syscall.c revision 155306
153152Smarcel/* $NetBSD: fault.c,v 1.45 2003/11/20 14:44:36 scw Exp $ */ 253152Smarcel 331567Ssef/*- 4101320Sjake * Copyright 2004 Olivier Houchard 5111178Sru * Copyright 2003 Wasabi Systems, Inc. 6101320Sjake * All rights reserved. 739908Ssef * 839908Ssef * Written by Steve C. Woodford for Wasabi Systems, Inc. 932367Ssef * 1031571Ssef * Redistribution and use in source and binary forms, with or without 1131571Ssef * modification, are permitted provided that the following conditions 1231567Ssef * are met: 1331567Ssef * 1. Redistributions of source code must retain the above copyright 1431567Ssef * notice, this list of conditions and the following disclaimer. 1531567Ssef * 2. Redistributions in binary form must reproduce the above copyright 1631567Ssef * notice, this list of conditions and the following disclaimer in the 1731567Ssef * documentation and/or other materials provided with the distribution. 1831567Ssef * 3. All advertising materials mentioning features or use of this software 1931567Ssef * must display the following acknowledgement: 2031567Ssef * This product includes software developed for the NetBSD Project by 2131567Ssef * Wasabi Systems, Inc. 2231567Ssef * 4. The name of Wasabi Systems, Inc. may not be used to endorse 2331567Ssef * or promote products derived from this software without specific prior 2431567Ssef * written permission. 2531567Ssef * 2631567Ssef * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND 2731567Ssef * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 2831567Ssef * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 2994432Sru * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL WASABI SYSTEMS, INC 3065849Sjkh * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 3131571Ssef * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 3231567Ssef * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38/*- 39 * Copyright (c) 1994-1997 Mark Brinicombe. 40 * Copyright (c) 1994 Brini. 41 * All rights reserved. 42 * 43 * This code is derived from software written for Brini by Mark Brinicombe 44 * 45 * Redistribution and use in source and binary forms, with or without 46 * modification, are permitted provided that the following conditions 47 * are met: 48 * 1. Redistributions of source code must retain the above copyright 49 * notice, this list of conditions and the following disclaimer. 50 * 2. Redistributions in binary form must reproduce the above copyright 51 * notice, this list of conditions and the following disclaimer in the 52 * documentation and/or other materials provided with the distribution. 53 * 3. All advertising materials mentioning features or use of this software 54 * must display the following acknowledgement: 55 * This product includes software developed by Brini. 56 * 4. The name of the company nor the name of the author may be used to 57 * endorse or promote products derived from this software without specific 58 * prior written permission. 59 * 60 * THIS SOFTWARE IS PROVIDED BY BRINI ``AS IS'' AND ANY EXPRESS OR IMPLIED 61 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF 62 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 63 * IN NO EVENT SHALL BRINI OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, 64 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 65 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 66 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 67 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 68 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 69 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 70 * SUCH DAMAGE. 71 * 72 * RiscBSD kernel project 73 * 74 * fault.c 75 * 76 * Fault handlers 77 * 78 * Created : 28/11/94 79 */ 80 81 82#include "opt_ktrace.h" 83 84#include <sys/cdefs.h> 85__FBSDID("$FreeBSD: head/sys/arm/arm/trap.c 155306 2006-02-04 17:01:19Z cognet $"); 86 87#include <sys/types.h> 88 89#include <sys/param.h> 90#include <sys/systm.h> 91#include <sys/proc.h> 92#include <sys/kernel.h> 93#include <sys/lock.h> 94#include <sys/mutex.h> 95#include <sys/syscall.h> 96#include <sys/sysent.h> 97#include <sys/signalvar.h> 98#include <sys/ktr.h> 99#ifdef KTRACE 100#include <sys/uio.h> 101#include <sys/ktrace.h> 102#endif 103#include <sys/ptrace.h> 104#include <sys/pioctl.h> 105 106#include <vm/vm.h> 107#include <vm/pmap.h> 108#include <vm/vm_kern.h> 109#include <vm/vm_map.h> 110#include <vm/vm_extern.h> 111 112#include <machine/cpuconf.h> 113#include <machine/vmparam.h> 114#include <machine/frame.h> 115#include <machine/katelib.h> 116#include <machine/cpu.h> 117#include <machine/intr.h> 118#include <machine/pcb.h> 119#include <machine/proc.h> 120#include <machine/swi.h> 121 122#include <security/audit/audit.h> 123 124#ifdef KDB 125#include <sys/kdb.h> 126#endif 127 128 129void swi_handler(trapframe_t *); 130void undefinedinstruction(trapframe_t *); 131 132#include <machine/disassem.h> 133#include <machine/machdep.h> 134 135extern char fusubailout[]; 136 137#ifdef DEBUG 138int last_fault_code; /* For the benefit of pmap_fault_fixup() */ 139#endif 140 141#if defined(CPU_ARM7TDMI) 142/* These CPUs may need data/prefetch abort fixups */ 143#define CPU_ABORT_FIXUP_REQUIRED 144#endif 145 146struct ksig { 147 int signb; 148 u_long code; 149}; 150struct data_abort { 151 int (*func)(trapframe_t *, u_int, u_int, struct thread *, struct ksig *); 152 const char *desc; 153}; 154 155static int dab_fatal(trapframe_t *, u_int, u_int, struct thread *, struct ksig *); 156static int dab_align(trapframe_t *, u_int, u_int, struct thread *, struct ksig *); 157static int dab_buserr(trapframe_t *, u_int, u_int, struct thread *, struct ksig *); 158 159static const struct data_abort data_aborts[] = { 160 {dab_fatal, "Vector Exception"}, 161 {dab_align, "Alignment Fault 1"}, 162 {dab_fatal, "Terminal Exception"}, 163 {dab_align, "Alignment Fault 3"}, 164 {dab_buserr, "External Linefetch Abort (S)"}, 165 {NULL, "Translation Fault (S)"}, 166 {dab_buserr, "External Linefetch Abort (P)"}, 167 {NULL, "Translation Fault (P)"}, 168 {dab_buserr, "External Non-Linefetch Abort (S)"}, 169 {NULL, "Domain Fault (S)"}, 170 {dab_buserr, "External Non-Linefetch Abort (P)"}, 171 {NULL, "Domain Fault (P)"}, 172 {dab_buserr, "External Translation Abort (L1)"}, 173 {NULL, "Permission Fault (S)"}, 174 {dab_buserr, "External Translation Abort (L2)"}, 175 {NULL, "Permission Fault (P)"} 176}; 177 178/* Determine if a fault came from user mode */ 179#define TRAP_USERMODE(tf) ((tf->tf_spsr & PSR_MODE) == PSR_USR32_MODE) 180 181/* Determine if 'x' is a permission fault */ 182#define IS_PERMISSION_FAULT(x) \ 183 (((1 << ((x) & FAULT_TYPE_MASK)) & \ 184 ((1 << FAULT_PERM_P) | (1 << FAULT_PERM_S))) != 0) 185 186static __inline void 187call_trapsignal(struct thread *td, int sig, u_long code) 188{ 189 ksiginfo_t ksi; 190 191 ksiginfo_init_trap(&ksi); 192 ksi.ksi_signo = sig; 193 ksi.ksi_code = (int)code; 194 trapsignal(td, &ksi); 195} 196 197static __inline int 198data_abort_fixup(trapframe_t *tf, u_int fsr, u_int far, struct thread *td, struct ksig *ksig) 199{ 200#ifdef CPU_ABORT_FIXUP_REQUIRED 201 int error; 202 203 /* Call the cpu specific data abort fixup routine */ 204 error = cpu_dataabt_fixup(tf); 205 if (__predict_true(error != ABORT_FIXUP_FAILED)) 206 return (error); 207 208 /* 209 * Oops, couldn't fix up the instruction 210 */ 211 printf("data_abort_fixup: fixup for %s mode data abort failed.\n", 212 TRAP_USERMODE(tf) ? "user" : "kernel"); 213 printf("pc = 0x%08x, opcode 0x%08x, insn = ", tf->tf_pc, 214 *((u_int *)tf->tf_pc)); 215 disassemble(tf->tf_pc); 216 217 /* Die now if this happened in kernel mode */ 218 if (!TRAP_USERMODE(tf)) 219 dab_fatal(tf, fsr, far, td, NULL, ksig); 220 221 return (error); 222#else 223 return (ABORT_FIXUP_OK); 224#endif /* CPU_ABORT_FIXUP_REQUIRED */ 225} 226 227void 228data_abort_handler(trapframe_t *tf) 229{ 230 struct vm_map *map; 231 struct pcb *pcb; 232 struct thread *td; 233 u_int user, far, fsr; 234 vm_prot_t ftype; 235 void *onfault; 236 vm_offset_t va; 237 u_int sticks = 0; 238 int error = 0; 239 struct ksig ksig; 240 struct proc *p; 241 242 243 /* Grab FAR/FSR before enabling interrupts */ 244 far = cpu_faultaddress(); 245 fsr = cpu_faultstatus(); 246#if 0 247 printf("data abort: %p (from %p %p)\n", (void*)far, (void*)tf->tf_pc, 248 (void*)tf->tf_svc_lr); 249#endif 250 251 /* Update vmmeter statistics */ 252#if 0 253 vmexp.traps++; 254#endif 255 256 td = curthread; 257 p = td->td_proc; 258 259 PCPU_LAZY_INC(cnt.v_trap); 260 /* Data abort came from user mode? */ 261 user = TRAP_USERMODE(tf); 262 263 if (user) { 264 sticks = td->td_sticks; td->td_frame = tf; 265 if (td->td_ucred != td->td_proc->p_ucred) 266 cred_update_thread(td); 267 if (td->td_pflags & TDP_SA) 268 thread_user_enter(td); 269 270 } 271 /* Grab the current pcb */ 272 pcb = td->td_pcb; 273 /* Re-enable interrupts if they were enabled previously */ 274 if (td->td_md.md_spinlock_count == 0 && 275 __predict_true(tf->tf_spsr & I32_bit) == 0) 276 enable_interrupts(I32_bit); 277 278 /* Invoke the appropriate handler, if necessary */ 279 if (__predict_false(data_aborts[fsr & FAULT_TYPE_MASK].func != NULL)) { 280 if ((data_aborts[fsr & FAULT_TYPE_MASK].func)(tf, fsr, far, 281 td, &ksig)) { 282 goto do_trapsignal; 283 } 284 goto out; 285 } 286 287 /* 288 * At this point, we're dealing with one of the following data aborts: 289 * 290 * FAULT_TRANS_S - Translation -- Section 291 * FAULT_TRANS_P - Translation -- Page 292 * FAULT_DOMAIN_S - Domain -- Section 293 * FAULT_DOMAIN_P - Domain -- Page 294 * FAULT_PERM_S - Permission -- Section 295 * FAULT_PERM_P - Permission -- Page 296 * 297 * These are the main virtual memory-related faults signalled by 298 * the MMU. 299 */ 300 301 /* fusubailout is used by [fs]uswintr to avoid page faulting */ 302 if (__predict_false(pcb->pcb_onfault == fusubailout)) { 303 tf->tf_r0 = EFAULT; 304 tf->tf_pc = (register_t)(intptr_t) pcb->pcb_onfault; 305 return; 306 } 307 308 /* 309 * Make sure the Program Counter is sane. We could fall foul of 310 * someone executing Thumb code, in which case the PC might not 311 * be word-aligned. This would cause a kernel alignment fault 312 * further down if we have to decode the current instruction. 313 * XXX: It would be nice to be able to support Thumb at some point. 314 */ 315 if (__predict_false((tf->tf_pc & 3) != 0)) { 316 if (user) { 317 /* 318 * Give the user an illegal instruction signal. 319 */ 320 /* Deliver a SIGILL to the process */ 321 ksig.signb = SIGILL; 322 ksig.code = 0; 323 goto do_trapsignal; 324 } 325 326 /* 327 * The kernel never executes Thumb code. 328 */ 329 printf("\ndata_abort_fault: Misaligned Kernel-mode " 330 "Program Counter\n"); 331 dab_fatal(tf, fsr, far, td, &ksig); 332 } 333 334 /* See if the cpu state needs to be fixed up */ 335 switch (data_abort_fixup(tf, fsr, far, td, &ksig)) { 336 case ABORT_FIXUP_RETURN: 337 return; 338 case ABORT_FIXUP_FAILED: 339 /* Deliver a SIGILL to the process */ 340 ksig.signb = SIGILL; 341 ksig.code = 0; 342 goto do_trapsignal; 343 default: 344 break; 345 } 346 347 va = trunc_page((vm_offset_t)far); 348 349 /* 350 * It is only a kernel address space fault iff: 351 * 1. user == 0 and 352 * 2. pcb_onfault not set or 353 * 3. pcb_onfault set and not LDRT/LDRBT/STRT/STRBT instruction. 354 */ 355 if (user == 0 && (va >= VM_MIN_KERNEL_ADDRESS || 356 (va < VM_MIN_ADDRESS && vector_page == ARM_VECTORS_LOW)) && 357 __predict_true((pcb->pcb_onfault == NULL || 358 (ReadWord(tf->tf_pc) & 0x05200000) != 0x04200000))) { 359 map = kernel_map; 360 361 /* Was the fault due to the FPE/IPKDB ? */ 362 if (__predict_false((tf->tf_spsr & PSR_MODE)==PSR_UND32_MODE)) { 363 364 /* 365 * Force exit via userret() 366 * This is necessary as the FPE is an extension to 367 * userland that actually runs in a priveledged mode 368 * but uses USR mode permissions for its accesses. 369 */ 370 user = 1; 371 ksig.signb = SIGSEGV; 372 ksig.code = 0; 373 goto do_trapsignal; 374 } 375 } else { 376 map = &td->td_proc->p_vmspace->vm_map; 377 } 378 379 /* 380 * We need to know whether the page should be mapped 381 * as R or R/W. The MMU does not give us the info as 382 * to whether the fault was caused by a read or a write. 383 * 384 * However, we know that a permission fault can only be 385 * the result of a write to a read-only location, so 386 * we can deal with those quickly. 387 * 388 * Otherwise we need to disassemble the instruction 389 * responsible to determine if it was a write. 390 */ 391 if (IS_PERMISSION_FAULT(fsr)) { 392 ftype = VM_PROT_WRITE; 393 } else { 394 u_int insn = ReadWord(tf->tf_pc); 395 396 if (((insn & 0x0c100000) == 0x04000000) || /* STR/STRB */ 397 ((insn & 0x0e1000b0) == 0x000000b0) || /* STRH/STRD */ 398 ((insn & 0x0a100000) == 0x08000000)) /* STM/CDT */ 399 { 400 ftype = VM_PROT_WRITE; 401 } 402 else 403 if ((insn & 0x0fb00ff0) == 0x01000090) /* SWP */ 404 ftype = VM_PROT_READ | VM_PROT_WRITE; 405 else 406 ftype = VM_PROT_READ; 407 } 408 409 /* 410 * See if the fault is as a result of ref/mod emulation, 411 * or domain mismatch. 412 */ 413#ifdef DEBUG 414 last_fault_code = fsr; 415#endif 416 if (pmap_fault_fixup(vmspace_pmap(td->td_proc->p_vmspace), va, ftype, 417 user)) { 418 goto out; 419 } 420 421 onfault = pcb->pcb_onfault; 422 pcb->pcb_onfault = NULL; 423 if (map != kernel_map) { 424 PROC_LOCK(p); 425 p->p_lock++; 426 PROC_UNLOCK(p); 427 } 428 error = vm_fault(map, va, ftype, (ftype & VM_PROT_WRITE) ? 429 VM_FAULT_DIRTY : VM_FAULT_NORMAL); 430 pcb->pcb_onfault = onfault; 431 if (__predict_true(error == 0)) { 432 goto out; 433 } 434 435 if (map != kernel_map) { 436 PROC_LOCK(p); 437 p->p_lock--; 438 PROC_UNLOCK(p); 439 } 440 if (user == 0) { 441 if (pcb->pcb_onfault) { 442 tf->tf_r0 = error; 443 tf->tf_pc = (register_t)(intptr_t) pcb->pcb_onfault; 444 return; 445 } 446 447 printf("\nvm_fault(%p, %x, %x, 0) -> %x\n", map, va, ftype, 448 error); 449 dab_fatal(tf, fsr, far, td, &ksig); 450 } 451 452 453 if (error == ENOMEM) { 454 printf("VM: pid %d (%s), uid %d killed: " 455 "out of swap\n", td->td_proc->p_pid, td->td_proc->p_comm, 456 (td->td_proc->p_ucred) ? 457 td->td_proc->p_ucred->cr_uid : -1); 458 ksig.signb = SIGKILL; 459 } else { 460 ksig.signb = SIGSEGV; 461 } 462 ksig.code = 0; 463do_trapsignal: 464 call_trapsignal(td, ksig.signb, ksig.code); 465out: 466 /* If returning to user mode, make sure to invoke userret() */ 467 if (user) 468 userret(td, tf, sticks); 469} 470 471/* 472 * dab_fatal() handles the following data aborts: 473 * 474 * FAULT_WRTBUF_0 - Vector Exception 475 * FAULT_WRTBUF_1 - Terminal Exception 476 * 477 * We should never see these on a properly functioning system. 478 * 479 * This function is also called by the other handlers if they 480 * detect a fatal problem. 481 * 482 * Note: If 'l' is NULL, we assume we're dealing with a prefetch abort. 483 */ 484static int 485dab_fatal(trapframe_t *tf, u_int fsr, u_int far, struct thread *td, struct ksig *ksig) 486{ 487 const char *mode; 488 489 mode = TRAP_USERMODE(tf) ? "user" : "kernel"; 490 491 if (td != NULL) { 492 printf("Fatal %s mode data abort: '%s'\n", mode, 493 data_aborts[fsr & FAULT_TYPE_MASK].desc); 494 printf("trapframe: %p\nFSR=%08x, FAR=", tf, fsr); 495 if ((fsr & FAULT_IMPRECISE) == 0) 496 printf("%08x, ", far); 497 else 498 printf("Invalid, "); 499 printf("spsr=%08x\n", tf->tf_spsr); 500 } else { 501 printf("Fatal %s mode prefetch abort at 0x%08x\n", 502 mode, tf->tf_pc); 503 printf("trapframe: %p, spsr=%08x\n", tf, tf->tf_spsr); 504 } 505 506 printf("r0 =%08x, r1 =%08x, r2 =%08x, r3 =%08x\n", 507 tf->tf_r0, tf->tf_r1, tf->tf_r2, tf->tf_r3); 508 printf("r4 =%08x, r5 =%08x, r6 =%08x, r7 =%08x\n", 509 tf->tf_r4, tf->tf_r5, tf->tf_r6, tf->tf_r7); 510 printf("r8 =%08x, r9 =%08x, r10=%08x, r11=%08x\n", 511 tf->tf_r8, tf->tf_r9, tf->tf_r10, tf->tf_r11); 512 printf("r12=%08x, ", tf->tf_r12); 513 514 if (TRAP_USERMODE(tf)) 515 printf("usp=%08x, ulr=%08x", 516 tf->tf_usr_sp, tf->tf_usr_lr); 517 else 518 printf("ssp=%08x, slr=%08x", 519 tf->tf_svc_sp, tf->tf_svc_lr); 520 printf(", pc =%08x\n\n", tf->tf_pc); 521 522#ifdef KDB 523 kdb_trap(fsr, 0, tf); 524#endif 525 panic("Fatal abort"); 526 /*NOTREACHED*/ 527} 528 529/* 530 * dab_align() handles the following data aborts: 531 * 532 * FAULT_ALIGN_0 - Alignment fault 533 * FAULT_ALIGN_0 - Alignment fault 534 * 535 * These faults are fatal if they happen in kernel mode. Otherwise, we 536 * deliver a bus error to the process. 537 */ 538static int 539dab_align(trapframe_t *tf, u_int fsr, u_int far, struct thread *td, struct ksig *ksig) 540{ 541 542 /* Alignment faults are always fatal if they occur in kernel mode */ 543 if (!TRAP_USERMODE(tf)) { 544 if (!td || !td->td_pcb->pcb_onfault) 545 dab_fatal(tf, fsr, far, td, ksig); 546 tf->tf_r0 = EFAULT; 547 tf->tf_pc = (int)td->td_pcb->pcb_onfault; 548 return (0); 549 } 550 551 /* pcb_onfault *must* be NULL at this point */ 552 553 /* See if the cpu state needs to be fixed up */ 554 (void) data_abort_fixup(tf, fsr, far, td, ksig); 555 556 /* Deliver a bus error signal to the process */ 557 ksig->code = 0; 558 ksig->signb = SIGBUS; 559 td->td_frame = tf; 560 561 return (1); 562} 563 564/* 565 * dab_buserr() handles the following data aborts: 566 * 567 * FAULT_BUSERR_0 - External Abort on Linefetch -- Section 568 * FAULT_BUSERR_1 - External Abort on Linefetch -- Page 569 * FAULT_BUSERR_2 - External Abort on Non-linefetch -- Section 570 * FAULT_BUSERR_3 - External Abort on Non-linefetch -- Page 571 * FAULT_BUSTRNL1 - External abort on Translation -- Level 1 572 * FAULT_BUSTRNL2 - External abort on Translation -- Level 2 573 * 574 * If pcb_onfault is set, flag the fault and return to the handler. 575 * If the fault occurred in user mode, give the process a SIGBUS. 576 * 577 * Note: On XScale, FAULT_BUSERR_0, FAULT_BUSERR_1, and FAULT_BUSERR_2 578 * can be flagged as imprecise in the FSR. This causes a real headache 579 * since some of the machine state is lost. In this case, tf->tf_pc 580 * may not actually point to the offending instruction. In fact, if 581 * we've taken a double abort fault, it generally points somewhere near 582 * the top of "data_abort_entry" in exception.S. 583 * 584 * In all other cases, these data aborts are considered fatal. 585 */ 586static int 587dab_buserr(trapframe_t *tf, u_int fsr, u_int far, struct thread *td, struct ksig *ksig) 588{ 589 struct pcb *pcb = td->td_pcb; 590 591#ifdef __XSCALE__ 592 if ((fsr & FAULT_IMPRECISE) != 0 && 593 (tf->tf_spsr & PSR_MODE) == PSR_ABT32_MODE) { 594 /* 595 * Oops, an imprecise, double abort fault. We've lost the 596 * r14_abt/spsr_abt values corresponding to the original 597 * abort, and the spsr saved in the trapframe indicates 598 * ABT mode. 599 */ 600 tf->tf_spsr &= ~PSR_MODE; 601 602 /* 603 * We use a simple heuristic to determine if the double abort 604 * happened as a result of a kernel or user mode access. 605 * If the current trapframe is at the top of the kernel stack, 606 * the fault _must_ have come from user mode. 607 */ 608 if (tf != ((trapframe_t *)pcb->un_32.pcb32_sp) - 1) { 609 /* 610 * Kernel mode. We're either about to die a 611 * spectacular death, or pcb_onfault will come 612 * to our rescue. Either way, the current value 613 * of tf->tf_pc is irrelevant. 614 */ 615 tf->tf_spsr |= PSR_SVC32_MODE; 616 if (pcb->pcb_onfault == NULL) 617 printf("\nKernel mode double abort!\n"); 618 } else { 619 /* 620 * User mode. We've lost the program counter at the 621 * time of the fault (not that it was accurate anyway; 622 * it's not called an imprecise fault for nothing). 623 * About all we can do is copy r14_usr to tf_pc and 624 * hope for the best. The process is about to get a 625 * SIGBUS, so it's probably history anyway. 626 */ 627 tf->tf_spsr |= PSR_USR32_MODE; 628 tf->tf_pc = tf->tf_usr_lr; 629 } 630 } 631 632 /* FAR is invalid for imprecise exceptions */ 633 if ((fsr & FAULT_IMPRECISE) != 0) 634 far = 0; 635#endif /* __XSCALE__ */ 636 637 if (pcb->pcb_onfault) { 638 tf->tf_r0 = EFAULT; 639 tf->tf_pc = (register_t)(intptr_t) pcb->pcb_onfault; 640 return (0); 641 } 642 643 /* See if the cpu state needs to be fixed up */ 644 (void) data_abort_fixup(tf, fsr, far, td, ksig); 645 646 /* 647 * At this point, if the fault happened in kernel mode, we're toast 648 */ 649 if (!TRAP_USERMODE(tf)) 650 dab_fatal(tf, fsr, far, td, ksig); 651 652 /* Deliver a bus error signal to the process */ 653 ksig->signb = SIGBUS; 654 ksig->code = 0; 655 td->td_frame = tf; 656 657 return (1); 658} 659 660static __inline int 661prefetch_abort_fixup(trapframe_t *tf, struct ksig *ksig) 662{ 663#ifdef CPU_ABORT_FIXUP_REQUIRED 664 int error; 665 666 /* Call the cpu specific prefetch abort fixup routine */ 667 error = cpu_prefetchabt_fixup(tf); 668 if (__predict_true(error != ABORT_FIXUP_FAILED)) 669 return (error); 670 671 /* 672 * Oops, couldn't fix up the instruction 673 */ 674 printf( 675 "prefetch_abort_fixup: fixup for %s mode prefetch abort failed.\n", 676 TRAP_USERMODE(tf) ? "user" : "kernel"); 677 printf("pc = 0x%08x, opcode 0x%08x, insn = ", tf->tf_pc, 678 *((u_int *)tf->tf_pc)); 679 disassemble(tf->tf_pc); 680 681 /* Die now if this happened in kernel mode */ 682 if (!TRAP_USERMODE(tf)) 683 dab_fatal(tf, 0, tf->tf_pc, NULL, ksig); 684 685 return (error); 686#else 687 return (ABORT_FIXUP_OK); 688#endif /* CPU_ABORT_FIXUP_REQUIRED */ 689} 690 691/* 692 * void prefetch_abort_handler(trapframe_t *tf) 693 * 694 * Abort handler called when instruction execution occurs at 695 * a non existent or restricted (access permissions) memory page. 696 * If the address is invalid and we were in SVC mode then panic as 697 * the kernel should never prefetch abort. 698 * If the address is invalid and the page is mapped then the user process 699 * does no have read permission so send it a signal. 700 * Otherwise fault the page in and try again. 701 */ 702void 703prefetch_abort_handler(trapframe_t *tf) 704{ 705 struct thread *td; 706 struct proc * p; 707 struct vm_map *map; 708 vm_offset_t fault_pc, va; 709 int error = 0; 710 u_int sticks = 0; 711 struct ksig ksig; 712 713 714#if 0 715 /* Update vmmeter statistics */ 716 uvmexp.traps++; 717#endif 718#if 0 719 printf("prefetch abort handler: %p %p\n", (void*)tf->tf_pc, 720 (void*)tf->tf_usr_lr); 721#endif 722 723 td = curthread; 724 p = td->td_proc; 725 PCPU_LAZY_INC(cnt.v_trap); 726 727 if (TRAP_USERMODE(tf)) { 728 td->td_frame = tf; 729 if (td->td_ucred != td->td_proc->p_ucred) 730 cred_update_thread(td); 731 if (td->td_proc->p_flag & P_SA) 732 thread_user_enter(td); 733 } 734 fault_pc = tf->tf_pc; 735 if (td->td_md.md_spinlock_count == 0 && 736 __predict_true((tf->tf_spsr & I32_bit) == 0)) 737 enable_interrupts(I32_bit); 738 739 740 /* See if the cpu state needs to be fixed up */ 741 switch (prefetch_abort_fixup(tf, &ksig)) { 742 case ABORT_FIXUP_RETURN: 743 return; 744 case ABORT_FIXUP_FAILED: 745 /* Deliver a SIGILL to the process */ 746 ksig.signb = SIGILL; 747 ksig.code = 0; 748 td->td_frame = tf; 749 goto do_trapsignal; 750 default: 751 break; 752 } 753 754 /* Prefetch aborts cannot happen in kernel mode */ 755 if (__predict_false(!TRAP_USERMODE(tf))) 756 dab_fatal(tf, 0, tf->tf_pc, NULL, &ksig); 757 sticks = td->td_sticks; 758 759 760 /* Ok validate the address, can only execute in USER space */ 761 if (__predict_false(fault_pc >= VM_MAXUSER_ADDRESS || 762 (fault_pc < VM_MIN_ADDRESS && vector_page == ARM_VECTORS_LOW))) { 763 ksig.signb = SIGSEGV; 764 ksig.code = 0; 765 goto do_trapsignal; 766 } 767 768 map = &td->td_proc->p_vmspace->vm_map; 769 va = trunc_page(fault_pc); 770 771 /* 772 * See if the pmap can handle this fault on its own... 773 */ 774#ifdef DEBUG 775 last_fault_code = -1; 776#endif 777 if (pmap_fault_fixup(map->pmap, va, VM_PROT_READ, 1)) 778 goto out; 779 780 if (map != kernel_map) { 781 PROC_LOCK(p); 782 p->p_lock++; 783 PROC_UNLOCK(p); 784 } 785 786 error = vm_fault(map, va, VM_PROT_READ | VM_PROT_EXECUTE, 787 VM_FAULT_NORMAL); 788 if (map != kernel_map) { 789 PROC_LOCK(p); 790 p->p_lock--; 791 PROC_UNLOCK(p); 792 } 793 794 if (__predict_true(error == 0)) 795 goto out; 796 797 if (error == ENOMEM) { 798 printf("VM: pid %d (%s), uid %d killed: " 799 "out of swap\n", td->td_proc->p_pid, td->td_proc->p_comm, 800 (td->td_proc->p_ucred) ? 801 td->td_proc->p_ucred->cr_uid : -1); 802 ksig.signb = SIGKILL; 803 } else { 804 ksig.signb = SIGSEGV; 805 } 806 ksig.code = 0; 807 808do_trapsignal: 809 call_trapsignal(td, ksig.signb, ksig.code); 810 811out: 812 userret(td, tf, sticks); 813 814} 815 816extern int badaddr_read_1(const uint8_t *, uint8_t *); 817extern int badaddr_read_2(const uint16_t *, uint16_t *); 818extern int badaddr_read_4(const uint32_t *, uint32_t *); 819/* 820 * Tentatively read an 8, 16, or 32-bit value from 'addr'. 821 * If the read succeeds, the value is written to 'rptr' and zero is returned. 822 * Else, return EFAULT. 823 */ 824int 825badaddr_read(void *addr, size_t size, void *rptr) 826{ 827 union { 828 uint8_t v1; 829 uint16_t v2; 830 uint32_t v4; 831 } u; 832 int rv; 833 834 cpu_drain_writebuf(); 835 836 /* Read from the test address. */ 837 switch (size) { 838 case sizeof(uint8_t): 839 rv = badaddr_read_1(addr, &u.v1); 840 if (rv == 0 && rptr) 841 *(uint8_t *) rptr = u.v1; 842 break; 843 844 case sizeof(uint16_t): 845 rv = badaddr_read_2(addr, &u.v2); 846 if (rv == 0 && rptr) 847 *(uint16_t *) rptr = u.v2; 848 break; 849 850 case sizeof(uint32_t): 851 rv = badaddr_read_4(addr, &u.v4); 852 if (rv == 0 && rptr) 853 *(uint32_t *) rptr = u.v4; 854 break; 855 856 default: 857 panic("badaddr: invalid size (%lu)", (u_long) size); 858 } 859 860 /* Return EFAULT if the address was invalid, else zero */ 861 return (rv); 862} 863 864#define MAXARGS 8 865static void 866syscall(struct thread *td, trapframe_t *frame, u_int32_t insn) 867{ 868 struct proc *p = td->td_proc; 869 int code, error; 870 u_int nap, nargs; 871 register_t *ap, *args, copyargs[MAXARGS]; 872 struct sysent *callp; 873 int locked = 0; 874 u_int sticks = 0; 875 876 PCPU_LAZY_INC(cnt.v_syscall); 877 sticks = td->td_sticks; 878 if (td->td_ucred != td->td_proc->p_ucred) 879 cred_update_thread(td); 880 switch (insn & SWI_OS_MASK) { 881 case 0: /* XXX: we need our own one. */ 882 nap = 4; 883 break; 884 default: 885 call_trapsignal(td, SIGILL, 0); 886 userret(td, frame, td->td_sticks); 887 return; 888 } 889 code = insn & 0x000fffff; 890 sticks = td->td_sticks; 891 ap = &frame->tf_r0; 892 if (code == SYS_syscall) { 893 code = *ap++; 894 895 nap--; 896 } else if (code == SYS___syscall) { 897 code = *ap++; 898 nap -= 2; 899 ap++; 900 } 901 if (p->p_sysent->sv_mask) 902 code &= p->p_sysent->sv_mask; 903 if (code >= p->p_sysent->sv_size) 904 callp = &p->p_sysent->sv_table[0]; 905 else 906 callp = &p->p_sysent->sv_table[code]; 907 nargs = callp->sy_narg & SYF_ARGMASK; 908 memcpy(copyargs, ap, nap * sizeof(register_t)); 909 if (nargs > nap) { 910 error = copyin((void *)frame->tf_usr_sp, copyargs + nap, 911 (nargs - nap) * sizeof(register_t)); 912 if (error) 913 goto bad; 914 } 915 args = copyargs; 916 error = 0; 917#ifdef KTRACE 918 if (KTRPOINT(td, KTR_SYSCALL)) 919 ktrsyscall(code, nargs, args); 920#endif 921 922 CTR4(KTR_SYSC, "syscall enter thread %p pid %d proc %s code %d", td, 923 td->td_proc->p_pid, td->td_proc->p_comm, code); 924 if ((callp->sy_narg & SYF_MPSAFE) == 0) 925 mtx_lock(&Giant); 926 locked = 1; 927 if (error == 0) { 928 td->td_retval[0] = 0; 929 td->td_retval[1] = 0; 930 STOPEVENT(p, S_SCE, (callp->sy_narg & SYF_ARGMASK)); 931 PTRACESTOP_SC(p, td, S_PT_SCE); 932 AUDIT_SYSCALL_ENTER(code, td); 933 error = (*callp->sy_call)(td, args); 934 AUDIT_SYSCALL_EXIT(error, td); 935 } 936 switch (error) { 937 case 0: 938#ifdef __ARMEB__ 939 if ((insn & 0x000fffff) && 940 (code != SYS_lseek)) { 941 /* 942 * 64-bit return, 32-bit syscall. Fixup byte order 943 */ 944 frame->tf_r0 = 0; 945 frame->tf_r1 = td->td_retval[0]; 946 } else { 947 frame->tf_r0 = td->td_retval[0]; 948 frame->tf_r1 = td->td_retval[1]; 949 } 950#else 951 frame->tf_r0 = td->td_retval[0]; 952 frame->tf_r1 = td->td_retval[1]; 953#endif 954 frame->tf_spsr &= ~PSR_C_bit; /* carry bit */ 955 break; 956 957 case ERESTART: 958 /* 959 * Reconstruct the pc to point at the swi. 960 */ 961 frame->tf_pc -= INSN_SIZE; 962 break; 963 case EJUSTRETURN: 964 /* nothing to do */ 965 break; 966 default: 967bad: 968 frame->tf_r0 = error; 969 frame->tf_spsr |= PSR_C_bit; /* carry bit */ 970 break; 971 } 972 if (locked && (callp->sy_narg & SYF_MPSAFE) == 0) 973 mtx_unlock(&Giant); 974 975 976 userret(td, frame, sticks); 977 CTR4(KTR_SYSC, "syscall exit thread %p pid %d proc %s code %d", td, 978 td->td_proc->p_pid, td->td_proc->p_comm, code); 979 980 STOPEVENT(p, S_SCX, code); 981 PTRACESTOP_SC(p, td, S_PT_SCX); 982#ifdef KTRACE 983 if (KTRPOINT(td, KTR_SYSRET)) 984 ktrsysret(code, error, td->td_retval[0]); 985#endif 986 mtx_assert(&sched_lock, MA_NOTOWNED); 987 mtx_assert(&Giant, MA_NOTOWNED); 988} 989 990void 991swi_handler(trapframe_t *frame) 992{ 993 struct thread *td = curthread; 994 uint32_t insn; 995 996 td->td_frame = frame; 997 998 if (td->td_proc->p_flag & P_SA) 999 thread_user_enter(td); 1000 /* 1001 * Make sure the program counter is correctly aligned so we 1002 * don't take an alignment fault trying to read the opcode. 1003 */ 1004 if (__predict_false(((frame->tf_pc - INSN_SIZE) & 3) != 0)) { 1005 call_trapsignal(td, SIGILL, 0); 1006 userret(td, frame, td->td_sticks); 1007 return; 1008 } 1009 insn = *(u_int32_t *)(frame->tf_pc - INSN_SIZE); 1010 /* 1011 * Enable interrupts if they were enabled before the exception. 1012 * Since all syscalls *should* come from user mode it will always 1013 * be safe to enable them, but check anyway. 1014 */ 1015 if (td->td_md.md_spinlock_count == 0 && !(frame->tf_spsr & I32_bit)) 1016 enable_interrupts(I32_bit); 1017 1018 syscall(td, frame, insn); 1019} 1020 1021