1 /* 2 * This program is free software; you can redistribute it and/or modify 3 * it under the terms of the GNU General Public License, version 2, as 4 * published by the Free Software Foundation. 5 * 6 * This program is distributed in the hope that it will be useful, 7 * but WITHOUT ANY WARRANTY; without even the implied warranty of 8 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 9 * GNU General Public License for more details. 10 * 11 * You should have received a copy of the GNU General Public License 12 * along with this program; if not, write to the Free Software 13 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. 14 * 15 * Copyright IBM Corp. 2007 16 * 17 * Authors: Hollis Blanchard <hollisb@us.ibm.com> 18 * Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com> 19 */ 20 21 #include <linux/errno.h> 22 #include <linux/err.h> 23 #include <linux/kvm_host.h> 24 #include <linux/module.h> 25 #include <linux/vmalloc.h> 26 #include <linux/hrtimer.h> 27 #include <linux/fs.h> 28 #include <linux/slab.h> 29 #include <asm/cputable.h> 30 #include <asm/uaccess.h> 31 #include <asm/kvm_ppc.h> 32 #include <asm/tlbflush.h> 33 #include "timing.h" 34 #include "../mm/mmu_decl.h" 35 36 #define CREATE_TRACE_POINTS 37 #include "trace.h" 38 39 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v) 40 { 41 return !(v->arch.shared->msr & MSR_WE) || 42 !!(v->arch.pending_exceptions); 43 } 44 45 int kvmppc_kvm_pv(struct kvm_vcpu *vcpu) 46 { 47 int nr = kvmppc_get_gpr(vcpu, 11); 48 int r; 49 unsigned long __maybe_unused param1 = kvmppc_get_gpr(vcpu, 3); 50 unsigned long __maybe_unused param2 = kvmppc_get_gpr(vcpu, 4); 51 unsigned long __maybe_unused param3 = kvmppc_get_gpr(vcpu, 5); 52 unsigned long __maybe_unused param4 = kvmppc_get_gpr(vcpu, 6); 53 unsigned long r2 = 0; 54 55 if (!(vcpu->arch.shared->msr & MSR_SF)) { 56 /* 32 bit mode */ 57 param1 &= 0xffffffff; 58 param2 &= 0xffffffff; 59 param3 &= 0xffffffff; 60 param4 &= 0xffffffff; 61 } 62 63 switch (nr) { 64 case HC_VENDOR_KVM | KVM_HC_PPC_MAP_MAGIC_PAGE: 65 { 66 vcpu->arch.magic_page_pa = param1; 67 vcpu->arch.magic_page_ea = param2; 68 69 r2 = KVM_MAGIC_FEAT_SR; 70 71 r = HC_EV_SUCCESS; 72 break; 73 } 74 case HC_VENDOR_KVM | KVM_HC_FEATURES: 75 r = HC_EV_SUCCESS; 76 #if defined(CONFIG_PPC_BOOK3S) /* XXX Missing magic page on BookE */ 77 r2 |= (1 << KVM_FEATURE_MAGIC_PAGE); 78 #endif 79 80 /* Second return value is in r4 */ 81 break; 82 default: 83 r = HC_EV_UNIMPLEMENTED; 84 break; 85 } 86 87 kvmppc_set_gpr(vcpu, 4, r2); 88 89 return r; 90 } 91 92 int kvmppc_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu) 93 { 94 enum emulation_result er; 95 int r; 96 97 er = kvmppc_emulate_instruction(run, vcpu); 98 switch (er) { 99 case EMULATE_DONE: 100 /* Future optimization: only reload non-volatiles if they were 101 * actually modified. */ 102 r = RESUME_GUEST_NV; 103 break; 104 case EMULATE_DO_MMIO: 105 run->exit_reason = KVM_EXIT_MMIO; 106 /* We must reload nonvolatiles because "update" load/store 107 * instructions modify register state. */ 108 /* Future optimization: only reload non-volatiles if they were 109 * actually modified. */ 110 r = RESUME_HOST_NV; 111 break; 112 case EMULATE_FAIL: 113 /* XXX Deliver Program interrupt to guest. */ 114 printk(KERN_EMERG "%s: emulation failed (%08x)\n", __func__, 115 kvmppc_get_last_inst(vcpu)); 116 r = RESUME_HOST; 117 break; 118 default: 119 BUG(); 120 } 121 122 return r; 123 } 124 125 int kvm_arch_hardware_enable(void *garbage) 126 { 127 return 0; 128 } 129 130 void kvm_arch_hardware_disable(void *garbage) 131 { 132 } 133 134 int kvm_arch_hardware_setup(void) 135 { 136 return 0; 137 } 138 139 void kvm_arch_hardware_unsetup(void) 140 { 141 } 142 143 void kvm_arch_check_processor_compat(void *rtn) 144 { 145 *(int *)rtn = kvmppc_core_check_processor_compat(); 146 } 147 148 struct kvm *kvm_arch_create_vm(void) 149 { 150 struct kvm *kvm; 151 152 kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL); 153 if (!kvm) 154 return ERR_PTR(-ENOMEM); 155 156 return kvm; 157 } 158 159 static void kvmppc_free_vcpus(struct kvm *kvm) 160 { 161 unsigned int i; 162 struct kvm_vcpu *vcpu; 163 164 kvm_for_each_vcpu(i, vcpu, kvm) 165 kvm_arch_vcpu_free(vcpu); 166 167 mutex_lock(&kvm->lock); 168 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++) 169 kvm->vcpus[i] = NULL; 170 171 atomic_set(&kvm->online_vcpus, 0); 172 mutex_unlock(&kvm->lock); 173 } 174 175 void kvm_arch_sync_events(struct kvm *kvm) 176 { 177 } 178 179 void kvm_arch_destroy_vm(struct kvm *kvm) 180 { 181 kvmppc_free_vcpus(kvm); 182 kvm_free_physmem(kvm); 183 cleanup_srcu_struct(&kvm->srcu); 184 kfree(kvm); 185 } 186 187 int kvm_dev_ioctl_check_extension(long ext) 188 { 189 int r; 190 191 switch (ext) { 192 case KVM_CAP_PPC_SEGSTATE: 193 case KVM_CAP_PPC_PAIRED_SINGLES: 194 case KVM_CAP_PPC_UNSET_IRQ: 195 case KVM_CAP_PPC_IRQ_LEVEL: 196 case KVM_CAP_ENABLE_CAP: 197 case KVM_CAP_PPC_OSI: 198 case KVM_CAP_PPC_GET_PVINFO: 199 r = 1; 200 break; 201 case KVM_CAP_COALESCED_MMIO: 202 r = KVM_COALESCED_MMIO_PAGE_OFFSET; 203 break; 204 default: 205 r = 0; 206 break; 207 } 208 return r; 209 210 } 211 212 long kvm_arch_dev_ioctl(struct file *filp, 213 unsigned int ioctl, unsigned long arg) 214 { 215 return -EINVAL; 216 } 217 218 int kvm_arch_prepare_memory_region(struct kvm *kvm, 219 struct kvm_memory_slot *memslot, 220 struct kvm_memory_slot old, 221 struct kvm_userspace_memory_region *mem, 222 int user_alloc) 223 { 224 return 0; 225 } 226 227 void kvm_arch_commit_memory_region(struct kvm *kvm, 228 struct kvm_userspace_memory_region *mem, 229 struct kvm_memory_slot old, 230 int user_alloc) 231 { 232 return; 233 } 234 235 236 void kvm_arch_flush_shadow(struct kvm *kvm) 237 { 238 } 239 240 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id) 241 { 242 struct kvm_vcpu *vcpu; 243 vcpu = kvmppc_core_vcpu_create(kvm, id); 244 if (!IS_ERR(vcpu)) 245 kvmppc_create_vcpu_debugfs(vcpu, id); 246 return vcpu; 247 } 248 249 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu) 250 { 251 /* Make sure we're not using the vcpu anymore */ 252 hrtimer_cancel(&vcpu->arch.dec_timer); 253 tasklet_kill(&vcpu->arch.tasklet); 254 255 kvmppc_remove_vcpu_debugfs(vcpu); 256 kvmppc_core_vcpu_free(vcpu); 257 } 258 259 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) 260 { 261 kvm_arch_vcpu_free(vcpu); 262 } 263 264 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) 265 { 266 return kvmppc_core_pending_dec(vcpu); 267 } 268 269 static void kvmppc_decrementer_func(unsigned long data) 270 { 271 struct kvm_vcpu *vcpu = (struct kvm_vcpu *)data; 272 273 kvmppc_core_queue_dec(vcpu); 274 275 if (waitqueue_active(&vcpu->wq)) { 276 wake_up_interruptible(&vcpu->wq); 277 vcpu->stat.halt_wakeup++; 278 } 279 } 280 281 /* 282 * low level hrtimer wake routine. Because this runs in hardirq context 283 * we schedule a tasklet to do the real work. 284 */ 285 enum hrtimer_restart kvmppc_decrementer_wakeup(struct hrtimer *timer) 286 { 287 struct kvm_vcpu *vcpu; 288 289 vcpu = container_of(timer, struct kvm_vcpu, arch.dec_timer); 290 tasklet_schedule(&vcpu->arch.tasklet); 291 292 return HRTIMER_NORESTART; 293 } 294 295 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu) 296 { 297 hrtimer_init(&vcpu->arch.dec_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS); 298 tasklet_init(&vcpu->arch.tasklet, kvmppc_decrementer_func, (ulong)vcpu); 299 vcpu->arch.dec_timer.function = kvmppc_decrementer_wakeup; 300 301 return 0; 302 } 303 304 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu) 305 { 306 kvmppc_mmu_destroy(vcpu); 307 } 308 309 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) 310 { 311 kvmppc_core_vcpu_load(vcpu, cpu); 312 } 313 314 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) 315 { 316 kvmppc_core_vcpu_put(vcpu); 317 } 318 319 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu, 320 struct kvm_guest_debug *dbg) 321 { 322 return -EINVAL; 323 } 324 325 static void kvmppc_complete_dcr_load(struct kvm_vcpu *vcpu, 326 struct kvm_run *run) 327 { 328 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, run->dcr.data); 329 } 330 331 static void kvmppc_complete_mmio_load(struct kvm_vcpu *vcpu, 332 struct kvm_run *run) 333 { 334 u64 uninitialized_var(gpr); 335 336 if (run->mmio.len > sizeof(gpr)) { 337 printk(KERN_ERR "bad MMIO length: %d\n", run->mmio.len); 338 return; 339 } 340 341 if (vcpu->arch.mmio_is_bigendian) { 342 switch (run->mmio.len) { 343 case 8: gpr = *(u64 *)run->mmio.data; break; 344 case 4: gpr = *(u32 *)run->mmio.data; break; 345 case 2: gpr = *(u16 *)run->mmio.data; break; 346 case 1: gpr = *(u8 *)run->mmio.data; break; 347 } 348 } else { 349 /* Convert BE data from userland back to LE. */ 350 switch (run->mmio.len) { 351 case 4: gpr = ld_le32((u32 *)run->mmio.data); break; 352 case 2: gpr = ld_le16((u16 *)run->mmio.data); break; 353 case 1: gpr = *(u8 *)run->mmio.data; break; 354 } 355 } 356 357 if (vcpu->arch.mmio_sign_extend) { 358 switch (run->mmio.len) { 359 #ifdef CONFIG_PPC64 360 case 4: 361 gpr = (s64)(s32)gpr; 362 break; 363 #endif 364 case 2: 365 gpr = (s64)(s16)gpr; 366 break; 367 case 1: 368 gpr = (s64)(s8)gpr; 369 break; 370 } 371 } 372 373 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr); 374 375 switch (vcpu->arch.io_gpr & KVM_REG_EXT_MASK) { 376 case KVM_REG_GPR: 377 kvmppc_set_gpr(vcpu, vcpu->arch.io_gpr, gpr); 378 break; 379 case KVM_REG_FPR: 380 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr; 381 break; 382 #ifdef CONFIG_PPC_BOOK3S 383 case KVM_REG_QPR: 384 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr; 385 break; 386 case KVM_REG_FQPR: 387 vcpu->arch.fpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr; 388 vcpu->arch.qpr[vcpu->arch.io_gpr & KVM_REG_MASK] = gpr; 389 break; 390 #endif 391 default: 392 BUG(); 393 } 394 } 395 396 int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu, 397 unsigned int rt, unsigned int bytes, int is_bigendian) 398 { 399 if (bytes > sizeof(run->mmio.data)) { 400 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__, 401 run->mmio.len); 402 } 403 404 run->mmio.phys_addr = vcpu->arch.paddr_accessed; 405 run->mmio.len = bytes; 406 run->mmio.is_write = 0; 407 408 vcpu->arch.io_gpr = rt; 409 vcpu->arch.mmio_is_bigendian = is_bigendian; 410 vcpu->mmio_needed = 1; 411 vcpu->mmio_is_write = 0; 412 vcpu->arch.mmio_sign_extend = 0; 413 414 return EMULATE_DO_MMIO; 415 } 416 417 /* Same as above, but sign extends */ 418 int kvmppc_handle_loads(struct kvm_run *run, struct kvm_vcpu *vcpu, 419 unsigned int rt, unsigned int bytes, int is_bigendian) 420 { 421 int r; 422 423 r = kvmppc_handle_load(run, vcpu, rt, bytes, is_bigendian); 424 vcpu->arch.mmio_sign_extend = 1; 425 426 return r; 427 } 428 429 int kvmppc_handle_store(struct kvm_run *run, struct kvm_vcpu *vcpu, 430 u64 val, unsigned int bytes, int is_bigendian) 431 { 432 void *data = run->mmio.data; 433 434 if (bytes > sizeof(run->mmio.data)) { 435 printk(KERN_ERR "%s: bad MMIO length: %d\n", __func__, 436 run->mmio.len); 437 } 438 439 run->mmio.phys_addr = vcpu->arch.paddr_accessed; 440 run->mmio.len = bytes; 441 run->mmio.is_write = 1; 442 vcpu->mmio_needed = 1; 443 vcpu->mmio_is_write = 1; 444 445 /* Store the value at the lowest bytes in 'data'. */ 446 if (is_bigendian) { 447 switch (bytes) { 448 case 8: *(u64 *)data = val; break; 449 case 4: *(u32 *)data = val; break; 450 case 2: *(u16 *)data = val; break; 451 case 1: *(u8 *)data = val; break; 452 } 453 } else { 454 /* Store LE value into 'data'. */ 455 switch (bytes) { 456 case 4: st_le32(data, val); break; 457 case 2: st_le16(data, val); break; 458 case 1: *(u8 *)data = val; break; 459 } 460 } 461 462 return EMULATE_DO_MMIO; 463 } 464 465 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run) 466 { 467 int r; 468 sigset_t sigsaved; 469 470 if (vcpu->sigset_active) 471 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved); 472 473 if (vcpu->mmio_needed) { 474 if (!vcpu->mmio_is_write) 475 kvmppc_complete_mmio_load(vcpu, run); 476 vcpu->mmio_needed = 0; 477 } else if (vcpu->arch.dcr_needed) { 478 if (!vcpu->arch.dcr_is_write) 479 kvmppc_complete_dcr_load(vcpu, run); 480 vcpu->arch.dcr_needed = 0; 481 } else if (vcpu->arch.osi_needed) { 482 u64 *gprs = run->osi.gprs; 483 int i; 484 485 for (i = 0; i < 32; i++) 486 kvmppc_set_gpr(vcpu, i, gprs[i]); 487 vcpu->arch.osi_needed = 0; 488 } 489 490 kvmppc_core_deliver_interrupts(vcpu); 491 492 local_irq_disable(); 493 kvm_guest_enter(); 494 r = __kvmppc_vcpu_run(run, vcpu); 495 kvm_guest_exit(); 496 local_irq_enable(); 497 498 if (vcpu->sigset_active) 499 sigprocmask(SIG_SETMASK, &sigsaved, NULL); 500 501 return r; 502 } 503 504 int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, struct kvm_interrupt *irq) 505 { 506 if (irq->irq == KVM_INTERRUPT_UNSET) 507 kvmppc_core_dequeue_external(vcpu, irq); 508 else 509 kvmppc_core_queue_external(vcpu, irq); 510 511 if (waitqueue_active(&vcpu->wq)) { 512 wake_up_interruptible(&vcpu->wq); 513 vcpu->stat.halt_wakeup++; 514 } 515 516 return 0; 517 } 518 519 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu, 520 struct kvm_enable_cap *cap) 521 { 522 int r; 523 524 if (cap->flags) 525 return -EINVAL; 526 527 switch (cap->cap) { 528 case KVM_CAP_PPC_OSI: 529 r = 0; 530 vcpu->arch.osi_enabled = true; 531 break; 532 default: 533 r = -EINVAL; 534 break; 535 } 536 537 return r; 538 } 539 540 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, 541 struct kvm_mp_state *mp_state) 542 { 543 return -EINVAL; 544 } 545 546 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, 547 struct kvm_mp_state *mp_state) 548 { 549 return -EINVAL; 550 } 551 552 long kvm_arch_vcpu_ioctl(struct file *filp, 553 unsigned int ioctl, unsigned long arg) 554 { 555 struct kvm_vcpu *vcpu = filp->private_data; 556 void __user *argp = (void __user *)arg; 557 long r; 558 559 switch (ioctl) { 560 case KVM_INTERRUPT: { 561 struct kvm_interrupt irq; 562 r = -EFAULT; 563 if (copy_from_user(&irq, argp, sizeof(irq))) 564 goto out; 565 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq); 566 goto out; 567 } 568 569 case KVM_ENABLE_CAP: 570 { 571 struct kvm_enable_cap cap; 572 r = -EFAULT; 573 if (copy_from_user(&cap, argp, sizeof(cap))) 574 goto out; 575 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap); 576 break; 577 } 578 default: 579 r = -EINVAL; 580 } 581 582 out: 583 return r; 584 } 585 586 static int kvm_vm_ioctl_get_pvinfo(struct kvm_ppc_pvinfo *pvinfo) 587 { 588 u32 inst_lis = 0x3c000000; 589 u32 inst_ori = 0x60000000; 590 u32 inst_nop = 0x60000000; 591 u32 inst_sc = 0x44000002; 592 u32 inst_imm_mask = 0xffff; 593 594 /* 595 * The hypercall to get into KVM from within guest context is as 596 * follows: 597 * 598 * lis r0, r0, KVM_SC_MAGIC_R0@h 599 * ori r0, KVM_SC_MAGIC_R0@l 600 * sc 601 * nop 602 */ 603 pvinfo->hcall[0] = inst_lis | ((KVM_SC_MAGIC_R0 >> 16) & inst_imm_mask); 604 pvinfo->hcall[1] = inst_ori | (KVM_SC_MAGIC_R0 & inst_imm_mask); 605 pvinfo->hcall[2] = inst_sc; 606 pvinfo->hcall[3] = inst_nop; 607 608 return 0; 609 } 610 611 long kvm_arch_vm_ioctl(struct file *filp, 612 unsigned int ioctl, unsigned long arg) 613 { 614 void __user *argp = (void __user *)arg; 615 long r; 616 617 switch (ioctl) { 618 case KVM_PPC_GET_PVINFO: { 619 struct kvm_ppc_pvinfo pvinfo; 620 memset(&pvinfo, 0, sizeof(pvinfo)); 621 r = kvm_vm_ioctl_get_pvinfo(&pvinfo); 622 if (copy_to_user(argp, &pvinfo, sizeof(pvinfo))) { 623 r = -EFAULT; 624 goto out; 625 } 626 627 break; 628 } 629 default: 630 r = -ENOTTY; 631 } 632 633 out: 634 return r; 635 } 636 637 int kvm_arch_init(void *opaque) 638 { 639 return 0; 640 } 641 642 void kvm_arch_exit(void) 643 { 644 } 645