823 1824 1825 /* Work out RMLS (real mode limit selector) field value for a given RMA size. 1826 Assumes POWER7 or PPC970. */ 1827 static inline int lpcr_rmls(unsigned long rma_size) 1828 { 1829 switch (rma_size) { 1830 case 32ul << 20: /* 32 MB */ 1831 if (cpu_has_feature(CPU_FTR_ARCH_206)) 1832 return 8; /* only supported on POWER7 */ 1833 return -1; 1834 case 64ul << 20: /* 64 MB */ 1835 return 3; 1836 case 128ul << 20: /* 128 MB */ 1837 return 7; 1838 case 256ul << 20: /* 256 MB */ 1839 return 4; 1840 case 1ul << 30: /* 1 GB */ 1841 return 2; 1842 case 16ul << 30: /* 16 GB */ 1843 return 1; 1844 case 256ul << 30: /* 256 GB */ 1845 return 0; 1846 default: 1847 return -1; 1848 } 1849 } 1850 1851 static int kvm_rma_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1852 { 1853 struct page *page; 1854 struct kvm_rma_info *ri = vma->vm_file->private_data; 1855 1856 if (vmf->pgoff >= kvm_rma_pages) 1857 return VM_FAULT_SIGBUS; 1858 1859 page = pfn_to_page(ri->base_pfn + vmf->pgoff); 1860 get_page(page); 1861 vmf->page = page; 1862 return 0; 1863 } 1864 1865 static const struct vm_operations_struct kvm_rma_vm_ops = { 1866 .fault = kvm_rma_fault, 1867 }; 1868 1869 static int kvm_rma_mmap(struct file *file, struct vm_area_struct *vma) 1870 { 1871 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP; 1872 vma->vm_ops = &kvm_rma_vm_ops; 1873 return 0; 1874 } 1875 1876 static int kvm_rma_release(struct inode *inode, struct file *filp) 1877 { 1878 struct kvm_rma_info *ri = filp->private_data; 1879 1880 kvm_release_rma(ri); 1881 return 0; 1882 } 1883 1884 static const struct file_operations kvm_rma_fops = { 1885 .mmap = kvm_rma_mmap, 1886 .release = kvm_rma_release, 1887 }; 1888 1889 static long kvm_vm_ioctl_allocate_rma(struct kvm *kvm, 1890 struct kvm_allocate_rma *ret) 1891 { 1892 long fd; 1893 struct kvm_rma_info *ri; 1894 /* 1895 * Only do this on PPC970 in HV mode 1896 */ 1897 if (!cpu_has_feature(CPU_FTR_HVMODE) || 1898 !cpu_has_feature(CPU_FTR_ARCH_201)) 1899 return -EINVAL; 1900 1901 if (!kvm_rma_pages) 1902 return -EINVAL; 1903 1904 ri = kvm_alloc_rma(); 1905 if (!ri) 1906 return -ENOMEM; 1907 1908 fd = anon_inode_getfd("kvm-rma", &kvm_rma_fops, ri, O_RDWR | O_CLOEXEC); 1909 if (fd < 0) 1910 kvm_release_rma(ri); 1911 1912 ret->rma_size = kvm_rma_pages << PAGE_SHIFT; 1913 return fd; 1914 } 1915 1916 static void kvmppc_add_seg_page_size(struct kvm_ppc_one_seg_page_size **sps, 1917 int linux_psize) 1918 { 1919 struct mmu_psize_def *def = &mmu_psize_defs[linux_psize]; 1920 1921 if (!def->shift) 1922 return; 1923 (*sps)->page_shift = def->shift; 1924 (*sps)->slb_enc = def->sllp; 1925 (*sps)->enc[0].page_shift = def->shift; 1926 /* 1927 * Only return base page encoding. We don't want to return 1928 * all the supporting pte_enc, because our H_ENTER doesn't 1929 * support MPSS yet. Once they do, we can start passing all 1930 * support pte_enc here 1931 */ 1932 (*sps)->enc[0].pte_enc = def->penc[linux_psize]; 1933 /* 1934 * Add 16MB MPSS support if host supports it 1935 */ 1936 if (linux_psize != MMU_PAGE_16M && def->penc[MMU_PAGE_16M] != -1) { 1937 (*sps)->enc[1].page_shift = 24; 1938 (*sps)->enc[1].pte_enc = def->penc[MMU_PAGE_16M]; 1939 } 1940 (*sps)++; 1941 } 1942 1943 static int kvm_vm_ioctl_get_smmu_info_hv(struct kvm *kvm, 1944 struct kvm_ppc_smmu_info *info) 1945 { 1946 struct kvm_ppc_one_seg_page_size *sps; 1947 1948 info->flags = KVM_PPC_PAGE_SIZES_REAL; 1949 if (mmu_has_feature(MMU_FTR_1T_SEGMENT)) 1950 info->flags |= KVM_PPC_1T_SEGMENTS; 1951 info->slb_size = mmu_slb_size; 1952 1953 /* We only support these sizes for now, and no muti-size segments */ 1954 sps = &info->sps[0]; 1955 kvmppc_add_seg_page_size(&sps, MMU_PAGE_4K); 1956 kvmppc_add_seg_page_size(&sps, MMU_PAGE_64K); 1957 kvmppc_add_seg_page_size(&sps, MMU_PAGE_16M); 1958 1959 return 0; 1960 } 1961 1962 /* 1963 * Get (and clear) the dirty memory log for a memory slot. 1964 */ 1965 static int kvm_vm_ioctl_get_dirty_log_hv(struct kvm *kvm, 1966 struct kvm_dirty_log *log) 1967 { 1968 struct kvm_memory_slot *memslot; 1969 int r; 1970 unsigned long n; 1971 1972 mutex_lock(&kvm->slots_lock); 1973 1974 r = -EINVAL; 1975 if (log->slot >= KVM_USER_MEM_SLOTS) 1976 goto out; 1977 1978 memslot = id_to_memslot(kvm->memslots, log->slot); 1979 r = -ENOENT; 1980 if (!memslot->dirty_bitmap) 1981 goto out; 1982 1983 n = kvm_dirty_bitmap_bytes(memslot); 1984 memset(memslot->dirty_bitmap, 0, n); 1985 1986 r = kvmppc_hv_get_dirty_log(kvm, memslot, memslot->dirty_bitmap); 1987 if (r) 1988 goto out; 1989 1990 r = -EFAULT; 1991 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n)) 1992 goto out; 1993 1994 r = 0; 1995 out: 1996 mutex_unlock(&kvm->slots_lock); 1997 return r; 1998 } 1999 2000 static void unpin_slot(struct kvm_memory_slot *memslot) 2001 { 2002 unsigned long *physp; 2003 unsigned long j, npages, pfn; 2004 struct page *page; 2005 2006 physp = memslot->arch.slot_phys; 2007 npages = memslot->npages; 2008 if (!physp) 2009 return; 2010 for (j = 0; j < npages; j++) { 2011 if (!(physp[j] & KVMPPC_GOT_PAGE)) 2012 continue; 2013 pfn = physp[j] >> PAGE_SHIFT; 2014 page = pfn_to_page(pfn); 2015 SetPageDirty(page); 2016 put_page(page); 2017 } 2018 } 2019 2020 static void kvmppc_core_free_memslot_hv(struct kvm_memory_slot *free, 2021 struct kvm_memory_slot *dont) 2022 { 2023 if (!dont || free->arch.rmap != dont->arch.rmap) { 2024 vfree(free->arch.rmap); 2025 free->arch.rmap = NULL; 2026 } 2027 if (!dont || free->arch.slot_phys != dont->arch.slot_phys) { 2028 unpin_slot(free); 2029 vfree(free->arch.slot_phys); 2030 free->arch.slot_phys = NULL; 2031 } 2032 } 2033 2034 static int kvmppc_core_create_memslot_hv(struct kvm_memory_slot *slot, 2035 unsigned long npages) 2036 { 2037 slot->arch.rmap = vzalloc(npages * sizeof(*slot->arch.rmap)); 2038 if (!slot->arch.rmap) 2039 return -ENOMEM; 2040 slot->arch.slot_phys = NULL; 2041 2042 return 0; 2043 } 2044 2045 static int kvmppc_core_prepare_memory_region_hv(struct kvm *kvm, 2046 struct kvm_memory_slot *memslot, 2047 struct kvm_userspace_memory_region *mem) 2048 { 2049 unsigned long *phys; 2050 2051 /* Allocate a slot_phys array if needed */ 2052 phys = memslot->arch.slot_phys; 2053 if (!kvm->arch.using_mmu_notifiers && !phys && memslot->npages) { 2054 phys = vzalloc(memslot->npages * sizeof(unsigned long)); 2055 if (!phys) 2056 return -ENOMEM; 2057 memslot->arch.slot_phys = phys; 2058 } 2059 2060 return 0; 2061 } 2062 2063 static void kvmppc_core_commit_memory_region_hv(struct kvm *kvm, 2064 struct kvm_userspace_memory_region *mem, 2065 const struct kvm_memory_slot *old) 2066 { 2067 unsigned long npages = mem->memory_size >> PAGE_SHIFT; 2068 struct kvm_memory_slot *memslot; 2069 2070 if (npages && old->npages) { 2071 /* 2072 * If modifying a memslot, reset all the rmap dirty bits. 2073 * If this is a new memslot, we don't need to do anything 2074 * since the rmap array starts out as all zeroes, 2075 * i.e. no pages are dirty. 2076 */ 2077 memslot = id_to_memslot(kvm->memslots, mem->slot); 2078 kvmppc_hv_get_dirty_log(kvm, memslot, NULL); 2079 } 2080 } 2081 2082 /* 2083 * Update LPCR values in kvm->arch and in vcores. 2084 * Caller must hold kvm->lock. 2085 */ 2086 void kvmppc_update_lpcr(struct kvm *kvm, unsigned long lpcr, unsigned long mask) 2087 { 2088 long int i; 2089 u32 cores_done = 0; 2090 2091 if ((kvm->arch.lpcr & mask) == lpcr) 2092 return; 2093 2094 kvm->arch.lpcr = (kvm->arch.lpcr & ~mask) | lpcr; 2095 2096 for (i = 0; i < KVM_MAX_VCORES; ++i) { 2097 struct kvmppc_vcore *vc = kvm->arch.vcores[i]; 2098 if (!vc) 2099 continue; 2100 spin_lock(&vc->lock); 2101 vc->lpcr = (vc->lpcr & ~mask) | lpcr; 2102 spin_unlock(&vc->lock); 2103 if (++cores_done >= kvm->arch.online_vcores) 2104 break; 2105 } 2106 } 2107 2108 static void kvmppc_mmu_destroy_hv(struct kvm_vcpu *vcpu) 2109 { 2110 return; 2111 } 2112 2113 static int kvmppc_hv_setup_htab_rma(struct kvm_vcpu *vcpu) 2114 { 2115 int err = 0; 2116 struct kvm *kvm = vcpu->kvm; 2117 struct kvm_rma_info *ri = NULL; 2118 unsigned long hva; 2119 struct kvm_memory_slot *memslot; 2120 struct vm_area_struct *vma; 2121 unsigned long lpcr = 0, senc; 2122 unsigned long lpcr_mask = 0; 2123 unsigned long psize, porder; 2124 unsigned long rma_size; 2125 unsigned long rmls; 2126 unsigned long *physp; 2127 unsigned long i, npages; 2128 int srcu_idx; 2129 2130 mutex_lock(&kvm->lock); 2131 if (kvm->arch.rma_setup_done) 2132 goto out; /* another vcpu beat us to it */ 2133 2134 /* Allocate hashed page table (if not done already) and reset it */ 2135 if (!kvm->arch.hpt_virt) { 2136 err = kvmppc_alloc_hpt(kvm, NULL); 2137 if (err) { 2138 pr_err("KVM: Couldn't alloc HPT\n"); 2139 goto out; 2140 } 2141 } 2142 2143 /* Look up the memslot for guest physical address 0 */ 2144 srcu_idx = srcu_read_lock(&kvm->srcu); 2145 memslot = gfn_to_memslot(kvm, 0); 2146 2147 /* We must have some memory at 0 by now */ 2148 err = -EINVAL; 2149 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) 2150 goto out_srcu; 2151 2152 /* Look up the VMA for the start of this memory slot */ 2153 hva = memslot->userspace_addr; 2154 down_read(¤t->mm->mmap_sem); 2155 vma = find_vma(current->mm, hva); 2156 if (!vma || vma->vm_start > hva || (vma->vm_flags & VM_IO)) 2157 goto up_out; 2158 2159 psize = vma_kernel_pagesize(vma); 2160 porder = __ilog2(psize); 2161 2162 /* Is this one of our preallocated RMAs? */ 2163 if (vma->vm_file && vma->vm_file->f_op == &kvm_rma_fops && 2164 hva == vma->vm_start) 2165 ri = vma->vm_file->private_data; 2166 2167 up_read(¤t->mm->mmap_sem); 2168 2169 if (!ri) { 2170 /* On POWER7, use VRMA; on PPC970, give up */ 2171 err = -EPERM; 2172 if (cpu_has_feature(CPU_FTR_ARCH_201)) { 2173 pr_err("KVM: CPU requires an RMO\n"); 2174 goto out_srcu; 2175 } 2176 2177 /* We can handle 4k, 64k or 16M pages in the VRMA */ 2178 err = -EINVAL; 2179 if (!(psize == 0x1000 || psize == 0x10000 || 2180 psize == 0x1000000)) 2181 goto out_srcu; 2182 2183 /* Update VRMASD field in the LPCR */ 2184 senc = slb_pgsize_encoding(psize); 2185 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T | 2186 (VRMA_VSID << SLB_VSID_SHIFT_1T); 2187 lpcr_mask = LPCR_VRMASD; 2188 /* the -4 is to account for senc values starting at 0x10 */ 2189 lpcr = senc << (LPCR_VRMASD_SH - 4); 2190 2191 /* Create HPTEs in the hash page table for the VRMA */ 2192 kvmppc_map_vrma(vcpu, memslot, porder); 2193 2194 } else { 2195 /* Set up to use an RMO region */ 2196 rma_size = kvm_rma_pages; 2197 if (rma_size > memslot->npages) 2198 rma_size = memslot->npages; 2199 rma_size <<= PAGE_SHIFT; 2200 rmls = lpcr_rmls(rma_size); 2201 err = -EINVAL; 2202 if ((long)rmls < 0) { 2203 pr_err("KVM: Can't use RMA of 0x%lx bytes\n", rma_size); 2204 goto out_srcu; 2205 } 2206 atomic_inc(&ri->use_count); 2207 kvm->arch.rma = ri; 2208 2209 /* Update LPCR and RMOR */ 2210 if (cpu_has_feature(CPU_FTR_ARCH_201)) { 2211 /* PPC970; insert RMLS value (split field) in HID4 */ 2212 lpcr_mask = (1ul << HID4_RMLS0_SH) | 2213 (3ul << HID4_RMLS2_SH) | HID4_RMOR; 2214 lpcr = ((rmls >> 2) << HID4_RMLS0_SH) | 2215 ((rmls & 3) << HID4_RMLS2_SH); 2216 /* RMOR is also in HID4 */ 2217 lpcr |= ((ri->base_pfn >> (26 - PAGE_SHIFT)) & 0xffff) 2218 << HID4_RMOR_SH; 2219 } else { 2220 /* POWER7 */ 2221 lpcr_mask = LPCR_VPM0 | LPCR_VRMA_L | LPCR_RMLS; 2222 lpcr = rmls << LPCR_RMLS_SH; 2223 kvm->arch.rmor = ri->base_pfn << PAGE_SHIFT; 2224 } 2225 pr_info("KVM: Using RMO at %lx size %lx (LPCR = %lx)\n", 2226 ri->base_pfn << PAGE_SHIFT, rma_size, lpcr); 2227 2228 /* Initialize phys addrs of pages in RMO */ 2229 npages = kvm_rma_pages; 2230 porder = __ilog2(npages); 2231 physp = memslot->arch.slot_phys; 2232 if (physp) { 2233 if (npages > memslot->npages) 2234 npages = memslot->npages; 2235 spin_lock(&kvm->arch.slot_phys_lock); 2236 for (i = 0; i < npages; ++i) 2237 physp[i] = ((ri->base_pfn + i) << PAGE_SHIFT) + 2238 porder; 2239 spin_unlock(&kvm->arch.slot_phys_lock); 2240 } 2241 } 2242 2243 kvmppc_update_lpcr(kvm, lpcr, lpcr_mask); 2244 2245 /* Order updates to kvm->arch.lpcr etc. vs. rma_setup_done */ 2246 smp_wmb(); 2247 kvm->arch.rma_setup_done = 1; 2248 err = 0; 2249 out_srcu: 2250 srcu_read_unlock(&kvm->srcu, srcu_idx); 2251 out: 2252 mutex_unlock(&kvm->lock); 2253 return err; 2254 2255 up_out: 2256 up_read(¤t->mm->mmap_sem); 2257 goto out_srcu; 2258 } 2259 2260 static int kvmppc_core_init_vm_hv(struct kvm *kvm) 2261 { 2262 unsigned long lpcr, lpid; 2263 2264 /* Allocate the guest's logical partition ID */ 2265 2266 lpid = kvmppc_alloc_lpid(); 2267 if ((long)lpid < 0) 2268 return -ENOMEM; 2269 kvm->arch.lpid = lpid; 2270 2271 /* 2272 * Since we don't flush the TLB when tearing down a VM, 2273 * and this lpid might have previously been used, 2274 * make sure we flush on each core before running the new VM. 2275 */ 2276 cpumask_setall(&kvm->arch.need_tlb_flush); 2277 2278 kvm->arch.rma = NULL; 2279 2280 kvm->arch.host_sdr1 = mfspr(SPRN_SDR1); 2281 2282 if (cpu_has_feature(CPU_FTR_ARCH_201)) { 2283 /* PPC970; HID4 is effectively the LPCR */ 2284 kvm->arch.host_lpid = 0; 2285 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_HID4); 2286 lpcr &= ~((3 << HID4_LPID1_SH) | (0xful << HID4_LPID5_SH)); 2287 lpcr |= ((lpid >> 4) << HID4_LPID1_SH) | 2288 ((lpid & 0xf) << HID4_LPID5_SH); 2289 } else { 2290 /* POWER7; init LPCR for virtual RMA mode */ 2291 kvm->arch.host_lpid = mfspr(SPRN_LPID); 2292 kvm->arch.host_lpcr = lpcr = mfspr(SPRN_LPCR); 2293 lpcr &= LPCR_PECE | LPCR_LPES; 2294 lpcr |= (4UL << LPCR_DPFD_SH) | LPCR_HDICE | 2295 LPCR_VPM0 | LPCR_VPM1; 2296 kvm->arch.vrma_slb_v = SLB_VSID_B_1T | 2297 (VRMA_VSID << SLB_VSID_SHIFT_1T); 2298 /* On POWER8 turn on online bit to enable PURR/SPURR */ 2299 if (cpu_has_feature(CPU_FTR_ARCH_207S)) 2300 lpcr |= LPCR_ONL; 2301 } 2302 kvm->arch.lpcr = lpcr; 2303 2304 kvm->arch.using_mmu_notifiers = !!cpu_has_feature(CPU_FTR_ARCH_206); 2305 spin_lock_init(&kvm->arch.slot_phys_lock); 2306 2307 /* 2308 * Don't allow secondary CPU threads to come online 2309 * while any KVM VMs exist. 2310 */ 2311 inhibit_secondary_onlining(); 2312 2313 return 0; 2314 } 2315 2316 static void kvmppc_free_vcores(struct kvm *kvm) 2317 { 2318 long int i; 2319 2320 for (i = 0; i < KVM_MAX_VCORES; ++i) 2321 kfree(kvm->arch.vcores[i]); 2322 kvm->arch.online_vcores = 0; 2323 } 2324 2325 static void kvmppc_core_destroy_vm_hv(struct kvm *kvm) 2326 { 2327 uninhibit_secondary_onlining(); 2328 2329 kvmppc_free_vcores(kvm); 2330 if (kvm->arch.rma) { 2331 kvm_release_rma(kvm->arch.rma); 2332 kvm->arch.rma = NULL; 2333 } 2334 2335 kvmppc_free_hpt(kvm); 2336 } 2337 2338 /* We don't need to emulate any privileged instructions or dcbz */ 2339 static int kvmppc_core_emulate_op_hv(struct kvm_run *run, struct kvm_vcpu *vcpu, 2340 unsigned int inst, int *advance) 2341 { 2342 return EMULATE_FAIL; 2343 } 2344 2345 static int kvmppc_core_emulate_mtspr_hv(struct kvm_vcpu *vcpu, int sprn, 2346 ulong spr_val) 2347 { 2348 return EMULATE_FAIL; 2349 } 2350 2351 static int kvmppc_core_emulate_mfspr_hv(struct kvm_vcpu *vcpu, int sprn, 2352 ulong *spr_val) 2353 { 2354 return EMULATE_FAIL; 2355 } 2356 2357 static int kvmppc_core_check_processor_compat_hv(void) 2358 { 2359 if (!cpu_has_feature(CPU_FTR_HVMODE)) 2360 return -EIO; 2361 return 0; 2362 } 2363 2364 static long kvm_arch_vm_ioctl_hv(struct file *filp, 2365 unsigned int ioctl, unsigned long arg) 2366 { 2367 struct kvm *kvm __maybe_unused = filp->private_data; 2368 void __user *argp = (void __user *)arg; 2369 long r; 2370 2371 switch (ioctl) { 2372 2373 case KVM_ALLOCATE_RMA: { 2374 struct kvm_allocate_rma rma; 2375 struct kvm *kvm = filp->private_data; 2376 2377 r = kvm_vm_ioctl_allocate_rma(kvm, &rma); 2378 if (r >= 0 && copy_to_user(argp, &rma, sizeof(rma))) 2379 r = -EFAULT; 2380 break; 2381 } 2382 2383 case KVM_PPC_ALLOCATE_HTAB: { 2384 u32 htab_order; 2385 2386 r = -EFAULT; 2387 if (get_user(htab_order, (u32 __user *)argp)) 2388 break; 2389 r = kvmppc_alloc_reset_hpt(kvm, &htab_order); 2390 if (r) 2391 break; 2392 r = -EFAULT; 2393 if (put_user(htab_order, (u32 __user *)argp)) 2394 break; 2395 r = 0; 2396 break; 2397 } 2398 2399 case KVM_PPC_GET_HTAB_FD: { 2400 struct kvm_get_htab_fd ghf; 2401 2402 r = -EFAULT; 2403 if (copy_from_user(&ghf, argp, sizeof(ghf))) 2404 break; 2405 r = kvm_vm_ioctl_get_htab_fd(kvm, &ghf); 2406 break; 2407 } 2408 2409 default: 2410 r = -ENOTTY; 2411 } 2412 2413 return r; 2414 } 2415 2416 static struct kvmppc_ops kvm_ops_hv = { 2417 .get_sregs = kvm_arch_vcpu_ioctl_get_sregs_hv, 2418 .set_sregs = kvm_arch_vcpu_ioctl_set_sregs_hv, 2419 .get_one_reg = kvmppc_get_one_reg_hv, 2420 .set_one_reg = kvmppc_set_one_reg_hv, 2421 .vcpu_load = kvmppc_core_vcpu_load_hv, 2422 .vcpu_put = kvmppc_core_vcpu_put_hv, 2423 .set_msr = kvmppc_set_msr_hv, 2424 .vcpu_run = kvmppc_vcpu_run_hv, 2425 .vcpu_create = kvmppc_core_vcpu_create_hv, 2426 .vcpu_free = kvmppc_core_vcpu_free_hv, 2427 .check_requests = kvmppc_core_check_requests_hv, 2428 .get_dirty_log = kvm_vm_ioctl_get_dirty_log_hv, 2429 .flush_memslot = kvmppc_core_flush_memslot_hv, 2430 .prepare_memory_region = kvmppc_core_prepare_memory_region_hv, 2431 .commit_memory_region = kvmppc_core_commit_memory_region_hv, 2432 .unmap_hva = kvm_unmap_hva_hv, 2433 .unmap_hva_range = kvm_unmap_hva_range_hv, 2434 .age_hva = kvm_age_hva_hv, 2435 .test_age_hva = kvm_test_age_hva_hv, 2436 .set_spte_hva = kvm_set_spte_hva_hv, 2437 .mmu_destroy = kvmppc_mmu_destroy_hv, 2438 .free_memslot = kvmppc_core_free_memslot_hv, 2439 .create_memslot = kvmppc_core_create_memslot_hv, 2440 .init_vm = kvmppc_core_init_vm_hv, 2441 .destroy_vm = kvmppc_core_destroy_vm_hv, 2442 .get_smmu_info = kvm_vm_ioctl_get_smmu_info_hv, 2443 .emulate_op = kvmppc_core_emulate_op_hv, 2444 .emulate_mtspr = kvmppc_core_emulate_mtspr_hv, 2445 .emulate_mfspr = kvmppc_core_emulate_mfspr_hv, 2446 .fast_vcpu_kick = kvmppc_fast_vcpu_kick_hv, 2447 .arch_vm_ioctl = kvm_arch_vm_ioctl_hv, 2448 }; 2449 2450 static int kvmppc_book3s_init_hv(void) 2451 { 2452 int r; 2453 /* 2454 * FIXME!! Do we need to check on all cpus ? 2455 */ 2456 r = kvmppc_core_check_processor_compat_hv(); 2457 if (r < 0) 2458 return -ENODEV; 2459 2460 kvm_ops_hv.owner = THIS_MODULE; 2461 kvmppc_hv_ops = &kvm_ops_hv; 2462 2463 r = kvmppc_mmu_hv_init(); 2464 return r; 2465 } 2466 2467 static void kvmppc_book3s_exit_hv(void) 2468 { 2469 kvmppc_hv_ops = NULL; 2470 } 2471 2472 module_init(kvmppc_book3s_init_hv); 2473 module_exit(kvmppc_book3s_exit_hv); 2474 MODULE_LICENSE("GPL"); 2475 MODULE_ALIAS_MISCDEV(KVM_MINOR); 2476 MODULE_ALIAS("devname:kvm"); 2477