syscall.c revision 157616
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 157616 2006-04-09 20:16:47Z 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	int error = 0;
238	struct ksig ksig;
239	struct proc *p;
240
241
242	/* Grab FAR/FSR before enabling interrupts */
243	far = cpu_faultaddress();
244	fsr = cpu_faultstatus();
245#if 0
246	printf("data abort: %p (from %p %p)\n", (void*)far, (void*)tf->tf_pc,
247	    (void*)tf->tf_svc_lr);
248#endif
249
250	/* Update vmmeter statistics */
251#if 0
252	vmexp.traps++;
253#endif
254
255	td = curthread;
256	p = td->td_proc;
257
258	PCPU_LAZY_INC(cnt.v_trap);
259	/* Data abort came from user mode? */
260	user = TRAP_USERMODE(tf);
261
262	if (user) {
263		td->td_pticks = 0;
264		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		if (__predict_true(tf->tf_spsr & I32_bit) == 0)
276			enable_interrupts(I32_bit);
277		if (__predict_true(tf->tf_spsr & F32_bit) == 0)
278			enable_interrupts(F32_bit);
279	}
280
281
282	/* Invoke the appropriate handler, if necessary */
283	if (__predict_false(data_aborts[fsr & FAULT_TYPE_MASK].func != NULL)) {
284		if ((data_aborts[fsr & FAULT_TYPE_MASK].func)(tf, fsr, far,
285		    td, &ksig)) {
286			goto do_trapsignal;
287		}
288		goto out;
289	}
290
291	/*
292	 * At this point, we're dealing with one of the following data aborts:
293	 *
294	 *  FAULT_TRANS_S  - Translation -- Section
295	 *  FAULT_TRANS_P  - Translation -- Page
296	 *  FAULT_DOMAIN_S - Domain -- Section
297	 *  FAULT_DOMAIN_P - Domain -- Page
298	 *  FAULT_PERM_S   - Permission -- Section
299	 *  FAULT_PERM_P   - Permission -- Page
300	 *
301	 * These are the main virtual memory-related faults signalled by
302	 * the MMU.
303	 */
304
305	/* fusubailout is used by [fs]uswintr to avoid page faulting */
306	if (__predict_false(pcb->pcb_onfault == fusubailout)) {
307		tf->tf_r0 = EFAULT;
308		tf->tf_pc = (register_t)(intptr_t) pcb->pcb_onfault;
309		return;
310	}
311
312	/*
313	 * Make sure the Program Counter is sane. We could fall foul of
314	 * someone executing Thumb code, in which case the PC might not
315	 * be word-aligned. This would cause a kernel alignment fault
316	 * further down if we have to decode the current instruction.
317	 * XXX: It would be nice to be able to support Thumb at some point.
318	 */
319	if (__predict_false((tf->tf_pc & 3) != 0)) {
320		if (user) {
321			/*
322			 * Give the user an illegal instruction signal.
323			 */
324			/* Deliver a SIGILL to the process */
325			ksig.signb = SIGILL;
326			ksig.code = 0;
327			goto do_trapsignal;
328		}
329
330		/*
331		 * The kernel never executes Thumb code.
332		 */
333		printf("\ndata_abort_fault: Misaligned Kernel-mode "
334		    "Program Counter\n");
335		dab_fatal(tf, fsr, far, td, &ksig);
336	}
337
338	/* See if the cpu state needs to be fixed up */
339	switch (data_abort_fixup(tf, fsr, far, td, &ksig)) {
340	case ABORT_FIXUP_RETURN:
341		return;
342	case ABORT_FIXUP_FAILED:
343		/* Deliver a SIGILL to the process */
344		ksig.signb = SIGILL;
345		ksig.code = 0;
346		goto do_trapsignal;
347	default:
348		break;
349	}
350
351	va = trunc_page((vm_offset_t)far);
352
353	/*
354	 * It is only a kernel address space fault iff:
355	 *	1. user == 0  and
356	 *	2. pcb_onfault not set or
357	 *	3. pcb_onfault set and not LDRT/LDRBT/STRT/STRBT instruction.
358	 */
359	if (user == 0 && (va >= VM_MIN_KERNEL_ADDRESS ||
360	    (va < VM_MIN_ADDRESS && vector_page == ARM_VECTORS_LOW)) &&
361	    __predict_true((pcb->pcb_onfault == NULL ||
362	     (ReadWord(tf->tf_pc) & 0x05200000) != 0x04200000))) {
363		map = kernel_map;
364
365		/* Was the fault due to the FPE/IPKDB ? */
366		if (__predict_false((tf->tf_spsr & PSR_MODE)==PSR_UND32_MODE)) {
367
368			/*
369			 * Force exit via userret()
370			 * This is necessary as the FPE is an extension to
371			 * userland that actually runs in a priveledged mode
372			 * but uses USR mode permissions for its accesses.
373			 */
374			user = 1;
375			ksig.signb = SIGSEGV;
376			ksig.code = 0;
377			goto do_trapsignal;
378		}
379	} else {
380		map = &td->td_proc->p_vmspace->vm_map;
381	}
382
383	/*
384	 * We need to know whether the page should be mapped
385	 * as R or R/W. The MMU does not give us the info as
386	 * to whether the fault was caused by a read or a write.
387	 *
388	 * However, we know that a permission fault can only be
389	 * the result of a write to a read-only location, so
390	 * we can deal with those quickly.
391	 *
392	 * Otherwise we need to disassemble the instruction
393	 * responsible to determine if it was a write.
394	 */
395	if (IS_PERMISSION_FAULT(fsr)) {
396		ftype = VM_PROT_WRITE;
397	} else {
398		u_int insn = ReadWord(tf->tf_pc);
399
400		if (((insn & 0x0c100000) == 0x04000000) ||	/* STR/STRB */
401		    ((insn & 0x0e1000b0) == 0x000000b0) ||	/* STRH/STRD */
402		    ((insn & 0x0a100000) == 0x08000000))	/* STM/CDT */
403		{
404			ftype = VM_PROT_WRITE;
405	}
406		else
407		if ((insn & 0x0fb00ff0) == 0x01000090)		/* SWP */
408			ftype = VM_PROT_READ | VM_PROT_WRITE;
409		else
410			ftype = VM_PROT_READ;
411	}
412
413	/*
414	 * See if the fault is as a result of ref/mod emulation,
415	 * or domain mismatch.
416	 */
417#ifdef DEBUG
418	last_fault_code = fsr;
419#endif
420	if (pmap_fault_fixup(vmspace_pmap(td->td_proc->p_vmspace), va, ftype,
421	    user)) {
422		goto out;
423	}
424
425	onfault = pcb->pcb_onfault;
426	pcb->pcb_onfault = NULL;
427	if (map != kernel_map) {
428		PROC_LOCK(p);
429		p->p_lock++;
430		PROC_UNLOCK(p);
431	}
432	error = vm_fault(map, va, ftype, (ftype & VM_PROT_WRITE) ?
433	    VM_FAULT_DIRTY : VM_FAULT_NORMAL);
434	pcb->pcb_onfault = onfault;
435
436	if (map != kernel_map) {
437		PROC_LOCK(p);
438		p->p_lock--;
439		PROC_UNLOCK(p);
440	}
441	if (__predict_true(error == 0))
442		goto out;
443	if (user == 0) {
444		if (pcb->pcb_onfault) {
445			tf->tf_r0 = error;
446			tf->tf_pc = (register_t)(intptr_t) pcb->pcb_onfault;
447			return;
448		}
449
450		printf("\nvm_fault(%p, %x, %x, 0) -> %x\n", map, va, ftype,
451		    error);
452		dab_fatal(tf, fsr, far, td, &ksig);
453	}
454
455
456	if (error == ENOMEM) {
457		printf("VM: pid %d (%s), uid %d killed: "
458		    "out of swap\n", td->td_proc->p_pid, td->td_proc->p_comm,
459		    (td->td_proc->p_ucred) ?
460		     td->td_proc->p_ucred->cr_uid : -1);
461		ksig.signb = SIGKILL;
462	} else {
463		ksig.signb = SIGSEGV;
464	}
465	ksig.code = 0;
466do_trapsignal:
467	call_trapsignal(td, ksig.signb, ksig.code);
468out:
469	/* If returning to user mode, make sure to invoke userret() */
470	if (user)
471		userret(td, tf);
472}
473
474/*
475 * dab_fatal() handles the following data aborts:
476 *
477 *  FAULT_WRTBUF_0 - Vector Exception
478 *  FAULT_WRTBUF_1 - Terminal Exception
479 *
480 * We should never see these on a properly functioning system.
481 *
482 * This function is also called by the other handlers if they
483 * detect a fatal problem.
484 *
485 * Note: If 'l' is NULL, we assume we're dealing with a prefetch abort.
486 */
487static int
488dab_fatal(trapframe_t *tf, u_int fsr, u_int far, struct thread *td, struct ksig *ksig)
489{
490	const char *mode;
491
492	mode = TRAP_USERMODE(tf) ? "user" : "kernel";
493
494	disable_interrupts(I32_bit|F32_bit);
495	if (td != NULL) {
496		printf("Fatal %s mode data abort: '%s'\n", mode,
497		    data_aborts[fsr & FAULT_TYPE_MASK].desc);
498		printf("trapframe: %p\nFSR=%08x, FAR=", tf, fsr);
499		if ((fsr & FAULT_IMPRECISE) == 0)
500			printf("%08x, ", far);
501		else
502			printf("Invalid,  ");
503		printf("spsr=%08x\n", tf->tf_spsr);
504	} else {
505		printf("Fatal %s mode prefetch abort at 0x%08x\n",
506		    mode, tf->tf_pc);
507		printf("trapframe: %p, spsr=%08x\n", tf, tf->tf_spsr);
508	}
509
510	printf("r0 =%08x, r1 =%08x, r2 =%08x, r3 =%08x\n",
511	    tf->tf_r0, tf->tf_r1, tf->tf_r2, tf->tf_r3);
512	printf("r4 =%08x, r5 =%08x, r6 =%08x, r7 =%08x\n",
513	    tf->tf_r4, tf->tf_r5, tf->tf_r6, tf->tf_r7);
514	printf("r8 =%08x, r9 =%08x, r10=%08x, r11=%08x\n",
515	    tf->tf_r8, tf->tf_r9, tf->tf_r10, tf->tf_r11);
516	printf("r12=%08x, ", tf->tf_r12);
517
518	if (TRAP_USERMODE(tf))
519		printf("usp=%08x, ulr=%08x",
520		    tf->tf_usr_sp, tf->tf_usr_lr);
521	else
522		printf("ssp=%08x, slr=%08x",
523		    tf->tf_svc_sp, tf->tf_svc_lr);
524	printf(", pc =%08x\n\n", tf->tf_pc);
525
526#ifdef KDB
527	kdb_trap(fsr, 0, tf);
528#endif
529	panic("Fatal abort");
530	/*NOTREACHED*/
531}
532
533/*
534 * dab_align() handles the following data aborts:
535 *
536 *  FAULT_ALIGN_0 - Alignment fault
537 *  FAULT_ALIGN_0 - Alignment fault
538 *
539 * These faults are fatal if they happen in kernel mode. Otherwise, we
540 * deliver a bus error to the process.
541 */
542static int
543dab_align(trapframe_t *tf, u_int fsr, u_int far, struct thread *td, struct ksig *ksig)
544{
545
546	/* Alignment faults are always fatal if they occur in kernel mode */
547	if (!TRAP_USERMODE(tf)) {
548		if (!td || !td->td_pcb->pcb_onfault)
549			dab_fatal(tf, fsr, far, td, ksig);
550		tf->tf_r0 = EFAULT;
551		tf->tf_pc = (int)td->td_pcb->pcb_onfault;
552		return (0);
553	}
554
555	/* pcb_onfault *must* be NULL at this point */
556
557	/* See if the cpu state needs to be fixed up */
558	(void) data_abort_fixup(tf, fsr, far, td, ksig);
559
560	/* Deliver a bus error signal to the process */
561	ksig->code = 0;
562	ksig->signb = SIGBUS;
563	td->td_frame = tf;
564
565	return (1);
566}
567
568/*
569 * dab_buserr() handles the following data aborts:
570 *
571 *  FAULT_BUSERR_0 - External Abort on Linefetch -- Section
572 *  FAULT_BUSERR_1 - External Abort on Linefetch -- Page
573 *  FAULT_BUSERR_2 - External Abort on Non-linefetch -- Section
574 *  FAULT_BUSERR_3 - External Abort on Non-linefetch -- Page
575 *  FAULT_BUSTRNL1 - External abort on Translation -- Level 1
576 *  FAULT_BUSTRNL2 - External abort on Translation -- Level 2
577 *
578 * If pcb_onfault is set, flag the fault and return to the handler.
579 * If the fault occurred in user mode, give the process a SIGBUS.
580 *
581 * Note: On XScale, FAULT_BUSERR_0, FAULT_BUSERR_1, and FAULT_BUSERR_2
582 * can be flagged as imprecise in the FSR. This causes a real headache
583 * since some of the machine state is lost. In this case, tf->tf_pc
584 * may not actually point to the offending instruction. In fact, if
585 * we've taken a double abort fault, it generally points somewhere near
586 * the top of "data_abort_entry" in exception.S.
587 *
588 * In all other cases, these data aborts are considered fatal.
589 */
590static int
591dab_buserr(trapframe_t *tf, u_int fsr, u_int far, struct thread *td, struct ksig *ksig)
592{
593	struct pcb *pcb = td->td_pcb;
594
595#ifdef __XSCALE__
596	if ((fsr & FAULT_IMPRECISE) != 0 &&
597	    (tf->tf_spsr & PSR_MODE) == PSR_ABT32_MODE) {
598		/*
599		 * Oops, an imprecise, double abort fault. We've lost the
600		 * r14_abt/spsr_abt values corresponding to the original
601		 * abort, and the spsr saved in the trapframe indicates
602		 * ABT mode.
603		 */
604		tf->tf_spsr &= ~PSR_MODE;
605
606		/*
607		 * We use a simple heuristic to determine if the double abort
608		 * happened as a result of a kernel or user mode access.
609		 * If the current trapframe is at the top of the kernel stack,
610		 * the fault _must_ have come from user mode.
611		 */
612		if (tf != ((trapframe_t *)pcb->un_32.pcb32_sp) - 1) {
613			/*
614			 * Kernel mode. We're either about to die a
615			 * spectacular death, or pcb_onfault will come
616			 * to our rescue. Either way, the current value
617			 * of tf->tf_pc is irrelevant.
618			 */
619			tf->tf_spsr |= PSR_SVC32_MODE;
620			if (pcb->pcb_onfault == NULL)
621				printf("\nKernel mode double abort!\n");
622		} else {
623			/*
624			 * User mode. We've lost the program counter at the
625			 * time of the fault (not that it was accurate anyway;
626			 * it's not called an imprecise fault for nothing).
627			 * About all we can do is copy r14_usr to tf_pc and
628			 * hope for the best. The process is about to get a
629			 * SIGBUS, so it's probably history anyway.
630			 */
631			tf->tf_spsr |= PSR_USR32_MODE;
632			tf->tf_pc = tf->tf_usr_lr;
633		}
634	}
635
636	/* FAR is invalid for imprecise exceptions */
637	if ((fsr & FAULT_IMPRECISE) != 0)
638		far = 0;
639#endif /* __XSCALE__ */
640
641	if (pcb->pcb_onfault) {
642		tf->tf_r0 = EFAULT;
643		tf->tf_pc = (register_t)(intptr_t) pcb->pcb_onfault;
644		return (0);
645	}
646
647	/* See if the cpu state needs to be fixed up */
648	(void) data_abort_fixup(tf, fsr, far, td, ksig);
649
650	/*
651	 * At this point, if the fault happened in kernel mode, we're toast
652	 */
653	if (!TRAP_USERMODE(tf))
654		dab_fatal(tf, fsr, far, td, ksig);
655
656	/* Deliver a bus error signal to the process */
657	ksig->signb = SIGBUS;
658	ksig->code = 0;
659	td->td_frame = tf;
660
661	return (1);
662}
663
664static __inline int
665prefetch_abort_fixup(trapframe_t *tf, struct ksig *ksig)
666{
667#ifdef CPU_ABORT_FIXUP_REQUIRED
668	int error;
669
670	/* Call the cpu specific prefetch abort fixup routine */
671	error = cpu_prefetchabt_fixup(tf);
672	if (__predict_true(error != ABORT_FIXUP_FAILED))
673		return (error);
674
675	/*
676	 * Oops, couldn't fix up the instruction
677	 */
678	printf(
679	    "prefetch_abort_fixup: fixup for %s mode prefetch abort failed.\n",
680	    TRAP_USERMODE(tf) ? "user" : "kernel");
681	printf("pc = 0x%08x, opcode 0x%08x, insn = ", tf->tf_pc,
682	    *((u_int *)tf->tf_pc));
683	disassemble(tf->tf_pc);
684
685	/* Die now if this happened in kernel mode */
686	if (!TRAP_USERMODE(tf))
687		dab_fatal(tf, 0, tf->tf_pc, NULL, ksig);
688
689	return (error);
690#else
691	return (ABORT_FIXUP_OK);
692#endif /* CPU_ABORT_FIXUP_REQUIRED */
693}
694
695/*
696 * void prefetch_abort_handler(trapframe_t *tf)
697 *
698 * Abort handler called when instruction execution occurs at
699 * a non existent or restricted (access permissions) memory page.
700 * If the address is invalid and we were in SVC mode then panic as
701 * the kernel should never prefetch abort.
702 * If the address is invalid and the page is mapped then the user process
703 * does no have read permission so send it a signal.
704 * Otherwise fault the page in and try again.
705 */
706void
707prefetch_abort_handler(trapframe_t *tf)
708{
709	struct thread *td;
710	struct proc * p;
711	struct vm_map *map;
712	vm_offset_t fault_pc, va;
713	int error = 0;
714	struct ksig ksig;
715
716
717#if 0
718	/* Update vmmeter statistics */
719	uvmexp.traps++;
720#endif
721#if 0
722	printf("prefetch abort handler: %p %p\n", (void*)tf->tf_pc,
723	    (void*)tf->tf_usr_lr);
724#endif
725
726 	td = curthread;
727	p = td->td_proc;
728	PCPU_LAZY_INC(cnt.v_trap);
729
730	if (TRAP_USERMODE(tf)) {
731		td->td_frame = tf;
732		if (td->td_ucred != td->td_proc->p_ucred)
733			cred_update_thread(td);
734		if (td->td_proc->p_flag & P_SA)
735			thread_user_enter(td);
736	}
737	fault_pc = tf->tf_pc;
738	if (td->td_md.md_spinlock_count == 0) {
739		if (__predict_true(tf->tf_spsr & I32_bit) == 0)
740			enable_interrupts(I32_bit);
741		if (__predict_true(tf->tf_spsr & F32_bit) == 0)
742			enable_interrupts(F32_bit);
743	}
744
745
746
747	/* See if the cpu state needs to be fixed up */
748	switch (prefetch_abort_fixup(tf, &ksig)) {
749	case ABORT_FIXUP_RETURN:
750		return;
751	case ABORT_FIXUP_FAILED:
752		/* Deliver a SIGILL to the process */
753		ksig.signb = SIGILL;
754		ksig.code = 0;
755		td->td_frame = tf;
756		goto do_trapsignal;
757	default:
758		break;
759	}
760
761	/* Prefetch aborts cannot happen in kernel mode */
762	if (__predict_false(!TRAP_USERMODE(tf)))
763		dab_fatal(tf, 0, tf->tf_pc, NULL, &ksig);
764	td->td_pticks = 0;
765
766
767	/* Ok validate the address, can only execute in USER space */
768	if (__predict_false(fault_pc >= VM_MAXUSER_ADDRESS ||
769	    (fault_pc < VM_MIN_ADDRESS && vector_page == ARM_VECTORS_LOW))) {
770		ksig.signb = SIGSEGV;
771		ksig.code = 0;
772		goto do_trapsignal;
773	}
774
775	map = &td->td_proc->p_vmspace->vm_map;
776	va = trunc_page(fault_pc);
777
778	/*
779	 * See if the pmap can handle this fault on its own...
780	 */
781#ifdef DEBUG
782	last_fault_code = -1;
783#endif
784	if (pmap_fault_fixup(map->pmap, va, VM_PROT_READ, 1))
785		goto out;
786
787	if (map != kernel_map) {
788		PROC_LOCK(p);
789		p->p_lock++;
790		PROC_UNLOCK(p);
791	}
792
793	error = vm_fault(map, va, VM_PROT_READ | VM_PROT_EXECUTE,
794	    VM_FAULT_NORMAL);
795	if (map != kernel_map) {
796		PROC_LOCK(p);
797		p->p_lock--;
798		PROC_UNLOCK(p);
799	}
800
801	if (__predict_true(error == 0))
802		goto out;
803
804	if (error == ENOMEM) {
805		printf("VM: pid %d (%s), uid %d killed: "
806		    "out of swap\n", td->td_proc->p_pid, td->td_proc->p_comm,
807		    (td->td_proc->p_ucred) ?
808		     td->td_proc->p_ucred->cr_uid : -1);
809		ksig.signb = SIGKILL;
810	} else {
811		ksig.signb = SIGSEGV;
812	}
813	ksig.code = 0;
814
815do_trapsignal:
816	call_trapsignal(td, ksig.signb, ksig.code);
817
818out:
819	userret(td, tf);
820
821}
822
823extern int badaddr_read_1(const uint8_t *, uint8_t *);
824extern int badaddr_read_2(const uint16_t *, uint16_t *);
825extern int badaddr_read_4(const uint32_t *, uint32_t *);
826/*
827 * Tentatively read an 8, 16, or 32-bit value from 'addr'.
828 * If the read succeeds, the value is written to 'rptr' and zero is returned.
829 * Else, return EFAULT.
830 */
831int
832badaddr_read(void *addr, size_t size, void *rptr)
833{
834	union {
835		uint8_t v1;
836		uint16_t v2;
837		uint32_t v4;
838	} u;
839	int rv;
840
841	cpu_drain_writebuf();
842
843	/* Read from the test address. */
844	switch (size) {
845	case sizeof(uint8_t):
846		rv = badaddr_read_1(addr, &u.v1);
847		if (rv == 0 && rptr)
848			*(uint8_t *) rptr = u.v1;
849		break;
850
851	case sizeof(uint16_t):
852		rv = badaddr_read_2(addr, &u.v2);
853		if (rv == 0 && rptr)
854			*(uint16_t *) rptr = u.v2;
855		break;
856
857	case sizeof(uint32_t):
858		rv = badaddr_read_4(addr, &u.v4);
859		if (rv == 0 && rptr)
860			*(uint32_t *) rptr = u.v4;
861		break;
862
863	default:
864		panic("badaddr: invalid size (%lu)", (u_long) size);
865	}
866
867	/* Return EFAULT if the address was invalid, else zero */
868	return (rv);
869}
870
871#define MAXARGS	8
872static void
873syscall(struct thread *td, trapframe_t *frame, u_int32_t insn)
874{
875	struct proc *p = td->td_proc;
876	int code, error;
877	u_int nap, nargs;
878	register_t *ap, *args, copyargs[MAXARGS];
879	struct sysent *callp;
880	int locked = 0;
881
882	PCPU_LAZY_INC(cnt.v_syscall);
883	td->td_pticks = 0;
884	if (td->td_ucred != td->td_proc->p_ucred)
885		cred_update_thread(td);
886	switch (insn & SWI_OS_MASK) {
887	case 0: /* XXX: we need our own one. */
888		nap = 4;
889		break;
890	default:
891		call_trapsignal(td, SIGILL, 0);
892		userret(td, frame);
893		return;
894	}
895	code = insn & 0x000fffff;
896	td->td_pticks = 0;
897	ap = &frame->tf_r0;
898	if (code == SYS_syscall) {
899		code = *ap++;
900
901		nap--;
902	} else if (code == SYS___syscall) {
903		code = *ap++;
904		nap -= 2;
905		ap++;
906	}
907	if (p->p_sysent->sv_mask)
908		code &= p->p_sysent->sv_mask;
909	if (code >= p->p_sysent->sv_size)
910		callp = &p->p_sysent->sv_table[0];
911	else
912		callp = &p->p_sysent->sv_table[code];
913	nargs = callp->sy_narg & SYF_ARGMASK;
914	memcpy(copyargs, ap, nap * sizeof(register_t));
915	if (nargs > nap) {
916		error = copyin((void *)frame->tf_usr_sp, copyargs + nap,
917		    (nargs - nap) * sizeof(register_t));
918		if (error)
919			goto bad;
920	}
921	args = copyargs;
922	error = 0;
923#ifdef KTRACE
924	if (KTRPOINT(td, KTR_SYSCALL))
925		ktrsyscall(code, nargs, args);
926#endif
927
928	CTR4(KTR_SYSC, "syscall enter thread %p pid %d proc %s code %d", td,
929	    td->td_proc->p_pid, td->td_proc->p_comm, code);
930	if ((callp->sy_narg & SYF_MPSAFE) == 0)
931		mtx_lock(&Giant);
932	locked = 1;
933	if (error == 0) {
934		td->td_retval[0] = 0;
935		td->td_retval[1] = 0;
936		STOPEVENT(p, S_SCE, (callp->sy_narg & SYF_ARGMASK));
937		PTRACESTOP_SC(p, td, S_PT_SCE);
938		AUDIT_SYSCALL_ENTER(code, td);
939		error = (*callp->sy_call)(td, args);
940		AUDIT_SYSCALL_EXIT(error, td);
941		KASSERT(td->td_ar == NULL,
942		    ("returning from syscall with td_ar set!"));
943	}
944	switch (error) {
945	case 0:
946#ifdef __ARMEB__
947		if ((insn & 0x000fffff) &&
948		    (code != SYS_lseek)) {
949			/*
950			 * 64-bit return, 32-bit syscall. Fixup byte order
951			 */
952			frame->tf_r0 = 0;
953			frame->tf_r1 = td->td_retval[0];
954		} else {
955                        frame->tf_r0 = td->td_retval[0];
956                        frame->tf_r1 = td->td_retval[1];
957		}
958#else
959      		frame->tf_r0 = td->td_retval[0];
960		frame->tf_r1 = td->td_retval[1];
961#endif
962		frame->tf_spsr &= ~PSR_C_bit;   /* carry bit */
963		break;
964
965	case ERESTART:
966		/*
967		 * Reconstruct the pc to point at the swi.
968		 */
969		frame->tf_pc -= INSN_SIZE;
970		break;
971	case EJUSTRETURN:
972		/* nothing to do */
973		break;
974	default:
975bad:
976		frame->tf_r0 = error;
977		frame->tf_spsr |= PSR_C_bit;    /* carry bit */
978		break;
979	}
980	if (locked && (callp->sy_narg & SYF_MPSAFE) == 0)
981		mtx_unlock(&Giant);
982
983
984	userret(td, frame);
985	CTR4(KTR_SYSC, "syscall exit thread %p pid %d proc %s code %d", td,
986	    td->td_proc->p_pid, td->td_proc->p_comm, code);
987
988	STOPEVENT(p, S_SCX, code);
989	PTRACESTOP_SC(p, td, S_PT_SCX);
990#ifdef KTRACE
991      	if (KTRPOINT(td, KTR_SYSRET))
992		ktrsysret(code, error, td->td_retval[0]);
993#endif
994	mtx_assert(&sched_lock, MA_NOTOWNED);
995	mtx_assert(&Giant, MA_NOTOWNED);
996}
997
998void
999swi_handler(trapframe_t *frame)
1000{
1001	struct thread *td = curthread;
1002	uint32_t insn;
1003
1004	td->td_frame = frame;
1005
1006	td->td_pticks = 0;
1007	if (td->td_proc->p_flag & P_SA)
1008		thread_user_enter(td);
1009	/*
1010      	 * Make sure the program counter is correctly aligned so we
1011	 * don't take an alignment fault trying to read the opcode.
1012	 */
1013	if (__predict_false(((frame->tf_pc - INSN_SIZE) & 3) != 0)) {
1014		call_trapsignal(td, SIGILL, 0);
1015		userret(td, frame);
1016		return;
1017	}
1018	insn = *(u_int32_t *)(frame->tf_pc - INSN_SIZE);
1019	/*
1020	 * Enable interrupts if they were enabled before the exception.
1021	 * Since all syscalls *should* come from user mode it will always
1022	 * be safe to enable them, but check anyway.
1023	 */
1024	if (td->td_md.md_spinlock_count == 0) {
1025		if (__predict_true(frame->tf_spsr & I32_bit) == 0)
1026			enable_interrupts(I32_bit);
1027		if (__predict_true(frame->tf_spsr & F32_bit) == 0)
1028			enable_interrupts(F32_bit);
1029	}
1030
1031	syscall(td, frame, insn);
1032}
1033
1034