syscall.c revision 250929
1/*	$NetBSD: fault.c,v 1.45 2003/11/20 14:44:36 scw Exp $	*/
2
3/*-
4 * Copyright 2004 Olivier Houchard
5 * Copyright 2003 Wasabi Systems, Inc.
6 * All rights reserved.
7 *
8 * Written by Steve C. Woodford for Wasabi Systems, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 *    notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 *    notice, this list of conditions and the following disclaimer in the
17 *    documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 *    must display the following acknowledgement:
20 *      This product includes software developed for the NetBSD Project by
21 *      Wasabi Systems, Inc.
22 * 4. The name of Wasabi Systems, Inc. may not be used to endorse
23 *    or promote products derived from this software without specific prior
24 *    written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 * 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 250929 2013-05-23 12:15:23Z gber $");
86
87#include <sys/param.h>
88#include <sys/systm.h>
89#include <sys/proc.h>
90#include <sys/kernel.h>
91#include <sys/lock.h>
92#include <sys/mutex.h>
93#include <sys/syscall.h>
94#include <sys/sysent.h>
95#include <sys/signalvar.h>
96#include <sys/ktr.h>
97#ifdef KTRACE
98#include <sys/uio.h>
99#include <sys/ktrace.h>
100#endif
101#include <sys/ptrace.h>
102#include <sys/pioctl.h>
103
104#include <vm/vm.h>
105#include <vm/pmap.h>
106#include <vm/vm_kern.h>
107#include <vm/vm_map.h>
108#include <vm/vm_extern.h>
109
110#include <machine/cpuconf.h>
111#include <machine/vmparam.h>
112#include <machine/frame.h>
113#include <machine/cpu.h>
114#include <machine/intr.h>
115#include <machine/pcb.h>
116#include <machine/proc.h>
117#include <machine/swi.h>
118
119#include <security/audit/audit.h>
120
121#ifdef KDB
122#include <sys/kdb.h>
123#endif
124
125
126void swi_handler(trapframe_t *);
127void undefinedinstruction(trapframe_t *);
128
129#include <machine/disassem.h>
130#include <machine/machdep.h>
131
132extern char fusubailout[];
133
134#ifdef DEBUG
135int last_fault_code;	/* For the benefit of pmap_fault_fixup() */
136#endif
137
138#if defined(CPU_ARM7TDMI)
139/* These CPUs may need data/prefetch abort fixups */
140#define	CPU_ABORT_FIXUP_REQUIRED
141#endif
142
143struct ksig {
144	int signb;
145	u_long code;
146};
147struct data_abort {
148	int (*func)(trapframe_t *, u_int, u_int, struct thread *, struct ksig *);
149	const char *desc;
150};
151
152static int dab_fatal(trapframe_t *, u_int, u_int, struct thread *, struct ksig *);
153static int dab_align(trapframe_t *, u_int, u_int, struct thread *, struct ksig *);
154static int dab_buserr(trapframe_t *, u_int, u_int, struct thread *, struct ksig *);
155
156static const struct data_abort data_aborts[] = {
157	{dab_fatal,	"Vector Exception"},
158	{dab_align,	"Alignment Fault 1"},
159	{dab_fatal,	"Terminal Exception"},
160	{dab_align,	"Alignment Fault 3"},
161	{dab_buserr,	"External Linefetch Abort (S)"},
162	{NULL,		"Translation Fault (S)"},
163#if (ARM_MMU_V6 + ARM_MMU_V7) != 0
164	{NULL,		"Translation Flag Fault"},
165#else
166	{dab_buserr,	"External Linefetch Abort (P)"},
167#endif
168	{NULL,		"Translation Fault (P)"},
169	{dab_buserr,	"External Non-Linefetch Abort (S)"},
170	{NULL,		"Domain Fault (S)"},
171	{dab_buserr,	"External Non-Linefetch Abort (P)"},
172	{NULL,		"Domain Fault (P)"},
173	{dab_buserr,	"External Translation Abort (L1)"},
174	{NULL,		"Permission Fault (S)"},
175	{dab_buserr,	"External Translation Abort (L2)"},
176	{NULL,		"Permission Fault (P)"}
177};
178
179/* Determine if a fault came from user mode */
180#define	TRAP_USERMODE(tf)	((tf->tf_spsr & PSR_MODE) == PSR_USR32_MODE)
181
182/* Determine if 'x' is a permission fault */
183#define	IS_PERMISSION_FAULT(x)					\
184	(((1 << ((x) & FAULT_TYPE_MASK)) &			\
185	  ((1 << FAULT_PERM_P) | (1 << FAULT_PERM_S))) != 0)
186
187static __inline void
188call_trapsignal(struct thread *td, int sig, u_long code)
189{
190	ksiginfo_t ksi;
191
192	ksiginfo_init_trap(&ksi);
193	ksi.ksi_signo = sig;
194	ksi.ksi_code = (int)code;
195	trapsignal(td, &ksi);
196}
197
198static __inline int
199data_abort_fixup(trapframe_t *tf, u_int fsr, u_int far, struct thread *td, struct ksig *ksig)
200{
201#ifdef CPU_ABORT_FIXUP_REQUIRED
202	int error;
203
204	/* Call the cpu specific data abort fixup routine */
205	error = cpu_dataabt_fixup(tf);
206	if (__predict_true(error != ABORT_FIXUP_FAILED))
207		return (error);
208
209	/*
210	 * Oops, couldn't fix up the instruction
211	 */
212	printf("data_abort_fixup: fixup for %s mode data abort failed.\n",
213	    TRAP_USERMODE(tf) ? "user" : "kernel");
214	printf("pc = 0x%08x, opcode 0x%08x, insn = ", tf->tf_pc,
215	    *((u_int *)tf->tf_pc));
216	disassemble(tf->tf_pc);
217
218	/* Die now if this happened in kernel mode */
219	if (!TRAP_USERMODE(tf))
220		dab_fatal(tf, fsr, far, td, NULL, ksig);
221
222	return (error);
223#else
224	return (ABORT_FIXUP_OK);
225#endif /* CPU_ABORT_FIXUP_REQUIRED */
226}
227
228void
229data_abort_handler(trapframe_t *tf)
230{
231	struct vm_map *map;
232	struct pcb *pcb;
233	struct thread *td;
234	u_int user, far, fsr;
235	vm_prot_t ftype;
236	void *onfault;
237	vm_offset_t va;
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: fault address=%p (from pc=%p lr=%p)\n",
248	       (void*)far, (void*)tf->tf_pc, (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_INC(cnt.v_trap);
260	/* Data abort came from user mode? */
261	user = TRAP_USERMODE(tf);
262
263	if (user) {
264		td->td_pticks = 0;
265		td->td_frame = tf;
266		if (td->td_ucred != td->td_proc->p_ucred)
267			cred_update_thread(td);
268
269	}
270	/* Grab the current pcb */
271	pcb = td->td_pcb;
272	/* Re-enable interrupts if they were enabled previously */
273	if (td->td_md.md_spinlock_count == 0) {
274		if (__predict_true(tf->tf_spsr & I32_bit) == 0)
275			enable_interrupts(I32_bit);
276		if (__predict_true(tf->tf_spsr & F32_bit) == 0)
277			enable_interrupts(F32_bit);
278	}
279
280
281	/* Invoke the appropriate handler, if necessary */
282	if (__predict_false(data_aborts[fsr & FAULT_TYPE_MASK].func != NULL)) {
283		if ((data_aborts[fsr & FAULT_TYPE_MASK].func)(tf, fsr, far,
284		    td, &ksig)) {
285			goto do_trapsignal;
286		}
287		goto out;
288	}
289
290	/*
291	 * At this point, we're dealing with one of the following data aborts:
292	 *
293	 *  FAULT_TRANS_S  - Translation -- Section
294	 *  FAULT_TRANS_P  - Translation -- Page
295	 *  FAULT_DOMAIN_S - Domain -- Section
296	 *  FAULT_DOMAIN_P - Domain -- Page
297	 *  FAULT_PERM_S   - Permission -- Section
298	 *  FAULT_PERM_P   - Permission -- Page
299	 *
300	 * These are the main virtual memory-related faults signalled by
301	 * the MMU.
302	 */
303
304	/* fusubailout is used by [fs]uswintr to avoid page faulting */
305	if (__predict_false(pcb->pcb_onfault == fusubailout)) {
306		tf->tf_r0 = EFAULT;
307		tf->tf_pc = (register_t)(intptr_t) pcb->pcb_onfault;
308		return;
309	}
310
311	/*
312	 * Make sure the Program Counter is sane. We could fall foul of
313	 * someone executing Thumb code, in which case the PC might not
314	 * be word-aligned. This would cause a kernel alignment fault
315	 * further down if we have to decode the current instruction.
316	 * XXX: It would be nice to be able to support Thumb at some point.
317	 */
318	if (__predict_false((tf->tf_pc & 3) != 0)) {
319		if (user) {
320			/*
321			 * Give the user an illegal instruction signal.
322			 */
323			/* Deliver a SIGILL to the process */
324			ksig.signb = SIGILL;
325			ksig.code = 0;
326			goto do_trapsignal;
327		}
328
329		/*
330		 * The kernel never executes Thumb code.
331		 */
332		printf("\ndata_abort_fault: Misaligned Kernel-mode "
333		    "Program Counter\n");
334		dab_fatal(tf, fsr, far, td, &ksig);
335	}
336
337	/* See if the cpu state needs to be fixed up */
338	switch (data_abort_fixup(tf, fsr, far, td, &ksig)) {
339	case ABORT_FIXUP_RETURN:
340		return;
341	case ABORT_FIXUP_FAILED:
342		/* Deliver a SIGILL to the process */
343		ksig.signb = SIGILL;
344		ksig.code = 0;
345		goto do_trapsignal;
346	default:
347		break;
348	}
349
350	va = trunc_page((vm_offset_t)far);
351
352	/*
353	 * It is only a kernel address space fault iff:
354	 *	1. user == 0  and
355	 *	2. pcb_onfault not set or
356	 *	3. pcb_onfault set and not LDRT/LDRBT/STRT/STRBT instruction.
357	 */
358	if (user == 0 && (va >= VM_MIN_KERNEL_ADDRESS ||
359	    (va < VM_MIN_ADDRESS && vector_page == ARM_VECTORS_LOW)) &&
360	    __predict_true((pcb->pcb_onfault == NULL ||
361	     (ReadWord(tf->tf_pc) & 0x05200000) != 0x04200000))) {
362		map = kernel_map;
363
364		/* Was the fault due to the FPE/IPKDB ? */
365		if (__predict_false((tf->tf_spsr & PSR_MODE)==PSR_UND32_MODE)) {
366
367			/*
368			 * Force exit via userret()
369			 * This is necessary as the FPE is an extension to
370			 * userland that actually runs in a priveledged mode
371			 * but uses USR mode permissions for its accesses.
372			 */
373			user = 1;
374			ksig.signb = SIGSEGV;
375			ksig.code = 0;
376			goto do_trapsignal;
377		}
378	} else {
379		map = &td->td_proc->p_vmspace->vm_map;
380	}
381
382	/*
383	 * We need to know whether the page should be mapped
384	 * as R or R/W. The MMU does not give us the info as
385	 * to whether the fault was caused by a read or a write.
386	 *
387	 * However, we know that a permission fault can only be
388	 * the result of a write to a read-only location, so
389	 * we can deal with those quickly.
390	 *
391	 * Otherwise we need to disassemble the instruction
392	 * responsible to determine if it was a write.
393	 */
394	if (IS_PERMISSION_FAULT(fsr))
395		ftype = VM_PROT_WRITE;
396	else {
397		u_int insn = ReadWord(tf->tf_pc);
398
399		if (((insn & 0x0c100000) == 0x04000000) ||	/* STR/STRB */
400		    ((insn & 0x0e1000b0) == 0x000000b0) ||	/* STRH/STRD */
401		    ((insn & 0x0a100000) == 0x08000000)) {	/* STM/CDT */
402			ftype = VM_PROT_WRITE;
403		} else {
404			if ((insn & 0x0fb00ff0) == 0x01000090)	/* SWP */
405				ftype = VM_PROT_READ | VM_PROT_WRITE;
406			else
407				ftype = VM_PROT_READ;
408		}
409	}
410
411	/*
412	 * See if the fault is as a result of ref/mod emulation,
413	 * or domain mismatch.
414	 */
415#ifdef DEBUG
416	last_fault_code = fsr;
417#endif
418	if (pmap_fault_fixup(vmspace_pmap(td->td_proc->p_vmspace), va, ftype,
419	    user)) {
420		goto out;
421	}
422
423	onfault = pcb->pcb_onfault;
424	pcb->pcb_onfault = NULL;
425	if (map != kernel_map) {
426		PROC_LOCK(p);
427		p->p_lock++;
428		PROC_UNLOCK(p);
429	}
430	error = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
431	pcb->pcb_onfault = onfault;
432
433	if (map != kernel_map) {
434		PROC_LOCK(p);
435		p->p_lock--;
436		PROC_UNLOCK(p);
437	}
438	if (__predict_true(error == 0))
439		goto out;
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_name,
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);
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	disable_interrupts(I32_bit|F32_bit);
492	if (td != NULL) {
493		printf("Fatal %s mode data abort: '%s'\n", mode,
494		    data_aborts[fsr & FAULT_TYPE_MASK].desc);
495		printf("trapframe: %p\nFSR=%08x, FAR=", tf, fsr);
496		if ((fsr & FAULT_IMPRECISE) == 0)
497			printf("%08x, ", far);
498		else
499			printf("Invalid,  ");
500		printf("spsr=%08x\n", tf->tf_spsr);
501	} else {
502		printf("Fatal %s mode prefetch abort at 0x%08x\n",
503		    mode, tf->tf_pc);
504		printf("trapframe: %p, spsr=%08x\n", tf, tf->tf_spsr);
505	}
506
507	printf("r0 =%08x, r1 =%08x, r2 =%08x, r3 =%08x\n",
508	    tf->tf_r0, tf->tf_r1, tf->tf_r2, tf->tf_r3);
509	printf("r4 =%08x, r5 =%08x, r6 =%08x, r7 =%08x\n",
510	    tf->tf_r4, tf->tf_r5, tf->tf_r6, tf->tf_r7);
511	printf("r8 =%08x, r9 =%08x, r10=%08x, r11=%08x\n",
512	    tf->tf_r8, tf->tf_r9, tf->tf_r10, tf->tf_r11);
513	printf("r12=%08x, ", tf->tf_r12);
514
515	if (TRAP_USERMODE(tf))
516		printf("usp=%08x, ulr=%08x",
517		    tf->tf_usr_sp, tf->tf_usr_lr);
518	else
519		printf("ssp=%08x, slr=%08x",
520		    tf->tf_svc_sp, tf->tf_svc_lr);
521	printf(", pc =%08x\n\n", tf->tf_pc);
522
523#ifdef KDB
524	if (debugger_on_panic || kdb_active)
525		kdb_trap(fsr, 0, tf);
526#endif
527	panic("Fatal abort");
528	/*NOTREACHED*/
529}
530
531/*
532 * dab_align() handles the following data aborts:
533 *
534 *  FAULT_ALIGN_0 - Alignment fault
535 *  FAULT_ALIGN_1 - Alignment fault
536 *
537 * These faults are fatal if they happen in kernel mode. Otherwise, we
538 * deliver a bus error to the process.
539 */
540static int
541dab_align(trapframe_t *tf, u_int fsr, u_int far, struct thread *td, struct ksig *ksig)
542{
543
544	/* Alignment faults are always fatal if they occur in kernel mode */
545	if (!TRAP_USERMODE(tf)) {
546		if (!td || !td->td_pcb->pcb_onfault)
547			dab_fatal(tf, fsr, far, td, ksig);
548		tf->tf_r0 = EFAULT;
549		tf->tf_pc = (int)td->td_pcb->pcb_onfault;
550		return (0);
551	}
552
553	/* pcb_onfault *must* be NULL at this point */
554
555	/* See if the cpu state needs to be fixed up */
556	(void) data_abort_fixup(tf, fsr, far, td, ksig);
557
558	/* Deliver a bus error signal to the process */
559	ksig->code = 0;
560	ksig->signb = SIGBUS;
561	td->td_frame = tf;
562
563	return (1);
564}
565
566/*
567 * dab_buserr() handles the following data aborts:
568 *
569 *  FAULT_BUSERR_0 - External Abort on Linefetch -- Section
570 *  FAULT_BUSERR_1 - External Abort on Linefetch -- Page
571 *  FAULT_BUSERR_2 - External Abort on Non-linefetch -- Section
572 *  FAULT_BUSERR_3 - External Abort on Non-linefetch -- Page
573 *  FAULT_BUSTRNL1 - External abort on Translation -- Level 1
574 *  FAULT_BUSTRNL2 - External abort on Translation -- Level 2
575 *
576 * If pcb_onfault is set, flag the fault and return to the handler.
577 * If the fault occurred in user mode, give the process a SIGBUS.
578 *
579 * Note: On XScale, FAULT_BUSERR_0, FAULT_BUSERR_1, and FAULT_BUSERR_2
580 * can be flagged as imprecise in the FSR. This causes a real headache
581 * since some of the machine state is lost. In this case, tf->tf_pc
582 * may not actually point to the offending instruction. In fact, if
583 * we've taken a double abort fault, it generally points somewhere near
584 * the top of "data_abort_entry" in exception.S.
585 *
586 * In all other cases, these data aborts are considered fatal.
587 */
588static int
589dab_buserr(trapframe_t *tf, u_int fsr, u_int far, struct thread *td, struct ksig *ksig)
590{
591	struct pcb *pcb = td->td_pcb;
592
593#ifdef __XSCALE__
594	if ((fsr & FAULT_IMPRECISE) != 0 &&
595	    (tf->tf_spsr & PSR_MODE) == PSR_ABT32_MODE) {
596		/*
597		 * Oops, an imprecise, double abort fault. We've lost the
598		 * r14_abt/spsr_abt values corresponding to the original
599		 * abort, and the spsr saved in the trapframe indicates
600		 * ABT mode.
601		 */
602		tf->tf_spsr &= ~PSR_MODE;
603
604		/*
605		 * We use a simple heuristic to determine if the double abort
606		 * happened as a result of a kernel or user mode access.
607		 * If the current trapframe is at the top of the kernel stack,
608		 * the fault _must_ have come from user mode.
609		 */
610		if (tf != ((trapframe_t *)pcb->un_32.pcb32_sp) - 1) {
611			/*
612			 * Kernel mode. We're either about to die a
613			 * spectacular death, or pcb_onfault will come
614			 * to our rescue. Either way, the current value
615			 * of tf->tf_pc is irrelevant.
616			 */
617			tf->tf_spsr |= PSR_SVC32_MODE;
618			if (pcb->pcb_onfault == NULL)
619				printf("\nKernel mode double abort!\n");
620		} else {
621			/*
622			 * User mode. We've lost the program counter at the
623			 * time of the fault (not that it was accurate anyway;
624			 * it's not called an imprecise fault for nothing).
625			 * About all we can do is copy r14_usr to tf_pc and
626			 * hope for the best. The process is about to get a
627			 * SIGBUS, so it's probably history anyway.
628			 */
629			tf->tf_spsr |= PSR_USR32_MODE;
630			tf->tf_pc = tf->tf_usr_lr;
631		}
632	}
633
634	/* FAR is invalid for imprecise exceptions */
635	if ((fsr & FAULT_IMPRECISE) != 0)
636		far = 0;
637#endif /* __XSCALE__ */
638
639	if (pcb->pcb_onfault) {
640		tf->tf_r0 = EFAULT;
641		tf->tf_pc = (register_t)(intptr_t) pcb->pcb_onfault;
642		return (0);
643	}
644
645	/* See if the cpu state needs to be fixed up */
646	(void) data_abort_fixup(tf, fsr, far, td, ksig);
647
648	/*
649	 * At this point, if the fault happened in kernel mode, we're toast
650	 */
651	if (!TRAP_USERMODE(tf))
652		dab_fatal(tf, fsr, far, td, ksig);
653
654	/* Deliver a bus error signal to the process */
655	ksig->signb = SIGBUS;
656	ksig->code = 0;
657	td->td_frame = tf;
658
659	return (1);
660}
661
662static __inline int
663prefetch_abort_fixup(trapframe_t *tf, struct ksig *ksig)
664{
665#ifdef CPU_ABORT_FIXUP_REQUIRED
666	int error;
667
668	/* Call the cpu specific prefetch abort fixup routine */
669	error = cpu_prefetchabt_fixup(tf);
670	if (__predict_true(error != ABORT_FIXUP_FAILED))
671		return (error);
672
673	/*
674	 * Oops, couldn't fix up the instruction
675	 */
676	printf(
677	    "prefetch_abort_fixup: fixup for %s mode prefetch abort failed.\n",
678	    TRAP_USERMODE(tf) ? "user" : "kernel");
679	printf("pc = 0x%08x, opcode 0x%08x, insn = ", tf->tf_pc,
680	    *((u_int *)tf->tf_pc));
681	disassemble(tf->tf_pc);
682
683	/* Die now if this happened in kernel mode */
684	if (!TRAP_USERMODE(tf))
685		dab_fatal(tf, 0, tf->tf_pc, NULL, ksig);
686
687	return (error);
688#else
689	return (ABORT_FIXUP_OK);
690#endif /* CPU_ABORT_FIXUP_REQUIRED */
691}
692
693/*
694 * void prefetch_abort_handler(trapframe_t *tf)
695 *
696 * Abort handler called when instruction execution occurs at
697 * a non existent or restricted (access permissions) memory page.
698 * If the address is invalid and we were in SVC mode then panic as
699 * the kernel should never prefetch abort.
700 * If the address is invalid and the page is mapped then the user process
701 * does no have read permission so send it a signal.
702 * Otherwise fault the page in and try again.
703 */
704void
705prefetch_abort_handler(trapframe_t *tf)
706{
707	struct thread *td;
708	struct proc * p;
709	struct vm_map *map;
710	vm_offset_t fault_pc, va;
711	int error = 0;
712	struct ksig ksig;
713
714
715#if 0
716	/* Update vmmeter statistics */
717	uvmexp.traps++;
718#endif
719#if 0
720	printf("prefetch abort handler: %p %p\n", (void*)tf->tf_pc,
721	    (void*)tf->tf_usr_lr);
722#endif
723
724 	td = curthread;
725	p = td->td_proc;
726	PCPU_INC(cnt.v_trap);
727
728	if (TRAP_USERMODE(tf)) {
729		td->td_frame = tf;
730		if (td->td_ucred != td->td_proc->p_ucred)
731			cred_update_thread(td);
732	}
733	fault_pc = tf->tf_pc;
734	if (td->td_md.md_spinlock_count == 0) {
735		if (__predict_true(tf->tf_spsr & I32_bit) == 0)
736			enable_interrupts(I32_bit);
737		if (__predict_true(tf->tf_spsr & F32_bit) == 0)
738			enable_interrupts(F32_bit);
739	}
740
741	/* See if the cpu state needs to be fixed up */
742	switch (prefetch_abort_fixup(tf, &ksig)) {
743	case ABORT_FIXUP_RETURN:
744		return;
745	case ABORT_FIXUP_FAILED:
746		/* Deliver a SIGILL to the process */
747		ksig.signb = SIGILL;
748		ksig.code = 0;
749		td->td_frame = tf;
750		goto do_trapsignal;
751	default:
752		break;
753	}
754
755	/* Prefetch aborts cannot happen in kernel mode */
756	if (__predict_false(!TRAP_USERMODE(tf)))
757		dab_fatal(tf, 0, tf->tf_pc, NULL, &ksig);
758	td->td_pticks = 0;
759
760
761	/* Ok validate the address, can only execute in USER space */
762	if (__predict_false(fault_pc >= VM_MAXUSER_ADDRESS ||
763	    (fault_pc < VM_MIN_ADDRESS && vector_page == ARM_VECTORS_LOW))) {
764		ksig.signb = SIGSEGV;
765		ksig.code = 0;
766		goto do_trapsignal;
767	}
768
769	map = &td->td_proc->p_vmspace->vm_map;
770	va = trunc_page(fault_pc);
771
772	/*
773	 * See if the pmap can handle this fault on its own...
774	 */
775#ifdef DEBUG
776	last_fault_code = -1;
777#endif
778	if (pmap_fault_fixup(map->pmap, va, VM_PROT_READ, 1))
779		goto out;
780
781	if (map != kernel_map) {
782		PROC_LOCK(p);
783		p->p_lock++;
784		PROC_UNLOCK(p);
785	}
786
787	error = vm_fault(map, va, VM_PROT_READ | VM_PROT_EXECUTE,
788	    VM_FAULT_NORMAL);
789	if (map != kernel_map) {
790		PROC_LOCK(p);
791		p->p_lock--;
792		PROC_UNLOCK(p);
793	}
794
795	if (__predict_true(error == 0))
796		goto out;
797
798	if (error == ENOMEM) {
799		printf("VM: pid %d (%s), uid %d killed: "
800		    "out of swap\n", td->td_proc->p_pid, td->td_name,
801		    (td->td_proc->p_ucred) ?
802		     td->td_proc->p_ucred->cr_uid : -1);
803		ksig.signb = SIGKILL;
804	} else {
805		ksig.signb = SIGSEGV;
806	}
807	ksig.code = 0;
808
809do_trapsignal:
810	call_trapsignal(td, ksig.signb, ksig.code);
811
812out:
813	userret(td, tf);
814
815}
816
817extern int badaddr_read_1(const uint8_t *, uint8_t *);
818extern int badaddr_read_2(const uint16_t *, uint16_t *);
819extern int badaddr_read_4(const uint32_t *, uint32_t *);
820/*
821 * Tentatively read an 8, 16, or 32-bit value from 'addr'.
822 * If the read succeeds, the value is written to 'rptr' and zero is returned.
823 * Else, return EFAULT.
824 */
825int
826badaddr_read(void *addr, size_t size, void *rptr)
827{
828	union {
829		uint8_t v1;
830		uint16_t v2;
831		uint32_t v4;
832	} u;
833	int rv;
834
835	cpu_drain_writebuf();
836
837	/* Read from the test address. */
838	switch (size) {
839	case sizeof(uint8_t):
840		rv = badaddr_read_1(addr, &u.v1);
841		if (rv == 0 && rptr)
842			*(uint8_t *) rptr = u.v1;
843		break;
844
845	case sizeof(uint16_t):
846		rv = badaddr_read_2(addr, &u.v2);
847		if (rv == 0 && rptr)
848			*(uint16_t *) rptr = u.v2;
849		break;
850
851	case sizeof(uint32_t):
852		rv = badaddr_read_4(addr, &u.v4);
853		if (rv == 0 && rptr)
854			*(uint32_t *) rptr = u.v4;
855		break;
856
857	default:
858		panic("badaddr: invalid size (%lu)", (u_long) size);
859	}
860
861	/* Return EFAULT if the address was invalid, else zero */
862	return (rv);
863}
864
865int
866cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
867{
868	struct proc *p;
869	register_t *ap;
870	int error;
871
872#ifdef __ARM_EABI__
873	sa->code = td->td_frame->tf_r7;
874#else
875	sa->code = sa->insn & 0x000fffff;
876#endif
877	ap = &td->td_frame->tf_r0;
878	if (sa->code == SYS_syscall) {
879		sa->code = *ap++;
880		sa->nap--;
881	} else if (sa->code == SYS___syscall) {
882		sa->code = ap[_QUAD_LOWWORD];
883		sa->nap -= 2;
884		ap += 2;
885	}
886	p = td->td_proc;
887	if (p->p_sysent->sv_mask)
888		sa->code &= p->p_sysent->sv_mask;
889	if (sa->code >= p->p_sysent->sv_size)
890		sa->callp = &p->p_sysent->sv_table[0];
891	else
892		sa->callp = &p->p_sysent->sv_table[sa->code];
893	sa->narg = sa->callp->sy_narg;
894	error = 0;
895	memcpy(sa->args, ap, sa->nap * sizeof(register_t));
896	if (sa->narg > sa->nap) {
897		error = copyin((void *)td->td_frame->tf_usr_sp, sa->args +
898		    sa->nap, (sa->narg - sa->nap) * sizeof(register_t));
899	}
900	if (error == 0) {
901		td->td_retval[0] = 0;
902		td->td_retval[1] = 0;
903	}
904	return (error);
905}
906
907#include "../../kern/subr_syscall.c"
908
909static void
910syscall(struct thread *td, trapframe_t *frame)
911{
912	struct syscall_args sa;
913	int error;
914
915#ifndef __ARM_EABI__
916	sa.insn = *(uint32_t *)(frame->tf_pc - INSN_SIZE);
917	switch (sa.insn & SWI_OS_MASK) {
918	case 0: /* XXX: we need our own one. */
919		break;
920	default:
921		call_trapsignal(td, SIGILL, 0);
922		userret(td, frame);
923		return;
924	}
925#endif
926	sa.nap = 4;
927
928	error = syscallenter(td, &sa);
929	KASSERT(error != 0 || td->td_ar == NULL,
930	    ("returning from syscall with td_ar set!"));
931	syscallret(td, error, &sa);
932}
933
934void
935swi_handler(trapframe_t *frame)
936{
937	struct thread *td = curthread;
938
939	td->td_frame = frame;
940
941	td->td_pticks = 0;
942	/*
943      	 * Make sure the program counter is correctly aligned so we
944	 * don't take an alignment fault trying to read the opcode.
945	 */
946	if (__predict_false(((frame->tf_pc - INSN_SIZE) & 3) != 0)) {
947		call_trapsignal(td, SIGILL, 0);
948		userret(td, frame);
949		return;
950	}
951	/*
952	 * Enable interrupts if they were enabled before the exception.
953	 * Since all syscalls *should* come from user mode it will always
954	 * be safe to enable them, but check anyway.
955	 */
956	if (td->td_md.md_spinlock_count == 0) {
957		if (__predict_true(frame->tf_spsr & I32_bit) == 0)
958			enable_interrupts(I32_bit);
959		if (__predict_true(frame->tf_spsr & F32_bit) == 0)
960			enable_interrupts(F32_bit);
961	}
962
963	syscall(td, frame);
964}
965
966