1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * S390 version 4 * Copyright IBM Corp. 1999 5 * Author(s): Hartmut Penner (hp@de.ibm.com) 6 * Ulrich Weigand (uweigand@de.ibm.com) 7 * 8 * Derived from "arch/i386/mm/fault.c" 9 * Copyright (C) 1995 Linus Torvalds 10 */ 11 12 #include <linux/kernel_stat.h> 13 #include <linux/perf_event.h> 14 #include <linux/signal.h> 15 #include <linux/sched.h> 16 #include <linux/sched/debug.h> 17 #include <linux/kernel.h> 18 #include <linux/errno.h> 19 #include <linux/string.h> 20 #include <linux/types.h> 21 #include <linux/ptrace.h> 22 #include <linux/mman.h> 23 #include <linux/mm.h> 24 #include <linux/compat.h> 25 #include <linux/smp.h> 26 #include <linux/kdebug.h> 27 #include <linux/init.h> 28 #include <linux/console.h> 29 #include <linux/extable.h> 30 #include <linux/hardirq.h> 31 #include <linux/kprobes.h> 32 #include <linux/uaccess.h> 33 #include <linux/hugetlb.h> 34 #include <linux/kfence.h> 35 #include <asm/asm-extable.h> 36 #include <asm/asm-offsets.h> 37 #include <asm/diag.h> 38 #include <asm/gmap.h> 39 #include <asm/irq.h> 40 #include <asm/mmu_context.h> 41 #include <asm/facility.h> 42 #include <asm/uv.h> 43 #include "../kernel/entry.h" 44 45 #define __FAIL_ADDR_MASK -4096L 46 #define __SUBCODE_MASK 0x0600 47 #define __PF_RES_FIELD 0x8000000000000000ULL 48 49 /* 50 * Allocate private vm_fault_reason from top. Please make sure it won't 51 * collide with vm_fault_reason. 52 */ 53 #define VM_FAULT_BADCONTEXT ((__force vm_fault_t)0x80000000) 54 #define VM_FAULT_BADMAP ((__force vm_fault_t)0x40000000) 55 #define VM_FAULT_BADACCESS ((__force vm_fault_t)0x20000000) 56 #define VM_FAULT_SIGNAL ((__force vm_fault_t)0x10000000) 57 #define VM_FAULT_PFAULT ((__force vm_fault_t)0x8000000) 58 59 enum fault_type { 60 KERNEL_FAULT, 61 USER_FAULT, 62 GMAP_FAULT, 63 }; 64 65 static unsigned long store_indication __read_mostly; 66 67 static int __init fault_init(void) 68 { 69 if (test_facility(75)) 70 store_indication = 0xc00; 71 return 0; 72 } 73 early_initcall(fault_init); 74 75 /* 76 * Find out which address space caused the exception. 77 */ 78 static enum fault_type get_fault_type(struct pt_regs *regs) 79 { 80 unsigned long trans_exc_code; 81 82 trans_exc_code = regs->int_parm_long & 3; 83 if (likely(trans_exc_code == 0)) { 84 /* primary space exception */ 85 if (user_mode(regs)) 86 return USER_FAULT; 87 if (!IS_ENABLED(CONFIG_PGSTE)) 88 return KERNEL_FAULT; 89 if (test_pt_regs_flag(regs, PIF_GUEST_FAULT)) 90 return GMAP_FAULT; 91 return KERNEL_FAULT; 92 } 93 if (trans_exc_code == 2) 94 return USER_FAULT; 95 if (trans_exc_code == 1) { 96 /* access register mode, not used in the kernel */ 97 return USER_FAULT; 98 } 99 /* home space exception -> access via kernel ASCE */ 100 return KERNEL_FAULT; 101 } 102 103 static unsigned long get_fault_address(struct pt_regs *regs) 104 { 105 unsigned long trans_exc_code = regs->int_parm_long; 106 107 return trans_exc_code & __FAIL_ADDR_MASK; 108 } 109 110 static bool fault_is_write(struct pt_regs *regs) 111 { 112 unsigned long trans_exc_code = regs->int_parm_long; 113 114 return (trans_exc_code & store_indication) == 0x400; 115 } 116 117 static int bad_address(void *p) 118 { 119 unsigned long dummy; 120 121 return get_kernel_nofault(dummy, (unsigned long *)p); 122 } 123 124 static void dump_pagetable(unsigned long asce, unsigned long address) 125 { 126 unsigned long *table = __va(asce & _ASCE_ORIGIN); 127 128 pr_alert("AS:%016lx ", asce); 129 switch (asce & _ASCE_TYPE_MASK) { 130 case _ASCE_TYPE_REGION1: 131 table += (address & _REGION1_INDEX) >> _REGION1_SHIFT; 132 if (bad_address(table)) 133 goto bad; 134 pr_cont("R1:%016lx ", *table); 135 if (*table & _REGION_ENTRY_INVALID) 136 goto out; 137 table = __va(*table & _REGION_ENTRY_ORIGIN); 138 fallthrough; 139 case _ASCE_TYPE_REGION2: 140 table += (address & _REGION2_INDEX) >> _REGION2_SHIFT; 141 if (bad_address(table)) 142 goto bad; 143 pr_cont("R2:%016lx ", *table); 144 if (*table & _REGION_ENTRY_INVALID) 145 goto out; 146 table = __va(*table & _REGION_ENTRY_ORIGIN); 147 fallthrough; 148 case _ASCE_TYPE_REGION3: 149 table += (address & _REGION3_INDEX) >> _REGION3_SHIFT; 150 if (bad_address(table)) 151 goto bad; 152 pr_cont("R3:%016lx ", *table); 153 if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE)) 154 goto out; 155 table = __va(*table & _REGION_ENTRY_ORIGIN); 156 fallthrough; 157 case _ASCE_TYPE_SEGMENT: 158 table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT; 159 if (bad_address(table)) 160 goto bad; 161 pr_cont("S:%016lx ", *table); 162 if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE)) 163 goto out; 164 table = __va(*table & _SEGMENT_ENTRY_ORIGIN); 165 } 166 table += (address & _PAGE_INDEX) >> _PAGE_SHIFT; 167 if (bad_address(table)) 168 goto bad; 169 pr_cont("P:%016lx ", *table); 170 out: 171 pr_cont("\n"); 172 return; 173 bad: 174 pr_cont("BAD\n"); 175 } 176 177 static void dump_fault_info(struct pt_regs *regs) 178 { 179 unsigned long asce; 180 181 pr_alert("Failing address: %016lx TEID: %016lx\n", 182 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long); 183 pr_alert("Fault in "); 184 switch (regs->int_parm_long & 3) { 185 case 3: 186 pr_cont("home space "); 187 break; 188 case 2: 189 pr_cont("secondary space "); 190 break; 191 case 1: 192 pr_cont("access register "); 193 break; 194 case 0: 195 pr_cont("primary space "); 196 break; 197 } 198 pr_cont("mode while using "); 199 switch (get_fault_type(regs)) { 200 case USER_FAULT: 201 asce = S390_lowcore.user_asce; 202 pr_cont("user "); 203 break; 204 case GMAP_FAULT: 205 asce = ((struct gmap *) S390_lowcore.gmap)->asce; 206 pr_cont("gmap "); 207 break; 208 case KERNEL_FAULT: 209 asce = S390_lowcore.kernel_asce; 210 pr_cont("kernel "); 211 break; 212 default: 213 unreachable(); 214 } 215 pr_cont("ASCE.\n"); 216 dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK); 217 } 218 219 int show_unhandled_signals = 1; 220 221 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault) 222 { 223 if ((task_pid_nr(current) > 1) && !show_unhandled_signals) 224 return; 225 if (!unhandled_signal(current, signr)) 226 return; 227 if (!printk_ratelimit()) 228 return; 229 printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ", 230 regs->int_code & 0xffff, regs->int_code >> 17); 231 print_vma_addr(KERN_CONT "in ", regs->psw.addr); 232 printk(KERN_CONT "\n"); 233 if (is_mm_fault) 234 dump_fault_info(regs); 235 show_regs(regs); 236 } 237 238 /* 239 * Send SIGSEGV to task. This is an external routine 240 * to keep the stack usage of do_page_fault small. 241 */ 242 static noinline void do_sigsegv(struct pt_regs *regs, int si_code) 243 { 244 report_user_fault(regs, SIGSEGV, 1); 245 force_sig_fault(SIGSEGV, si_code, 246 (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK)); 247 } 248 249 static noinline void do_no_context(struct pt_regs *regs, vm_fault_t fault) 250 { 251 enum fault_type fault_type; 252 unsigned long address; 253 bool is_write; 254 255 if (fixup_exception(regs)) 256 return; 257 fault_type = get_fault_type(regs); 258 if ((fault_type == KERNEL_FAULT) && (fault == VM_FAULT_BADCONTEXT)) { 259 address = get_fault_address(regs); 260 is_write = fault_is_write(regs); 261 if (kfence_handle_page_fault(address, is_write, regs)) 262 return; 263 } 264 /* 265 * Oops. The kernel tried to access some bad page. We'll have to 266 * terminate things with extreme prejudice. 267 */ 268 if (fault_type == KERNEL_FAULT) 269 printk(KERN_ALERT "Unable to handle kernel pointer dereference" 270 " in virtual kernel address space\n"); 271 else 272 printk(KERN_ALERT "Unable to handle kernel paging request" 273 " in virtual user address space\n"); 274 dump_fault_info(regs); 275 die(regs, "Oops"); 276 } 277 278 static noinline void do_low_address(struct pt_regs *regs) 279 { 280 /* Low-address protection hit in kernel mode means 281 NULL pointer write access in kernel mode. */ 282 if (regs->psw.mask & PSW_MASK_PSTATE) { 283 /* Low-address protection hit in user mode 'cannot happen'. */ 284 die (regs, "Low-address protection"); 285 } 286 287 do_no_context(regs, VM_FAULT_BADACCESS); 288 } 289 290 static noinline void do_sigbus(struct pt_regs *regs) 291 { 292 /* 293 * Send a sigbus, regardless of whether we were in kernel 294 * or user mode. 295 */ 296 force_sig_fault(SIGBUS, BUS_ADRERR, 297 (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK)); 298 } 299 300 static noinline void do_fault_error(struct pt_regs *regs, vm_fault_t fault) 301 { 302 int si_code; 303 304 switch (fault) { 305 case VM_FAULT_BADACCESS: 306 case VM_FAULT_BADMAP: 307 /* Bad memory access. Check if it is kernel or user space. */ 308 if (user_mode(regs)) { 309 /* User mode accesses just cause a SIGSEGV */ 310 si_code = (fault == VM_FAULT_BADMAP) ? 311 SEGV_MAPERR : SEGV_ACCERR; 312 do_sigsegv(regs, si_code); 313 break; 314 } 315 fallthrough; 316 case VM_FAULT_BADCONTEXT: 317 case VM_FAULT_PFAULT: 318 do_no_context(regs, fault); 319 break; 320 case VM_FAULT_SIGNAL: 321 if (!user_mode(regs)) 322 do_no_context(regs, fault); 323 break; 324 default: /* fault & VM_FAULT_ERROR */ 325 if (fault & VM_FAULT_OOM) { 326 if (!user_mode(regs)) 327 do_no_context(regs, fault); 328 else 329 pagefault_out_of_memory(); 330 } else if (fault & VM_FAULT_SIGSEGV) { 331 /* Kernel mode? Handle exceptions or die */ 332 if (!user_mode(regs)) 333 do_no_context(regs, fault); 334 else 335 do_sigsegv(regs, SEGV_MAPERR); 336 } else if (fault & VM_FAULT_SIGBUS) { 337 /* Kernel mode? Handle exceptions or die */ 338 if (!user_mode(regs)) 339 do_no_context(regs, fault); 340 else 341 do_sigbus(regs); 342 } else 343 BUG(); 344 break; 345 } 346 } 347 348 /* 349 * This routine handles page faults. It determines the address, 350 * and the problem, and then passes it off to one of the appropriate 351 * routines. 352 * 353 * interruption code (int_code): 354 * 04 Protection -> Write-Protection (suppression) 355 * 10 Segment translation -> Not present (nullification) 356 * 11 Page translation -> Not present (nullification) 357 * 3b Region third trans. -> Not present (nullification) 358 */ 359 static inline vm_fault_t do_exception(struct pt_regs *regs, int access) 360 { 361 struct gmap *gmap; 362 struct task_struct *tsk; 363 struct mm_struct *mm; 364 struct vm_area_struct *vma; 365 enum fault_type type; 366 unsigned long address; 367 unsigned int flags; 368 vm_fault_t fault; 369 bool is_write; 370 371 tsk = current; 372 /* 373 * The instruction that caused the program check has 374 * been nullified. Don't signal single step via SIGTRAP. 375 */ 376 clear_thread_flag(TIF_PER_TRAP); 377 378 if (kprobe_page_fault(regs, 14)) 379 return 0; 380 381 mm = tsk->mm; 382 address = get_fault_address(regs); 383 is_write = fault_is_write(regs); 384 385 /* 386 * Verify that the fault happened in user space, that 387 * we are not in an interrupt and that there is a 388 * user context. 389 */ 390 fault = VM_FAULT_BADCONTEXT; 391 type = get_fault_type(regs); 392 switch (type) { 393 case KERNEL_FAULT: 394 goto out; 395 case USER_FAULT: 396 case GMAP_FAULT: 397 if (faulthandler_disabled() || !mm) 398 goto out; 399 break; 400 } 401 402 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); 403 flags = FAULT_FLAG_DEFAULT; 404 if (user_mode(regs)) 405 flags |= FAULT_FLAG_USER; 406 if (is_write) 407 access = VM_WRITE; 408 if (access == VM_WRITE) 409 flags |= FAULT_FLAG_WRITE; 410 #ifdef CONFIG_PER_VMA_LOCK 411 if (!(flags & FAULT_FLAG_USER)) 412 goto lock_mmap; 413 vma = lock_vma_under_rcu(mm, address); 414 if (!vma) 415 goto lock_mmap; 416 if (!(vma->vm_flags & access)) { 417 vma_end_read(vma); 418 goto lock_mmap; 419 } 420 fault = handle_mm_fault(vma, address, flags | FAULT_FLAG_VMA_LOCK, regs); 421 vma_end_read(vma); 422 if (!(fault & VM_FAULT_RETRY)) { 423 count_vm_vma_lock_event(VMA_LOCK_SUCCESS); 424 if (likely(!(fault & VM_FAULT_ERROR))) 425 fault = 0; 426 goto out; 427 } 428 count_vm_vma_lock_event(VMA_LOCK_RETRY); 429 /* Quick path to respond to signals */ 430 if (fault_signal_pending(fault, regs)) { 431 fault = VM_FAULT_SIGNAL; 432 goto out; 433 } 434 lock_mmap: 435 #endif /* CONFIG_PER_VMA_LOCK */ 436 mmap_read_lock(mm); 437 438 gmap = NULL; 439 if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) { 440 gmap = (struct gmap *) S390_lowcore.gmap; 441 current->thread.gmap_addr = address; 442 current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE); 443 current->thread.gmap_int_code = regs->int_code & 0xffff; 444 address = __gmap_translate(gmap, address); 445 if (address == -EFAULT) { 446 fault = VM_FAULT_BADMAP; 447 goto out_up; 448 } 449 if (gmap->pfault_enabled) 450 flags |= FAULT_FLAG_RETRY_NOWAIT; 451 } 452 453 retry: 454 fault = VM_FAULT_BADMAP; 455 vma = find_vma(mm, address); 456 if (!vma) 457 goto out_up; 458 459 if (unlikely(vma->vm_start > address)) { 460 if (!(vma->vm_flags & VM_GROWSDOWN)) 461 goto out_up; 462 vma = expand_stack(mm, address); 463 if (!vma) 464 goto out; 465 } 466 467 /* 468 * Ok, we have a good vm_area for this memory access, so 469 * we can handle it.. 470 */ 471 fault = VM_FAULT_BADACCESS; 472 if (unlikely(!(vma->vm_flags & access))) 473 goto out_up; 474 475 /* 476 * If for any reason at all we couldn't handle the fault, 477 * make sure we exit gracefully rather than endlessly redo 478 * the fault. 479 */ 480 fault = handle_mm_fault(vma, address, flags, regs); 481 if (fault_signal_pending(fault, regs)) { 482 fault = VM_FAULT_SIGNAL; 483 if (flags & FAULT_FLAG_RETRY_NOWAIT) 484 goto out_up; 485 goto out; 486 } 487 488 /* The fault is fully completed (including releasing mmap lock) */ 489 if (fault & VM_FAULT_COMPLETED) { 490 if (gmap) { 491 mmap_read_lock(mm); 492 goto out_gmap; 493 } 494 fault = 0; 495 goto out; 496 } 497 498 if (unlikely(fault & VM_FAULT_ERROR)) 499 goto out_up; 500 501 if (fault & VM_FAULT_RETRY) { 502 if (IS_ENABLED(CONFIG_PGSTE) && gmap && 503 (flags & FAULT_FLAG_RETRY_NOWAIT)) { 504 /* 505 * FAULT_FLAG_RETRY_NOWAIT has been set, mmap_lock has 506 * not been released 507 */ 508 current->thread.gmap_pfault = 1; 509 fault = VM_FAULT_PFAULT; 510 goto out_up; 511 } 512 flags &= ~FAULT_FLAG_RETRY_NOWAIT; 513 flags |= FAULT_FLAG_TRIED; 514 mmap_read_lock(mm); 515 goto retry; 516 } 517 out_gmap: 518 if (IS_ENABLED(CONFIG_PGSTE) && gmap) { 519 address = __gmap_link(gmap, current->thread.gmap_addr, 520 address); 521 if (address == -EFAULT) { 522 fault = VM_FAULT_BADMAP; 523 goto out_up; 524 } 525 if (address == -ENOMEM) { 526 fault = VM_FAULT_OOM; 527 goto out_up; 528 } 529 } 530 fault = 0; 531 out_up: 532 mmap_read_unlock(mm); 533 out: 534 return fault; 535 } 536 537 void do_protection_exception(struct pt_regs *regs) 538 { 539 unsigned long trans_exc_code; 540 int access; 541 vm_fault_t fault; 542 543 trans_exc_code = regs->int_parm_long; 544 /* 545 * Protection exceptions are suppressing, decrement psw address. 546 * The exception to this rule are aborted transactions, for these 547 * the PSW already points to the correct location. 548 */ 549 if (!(regs->int_code & 0x200)) 550 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16); 551 /* 552 * Check for low-address protection. This needs to be treated 553 * as a special case because the translation exception code 554 * field is not guaranteed to contain valid data in this case. 555 */ 556 if (unlikely(!(trans_exc_code & 4))) { 557 do_low_address(regs); 558 return; 559 } 560 if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) { 561 regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) | 562 (regs->psw.addr & PAGE_MASK); 563 access = VM_EXEC; 564 fault = VM_FAULT_BADACCESS; 565 } else { 566 access = VM_WRITE; 567 fault = do_exception(regs, access); 568 } 569 if (unlikely(fault)) 570 do_fault_error(regs, fault); 571 } 572 NOKPROBE_SYMBOL(do_protection_exception); 573 574 void do_dat_exception(struct pt_regs *regs) 575 { 576 int access; 577 vm_fault_t fault; 578 579 access = VM_ACCESS_FLAGS; 580 fault = do_exception(regs, access); 581 if (unlikely(fault)) 582 do_fault_error(regs, fault); 583 } 584 NOKPROBE_SYMBOL(do_dat_exception); 585 586 #ifdef CONFIG_PFAULT 587 /* 588 * 'pfault' pseudo page faults routines. 589 */ 590 static int pfault_disable; 591 592 static int __init nopfault(char *str) 593 { 594 pfault_disable = 1; 595 return 1; 596 } 597 598 __setup("nopfault", nopfault); 599 600 struct pfault_refbk { 601 u16 refdiagc; 602 u16 reffcode; 603 u16 refdwlen; 604 u16 refversn; 605 u64 refgaddr; 606 u64 refselmk; 607 u64 refcmpmk; 608 u64 reserved; 609 } __attribute__ ((packed, aligned(8))); 610 611 static struct pfault_refbk pfault_init_refbk = { 612 .refdiagc = 0x258, 613 .reffcode = 0, 614 .refdwlen = 5, 615 .refversn = 2, 616 .refgaddr = __LC_LPP, 617 .refselmk = 1ULL << 48, 618 .refcmpmk = 1ULL << 48, 619 .reserved = __PF_RES_FIELD 620 }; 621 622 int pfault_init(void) 623 { 624 int rc; 625 626 if (pfault_disable) 627 return -1; 628 diag_stat_inc(DIAG_STAT_X258); 629 asm volatile( 630 " diag %1,%0,0x258\n" 631 "0: j 2f\n" 632 "1: la %0,8\n" 633 "2:\n" 634 EX_TABLE(0b,1b) 635 : "=d" (rc) 636 : "a" (&pfault_init_refbk), "m" (pfault_init_refbk) : "cc"); 637 return rc; 638 } 639 640 static struct pfault_refbk pfault_fini_refbk = { 641 .refdiagc = 0x258, 642 .reffcode = 1, 643 .refdwlen = 5, 644 .refversn = 2, 645 }; 646 647 void pfault_fini(void) 648 { 649 650 if (pfault_disable) 651 return; 652 diag_stat_inc(DIAG_STAT_X258); 653 asm volatile( 654 " diag %0,0,0x258\n" 655 "0: nopr %%r7\n" 656 EX_TABLE(0b,0b) 657 : : "a" (&pfault_fini_refbk), "m" (pfault_fini_refbk) : "cc"); 658 } 659 660 static DEFINE_SPINLOCK(pfault_lock); 661 static LIST_HEAD(pfault_list); 662 663 #define PF_COMPLETE 0x0080 664 665 /* 666 * The mechanism of our pfault code: if Linux is running as guest, runs a user 667 * space process and the user space process accesses a page that the host has 668 * paged out we get a pfault interrupt. 669 * 670 * This allows us, within the guest, to schedule a different process. Without 671 * this mechanism the host would have to suspend the whole virtual cpu until 672 * the page has been paged in. 673 * 674 * So when we get such an interrupt then we set the state of the current task 675 * to uninterruptible and also set the need_resched flag. Both happens within 676 * interrupt context(!). If we later on want to return to user space we 677 * recognize the need_resched flag and then call schedule(). It's not very 678 * obvious how this works... 679 * 680 * Of course we have a lot of additional fun with the completion interrupt (-> 681 * host signals that a page of a process has been paged in and the process can 682 * continue to run). This interrupt can arrive on any cpu and, since we have 683 * virtual cpus, actually appear before the interrupt that signals that a page 684 * is missing. 685 */ 686 static void pfault_interrupt(struct ext_code ext_code, 687 unsigned int param32, unsigned long param64) 688 { 689 struct task_struct *tsk; 690 __u16 subcode; 691 pid_t pid; 692 693 /* 694 * Get the external interruption subcode & pfault initial/completion 695 * signal bit. VM stores this in the 'cpu address' field associated 696 * with the external interrupt. 697 */ 698 subcode = ext_code.subcode; 699 if ((subcode & 0xff00) != __SUBCODE_MASK) 700 return; 701 inc_irq_stat(IRQEXT_PFL); 702 /* Get the token (= pid of the affected task). */ 703 pid = param64 & LPP_PID_MASK; 704 rcu_read_lock(); 705 tsk = find_task_by_pid_ns(pid, &init_pid_ns); 706 if (tsk) 707 get_task_struct(tsk); 708 rcu_read_unlock(); 709 if (!tsk) 710 return; 711 spin_lock(&pfault_lock); 712 if (subcode & PF_COMPLETE) { 713 /* signal bit is set -> a page has been swapped in by VM */ 714 if (tsk->thread.pfault_wait == 1) { 715 /* Initial interrupt was faster than the completion 716 * interrupt. pfault_wait is valid. Set pfault_wait 717 * back to zero and wake up the process. This can 718 * safely be done because the task is still sleeping 719 * and can't produce new pfaults. */ 720 tsk->thread.pfault_wait = 0; 721 list_del(&tsk->thread.list); 722 wake_up_process(tsk); 723 put_task_struct(tsk); 724 } else { 725 /* Completion interrupt was faster than initial 726 * interrupt. Set pfault_wait to -1 so the initial 727 * interrupt doesn't put the task to sleep. 728 * If the task is not running, ignore the completion 729 * interrupt since it must be a leftover of a PFAULT 730 * CANCEL operation which didn't remove all pending 731 * completion interrupts. */ 732 if (task_is_running(tsk)) 733 tsk->thread.pfault_wait = -1; 734 } 735 } else { 736 /* signal bit not set -> a real page is missing. */ 737 if (WARN_ON_ONCE(tsk != current)) 738 goto out; 739 if (tsk->thread.pfault_wait == 1) { 740 /* Already on the list with a reference: put to sleep */ 741 goto block; 742 } else if (tsk->thread.pfault_wait == -1) { 743 /* Completion interrupt was faster than the initial 744 * interrupt (pfault_wait == -1). Set pfault_wait 745 * back to zero and exit. */ 746 tsk->thread.pfault_wait = 0; 747 } else { 748 /* Initial interrupt arrived before completion 749 * interrupt. Let the task sleep. 750 * An extra task reference is needed since a different 751 * cpu may set the task state to TASK_RUNNING again 752 * before the scheduler is reached. */ 753 get_task_struct(tsk); 754 tsk->thread.pfault_wait = 1; 755 list_add(&tsk->thread.list, &pfault_list); 756 block: 757 /* Since this must be a userspace fault, there 758 * is no kernel task state to trample. Rely on the 759 * return to userspace schedule() to block. */ 760 __set_current_state(TASK_UNINTERRUPTIBLE); 761 set_tsk_need_resched(tsk); 762 set_preempt_need_resched(); 763 } 764 } 765 out: 766 spin_unlock(&pfault_lock); 767 put_task_struct(tsk); 768 } 769 770 static int pfault_cpu_dead(unsigned int cpu) 771 { 772 struct thread_struct *thread, *next; 773 struct task_struct *tsk; 774 775 spin_lock_irq(&pfault_lock); 776 list_for_each_entry_safe(thread, next, &pfault_list, list) { 777 thread->pfault_wait = 0; 778 list_del(&thread->list); 779 tsk = container_of(thread, struct task_struct, thread); 780 wake_up_process(tsk); 781 put_task_struct(tsk); 782 } 783 spin_unlock_irq(&pfault_lock); 784 return 0; 785 } 786 787 static int __init pfault_irq_init(void) 788 { 789 int rc; 790 791 rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt); 792 if (rc) 793 goto out_extint; 794 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP; 795 if (rc) 796 goto out_pfault; 797 irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL); 798 cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead", 799 NULL, pfault_cpu_dead); 800 return 0; 801 802 out_pfault: 803 unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt); 804 out_extint: 805 pfault_disable = 1; 806 return rc; 807 } 808 early_initcall(pfault_irq_init); 809 810 #endif /* CONFIG_PFAULT */ 811 812 #if IS_ENABLED(CONFIG_PGSTE) 813 814 void do_secure_storage_access(struct pt_regs *regs) 815 { 816 unsigned long addr = regs->int_parm_long & __FAIL_ADDR_MASK; 817 struct vm_area_struct *vma; 818 struct mm_struct *mm; 819 struct page *page; 820 struct gmap *gmap; 821 int rc; 822 823 /* 824 * bit 61 tells us if the address is valid, if it's not we 825 * have a major problem and should stop the kernel or send a 826 * SIGSEGV to the process. Unfortunately bit 61 is not 827 * reliable without the misc UV feature so we need to check 828 * for that as well. 829 */ 830 if (test_bit_inv(BIT_UV_FEAT_MISC, &uv_info.uv_feature_indications) && 831 !test_bit_inv(61, ®s->int_parm_long)) { 832 /* 833 * When this happens, userspace did something that it 834 * was not supposed to do, e.g. branching into secure 835 * memory. Trigger a segmentation fault. 836 */ 837 if (user_mode(regs)) { 838 send_sig(SIGSEGV, current, 0); 839 return; 840 } 841 842 /* 843 * The kernel should never run into this case and we 844 * have no way out of this situation. 845 */ 846 panic("Unexpected PGM 0x3d with TEID bit 61=0"); 847 } 848 849 switch (get_fault_type(regs)) { 850 case GMAP_FAULT: 851 mm = current->mm; 852 gmap = (struct gmap *)S390_lowcore.gmap; 853 mmap_read_lock(mm); 854 addr = __gmap_translate(gmap, addr); 855 mmap_read_unlock(mm); 856 if (IS_ERR_VALUE(addr)) { 857 do_fault_error(regs, VM_FAULT_BADMAP); 858 break; 859 } 860 fallthrough; 861 case USER_FAULT: 862 mm = current->mm; 863 mmap_read_lock(mm); 864 vma = find_vma(mm, addr); 865 if (!vma) { 866 mmap_read_unlock(mm); 867 do_fault_error(regs, VM_FAULT_BADMAP); 868 break; 869 } 870 page = follow_page(vma, addr, FOLL_WRITE | FOLL_GET); 871 if (IS_ERR_OR_NULL(page)) { 872 mmap_read_unlock(mm); 873 break; 874 } 875 if (arch_make_page_accessible(page)) 876 send_sig(SIGSEGV, current, 0); 877 put_page(page); 878 mmap_read_unlock(mm); 879 break; 880 case KERNEL_FAULT: 881 page = phys_to_page(addr); 882 if (unlikely(!try_get_page(page))) 883 break; 884 rc = arch_make_page_accessible(page); 885 put_page(page); 886 if (rc) 887 BUG(); 888 break; 889 default: 890 do_fault_error(regs, VM_FAULT_BADMAP); 891 WARN_ON_ONCE(1); 892 } 893 } 894 NOKPROBE_SYMBOL(do_secure_storage_access); 895 896 void do_non_secure_storage_access(struct pt_regs *regs) 897 { 898 unsigned long gaddr = regs->int_parm_long & __FAIL_ADDR_MASK; 899 struct gmap *gmap = (struct gmap *)S390_lowcore.gmap; 900 901 if (get_fault_type(regs) != GMAP_FAULT) { 902 do_fault_error(regs, VM_FAULT_BADMAP); 903 WARN_ON_ONCE(1); 904 return; 905 } 906 907 if (gmap_convert_to_secure(gmap, gaddr) == -EINVAL) 908 send_sig(SIGSEGV, current, 0); 909 } 910 NOKPROBE_SYMBOL(do_non_secure_storage_access); 911 912 void do_secure_storage_violation(struct pt_regs *regs) 913 { 914 unsigned long gaddr = regs->int_parm_long & __FAIL_ADDR_MASK; 915 struct gmap *gmap = (struct gmap *)S390_lowcore.gmap; 916 917 /* 918 * If the VM has been rebooted, its address space might still contain 919 * secure pages from the previous boot. 920 * Clear the page so it can be reused. 921 */ 922 if (!gmap_destroy_page(gmap, gaddr)) 923 return; 924 /* 925 * Either KVM messed up the secure guest mapping or the same 926 * page is mapped into multiple secure guests. 927 * 928 * This exception is only triggered when a guest 2 is running 929 * and can therefore never occur in kernel context. 930 */ 931 printk_ratelimited(KERN_WARNING 932 "Secure storage violation in task: %s, pid %d\n", 933 current->comm, current->pid); 934 send_sig(SIGSEGV, current, 0); 935 } 936 937 #endif /* CONFIG_PGSTE */ 938