syscall.c revision 151316
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 151316 2005-10-14 12:43:47Z davidxu $");
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#ifdef KDB
123#include <sys/kdb.h>
124#endif
125
126
127void swi_handler(trapframe_t *);
128void undefinedinstruction(trapframe_t *);
129
130#include <machine/disassem.h>
131#include <machine/machdep.h>
132
133extern char fusubailout[];
134
135#ifdef DEBUG
136int last_fault_code;	/* For the benefit of pmap_fault_fixup() */
137#endif
138
139#if defined(CPU_ARM7TDMI)
140/* These CPUs may need data/prefetch abort fixups */
141#define	CPU_ABORT_FIXUP_REQUIRED
142#endif
143
144struct ksig {
145	int signb;
146	u_long code;
147};
148struct data_abort {
149	int (*func)(trapframe_t *, u_int, u_int, struct thread *, struct ksig *);
150	const char *desc;
151};
152
153static int dab_fatal(trapframe_t *, u_int, u_int, struct thread *, struct ksig *);
154static int dab_align(trapframe_t *, u_int, u_int, struct thread *, struct ksig *);
155static int dab_buserr(trapframe_t *, u_int, u_int, struct thread *, struct ksig *);
156
157static const struct data_abort data_aborts[] = {
158	{dab_fatal,	"Vector Exception"},
159	{dab_align,	"Alignment Fault 1"},
160	{dab_fatal,	"Terminal Exception"},
161	{dab_align,	"Alignment Fault 3"},
162	{dab_buserr,	"External Linefetch Abort (S)"},
163	{NULL,		"Translation Fault (S)"},
164	{dab_buserr,	"External Linefetch Abort (P)"},
165	{NULL,		"Translation Fault (P)"},
166	{dab_buserr,	"External Non-Linefetch Abort (S)"},
167	{NULL,		"Domain Fault (S)"},
168	{dab_buserr,	"External Non-Linefetch Abort (P)"},
169	{NULL,		"Domain Fault (P)"},
170	{dab_buserr,	"External Translation Abort (L1)"},
171	{NULL,		"Permission Fault (S)"},
172	{dab_buserr,	"External Translation Abort (L2)"},
173	{NULL,		"Permission Fault (P)"}
174};
175
176/* Determine if a fault came from user mode */
177#define	TRAP_USERMODE(tf)	((tf->tf_spsr & PSR_MODE) == PSR_USR32_MODE)
178
179/* Determine if 'x' is a permission fault */
180#define	IS_PERMISSION_FAULT(x)					\
181	(((1 << ((x) & FAULT_TYPE_MASK)) &			\
182	  ((1 << FAULT_PERM_P) | (1 << FAULT_PERM_S))) != 0)
183
184static __inline void
185call_trapsignal(struct thread *td, int sig, u_long code)
186{
187	ksiginfo_t ksi;
188
189	ksiginfo_init_trap(&ksi);
190	ksi.ksi_signo = sig;
191	ksi.ksi_code = (int)code;
192	trapsignal(td, &ksi);
193}
194
195static __inline int
196data_abort_fixup(trapframe_t *tf, u_int fsr, u_int far, struct thread *td, struct ksig *ksig)
197{
198#ifdef CPU_ABORT_FIXUP_REQUIRED
199	int error;
200
201	/* Call the cpu specific data abort fixup routine */
202	error = cpu_dataabt_fixup(tf);
203	if (__predict_true(error != ABORT_FIXUP_FAILED))
204		return (error);
205
206	/*
207	 * Oops, couldn't fix up the instruction
208	 */
209	printf("data_abort_fixup: fixup for %s mode data abort failed.\n",
210	    TRAP_USERMODE(tf) ? "user" : "kernel");
211	printf("pc = 0x%08x, opcode 0x%08x, insn = ", tf->tf_pc,
212	    *((u_int *)tf->tf_pc));
213	disassemble(tf->tf_pc);
214
215	/* Die now if this happened in kernel mode */
216	if (!TRAP_USERMODE(tf))
217		dab_fatal(tf, fsr, far, td, NULL, ksig);
218
219	return (error);
220#else
221	return (ABORT_FIXUP_OK);
222#endif /* CPU_ABORT_FIXUP_REQUIRED */
223}
224
225void
226data_abort_handler(trapframe_t *tf)
227{
228	struct vm_map *map;
229	struct pcb *pcb;
230	struct thread *td;
231	u_int user, far, fsr;
232	vm_prot_t ftype;
233	void *onfault;
234	vm_offset_t va;
235	u_int sticks = 0;
236	int error = 0;
237	struct ksig ksig;
238	struct proc *p;
239
240
241	/* Grab FAR/FSR before enabling interrupts */
242	far = cpu_faultaddress();
243	fsr = cpu_faultstatus();
244#if 0
245	printf("data abort: %p (from %p %p)\n", (void*)far, (void*)tf->tf_pc,
246	    (void*)tf->tf_svc_lr);
247#endif
248
249	/* Update vmmeter statistics */
250#if 0
251	vmexp.traps++;
252#endif
253
254	td = curthread;
255	p = td->td_proc;
256
257	PCPU_LAZY_INC(cnt.v_trap);
258	/* Data abort came from user mode? */
259	user = TRAP_USERMODE(tf);
260
261	if (user) {
262		sticks = td->td_sticks;                                                         td->td_frame = tf;
263		if (td->td_ucred != td->td_proc->p_ucred)
264			cred_update_thread(td);
265		if (td->td_pflags & TDP_SA)
266			thread_user_enter(td);
267
268	}
269	/* Grab the current pcb */
270	pcb = td->td_pcb;
271	/* Re-enable interrupts if they were enabled previously */
272	if (td->td_critnest == 0 && __predict_true(tf->tf_spsr & I32_bit) == 0)
273		enable_interrupts(I32_bit);
274
275	/* Invoke the appropriate handler, if necessary */
276	if (__predict_false(data_aborts[fsr & FAULT_TYPE_MASK].func != NULL)) {
277		if ((data_aborts[fsr & FAULT_TYPE_MASK].func)(tf, fsr, far,
278		    td, &ksig)) {
279			goto do_trapsignal;
280		}
281		goto out;
282	}
283
284	/*
285	 * At this point, we're dealing with one of the following data aborts:
286	 *
287	 *  FAULT_TRANS_S  - Translation -- Section
288	 *  FAULT_TRANS_P  - Translation -- Page
289	 *  FAULT_DOMAIN_S - Domain -- Section
290	 *  FAULT_DOMAIN_P - Domain -- Page
291	 *  FAULT_PERM_S   - Permission -- Section
292	 *  FAULT_PERM_P   - Permission -- Page
293	 *
294	 * These are the main virtual memory-related faults signalled by
295	 * the MMU.
296	 */
297
298	/* fusubailout is used by [fs]uswintr to avoid page faulting */
299	if (__predict_false(pcb->pcb_onfault == fusubailout)) {
300		tf->tf_r0 = EFAULT;
301		tf->tf_pc = (register_t)(intptr_t) pcb->pcb_onfault;
302		return;
303	}
304
305	/*
306	 * Make sure the Program Counter is sane. We could fall foul of
307	 * someone executing Thumb code, in which case the PC might not
308	 * be word-aligned. This would cause a kernel alignment fault
309	 * further down if we have to decode the current instruction.
310	 * XXX: It would be nice to be able to support Thumb at some point.
311	 */
312	if (__predict_false((tf->tf_pc & 3) != 0)) {
313		if (user) {
314			/*
315			 * Give the user an illegal instruction signal.
316			 */
317			/* Deliver a SIGILL to the process */
318			ksig.signb = SIGILL;
319			ksig.code = 0;
320			goto do_trapsignal;
321		}
322
323		/*
324		 * The kernel never executes Thumb code.
325		 */
326		printf("\ndata_abort_fault: Misaligned Kernel-mode "
327		    "Program Counter\n");
328		dab_fatal(tf, fsr, far, td, &ksig);
329	}
330
331	/* See if the cpu state needs to be fixed up */
332	switch (data_abort_fixup(tf, fsr, far, td, &ksig)) {
333	case ABORT_FIXUP_RETURN:
334		return;
335	case ABORT_FIXUP_FAILED:
336		/* Deliver a SIGILL to the process */
337		ksig.signb = SIGILL;
338		ksig.code = 0;
339		goto do_trapsignal;
340	default:
341		break;
342	}
343
344	va = trunc_page((vm_offset_t)far);
345
346	/*
347	 * It is only a kernel address space fault iff:
348	 *	1. user == 0  and
349	 *	2. pcb_onfault not set or
350	 *	3. pcb_onfault set and not LDRT/LDRBT/STRT/STRBT instruction.
351	 */
352	if (user == 0 && (va >= VM_MIN_KERNEL_ADDRESS ||
353	    (va < VM_MIN_ADDRESS && vector_page == ARM_VECTORS_LOW)) &&
354	    __predict_true((pcb->pcb_onfault == NULL ||
355	     (ReadWord(tf->tf_pc) & 0x05200000) != 0x04200000))) {
356		map = kernel_map;
357
358		/* Was the fault due to the FPE/IPKDB ? */
359		if (__predict_false((tf->tf_spsr & PSR_MODE)==PSR_UND32_MODE)) {
360
361			/*
362			 * Force exit via userret()
363			 * This is necessary as the FPE is an extension to
364			 * userland that actually runs in a priveledged mode
365			 * but uses USR mode permissions for its accesses.
366			 */
367			user = 1;
368			ksig.signb = SIGSEGV;
369			ksig.code = 0;
370			goto do_trapsignal;
371		}
372	} else {
373		map = &td->td_proc->p_vmspace->vm_map;
374	}
375
376	/*
377	 * We need to know whether the page should be mapped
378	 * as R or R/W. The MMU does not give us the info as
379	 * to whether the fault was caused by a read or a write.
380	 *
381	 * However, we know that a permission fault can only be
382	 * the result of a write to a read-only location, so
383	 * we can deal with those quickly.
384	 *
385	 * Otherwise we need to disassemble the instruction
386	 * responsible to determine if it was a write.
387	 */
388	if (IS_PERMISSION_FAULT(fsr)) {
389		ftype = VM_PROT_WRITE;
390	} else {
391		u_int insn = ReadWord(tf->tf_pc);
392
393		if (((insn & 0x0c100000) == 0x04000000) ||	/* STR/STRB */
394		    ((insn & 0x0e1000b0) == 0x000000b0) ||	/* STRH/STRD */
395		    ((insn & 0x0a100000) == 0x08000000))	/* STM/CDT */
396		{
397			ftype = VM_PROT_WRITE;
398	}
399		else
400		if ((insn & 0x0fb00ff0) == 0x01000090)		/* SWP */
401			ftype = VM_PROT_READ | VM_PROT_WRITE;
402		else
403			ftype = VM_PROT_READ;
404	}
405
406	/*
407	 * See if the fault is as a result of ref/mod emulation,
408	 * or domain mismatch.
409	 */
410#ifdef DEBUG
411	last_fault_code = fsr;
412#endif
413	if (pmap_fault_fixup(vmspace_pmap(td->td_proc->p_vmspace), va, ftype,
414	    user)) {
415		goto out;
416	}
417
418	onfault = pcb->pcb_onfault;
419	pcb->pcb_onfault = NULL;
420	if (map != kernel_map) {
421		PROC_LOCK(p);
422		p->p_lock++;
423		PROC_UNLOCK(p);
424	}
425	error = vm_fault(map, va, ftype, (ftype & VM_PROT_WRITE) ?
426	    VM_FAULT_DIRTY : VM_FAULT_NORMAL);
427	pcb->pcb_onfault = onfault;
428	if (__predict_true(error == 0)) {
429		goto out;
430	}
431
432	if (map != kernel_map) {
433		PROC_LOCK(p);
434		p->p_lock--;
435		PROC_UNLOCK(p);
436	}
437	if (user == 0) {
438		if (pcb->pcb_onfault) {
439			tf->tf_r0 = error;
440			tf->tf_pc = (register_t)(intptr_t) pcb->pcb_onfault;
441			return;
442		}
443
444		printf("\nvm_fault(%p, %x, %x, 0) -> %x\n", map, va, ftype,
445		    error);
446		dab_fatal(tf, fsr, far, td, &ksig);
447	}
448
449
450	if (error == ENOMEM) {
451		printf("VM: pid %d (%s), uid %d killed: "
452		    "out of swap\n", td->td_proc->p_pid, td->td_proc->p_comm,
453		    (td->td_proc->p_ucred) ?
454		     td->td_proc->p_ucred->cr_uid : -1);
455		ksig.signb = SIGKILL;
456	} else {
457		ksig.signb = SIGSEGV;
458	}
459	ksig.code = 0;
460do_trapsignal:
461	call_trapsignal(td, ksig.signb, ksig.code);
462out:
463	/* If returning to user mode, make sure to invoke userret() */
464	if (user)
465		userret(td, tf, sticks);
466}
467
468/*
469 * dab_fatal() handles the following data aborts:
470 *
471 *  FAULT_WRTBUF_0 - Vector Exception
472 *  FAULT_WRTBUF_1 - Terminal Exception
473 *
474 * We should never see these on a properly functioning system.
475 *
476 * This function is also called by the other handlers if they
477 * detect a fatal problem.
478 *
479 * Note: If 'l' is NULL, we assume we're dealing with a prefetch abort.
480 */
481static int
482dab_fatal(trapframe_t *tf, u_int fsr, u_int far, struct thread *td, struct ksig *ksig)
483{
484	const char *mode;
485
486	mode = TRAP_USERMODE(tf) ? "user" : "kernel";
487
488	if (td != NULL) {
489		printf("Fatal %s mode data abort: '%s'\n", mode,
490		    data_aborts[fsr & FAULT_TYPE_MASK].desc);
491		printf("trapframe: %p\nFSR=%08x, FAR=", tf, fsr);
492		if ((fsr & FAULT_IMPRECISE) == 0)
493			printf("%08x, ", far);
494		else
495			printf("Invalid,  ");
496		printf("spsr=%08x\n", tf->tf_spsr);
497	} else {
498		printf("Fatal %s mode prefetch abort at 0x%08x\n",
499		    mode, tf->tf_pc);
500		printf("trapframe: %p, spsr=%08x\n", tf, tf->tf_spsr);
501	}
502
503	printf("r0 =%08x, r1 =%08x, r2 =%08x, r3 =%08x\n",
504	    tf->tf_r0, tf->tf_r1, tf->tf_r2, tf->tf_r3);
505	printf("r4 =%08x, r5 =%08x, r6 =%08x, r7 =%08x\n",
506	    tf->tf_r4, tf->tf_r5, tf->tf_r6, tf->tf_r7);
507	printf("r8 =%08x, r9 =%08x, r10=%08x, r11=%08x\n",
508	    tf->tf_r8, tf->tf_r9, tf->tf_r10, tf->tf_r11);
509	printf("r12=%08x, ", tf->tf_r12);
510
511	if (TRAP_USERMODE(tf))
512		printf("usp=%08x, ulr=%08x",
513		    tf->tf_usr_sp, tf->tf_usr_lr);
514	else
515		printf("ssp=%08x, slr=%08x",
516		    tf->tf_svc_sp, tf->tf_svc_lr);
517	printf(", pc =%08x\n\n", tf->tf_pc);
518
519#ifdef KDB
520	kdb_trap(fsr, 0, tf);
521#endif
522	panic("Fatal abort");
523	/*NOTREACHED*/
524}
525
526/*
527 * dab_align() handles the following data aborts:
528 *
529 *  FAULT_ALIGN_0 - Alignment fault
530 *  FAULT_ALIGN_0 - Alignment fault
531 *
532 * These faults are fatal if they happen in kernel mode. Otherwise, we
533 * deliver a bus error to the process.
534 */
535static int
536dab_align(trapframe_t *tf, u_int fsr, u_int far, struct thread *td, struct ksig *ksig)
537{
538
539	/* Alignment faults are always fatal if they occur in kernel mode */
540	if (!TRAP_USERMODE(tf)) {
541		if (!td || !td->td_pcb->pcb_onfault)
542			dab_fatal(tf, fsr, far, td, ksig);
543		tf->tf_r0 = EFAULT;
544		tf->tf_pc = (int)td->td_pcb->pcb_onfault;
545		return (0);
546	}
547
548	/* pcb_onfault *must* be NULL at this point */
549
550	/* See if the cpu state needs to be fixed up */
551	(void) data_abort_fixup(tf, fsr, far, td, ksig);
552
553	/* Deliver a bus error signal to the process */
554	ksig->code = 0;
555	ksig->signb = SIGBUS;
556	td->td_frame = tf;
557
558	return (1);
559}
560
561/*
562 * dab_buserr() handles the following data aborts:
563 *
564 *  FAULT_BUSERR_0 - External Abort on Linefetch -- Section
565 *  FAULT_BUSERR_1 - External Abort on Linefetch -- Page
566 *  FAULT_BUSERR_2 - External Abort on Non-linefetch -- Section
567 *  FAULT_BUSERR_3 - External Abort on Non-linefetch -- Page
568 *  FAULT_BUSTRNL1 - External abort on Translation -- Level 1
569 *  FAULT_BUSTRNL2 - External abort on Translation -- Level 2
570 *
571 * If pcb_onfault is set, flag the fault and return to the handler.
572 * If the fault occurred in user mode, give the process a SIGBUS.
573 *
574 * Note: On XScale, FAULT_BUSERR_0, FAULT_BUSERR_1, and FAULT_BUSERR_2
575 * can be flagged as imprecise in the FSR. This causes a real headache
576 * since some of the machine state is lost. In this case, tf->tf_pc
577 * may not actually point to the offending instruction. In fact, if
578 * we've taken a double abort fault, it generally points somewhere near
579 * the top of "data_abort_entry" in exception.S.
580 *
581 * In all other cases, these data aborts are considered fatal.
582 */
583static int
584dab_buserr(trapframe_t *tf, u_int fsr, u_int far, struct thread *td, struct ksig *ksig)
585{
586	struct pcb *pcb = td->td_pcb;
587
588#ifdef __XSCALE__
589	if ((fsr & FAULT_IMPRECISE) != 0 &&
590	    (tf->tf_spsr & PSR_MODE) == PSR_ABT32_MODE) {
591		/*
592		 * Oops, an imprecise, double abort fault. We've lost the
593		 * r14_abt/spsr_abt values corresponding to the original
594		 * abort, and the spsr saved in the trapframe indicates
595		 * ABT mode.
596		 */
597		tf->tf_spsr &= ~PSR_MODE;
598
599		/*
600		 * We use a simple heuristic to determine if the double abort
601		 * happened as a result of a kernel or user mode access.
602		 * If the current trapframe is at the top of the kernel stack,
603		 * the fault _must_ have come from user mode.
604		 */
605		if (tf != ((trapframe_t *)pcb->un_32.pcb32_sp) - 1) {
606			/*
607			 * Kernel mode. We're either about to die a
608			 * spectacular death, or pcb_onfault will come
609			 * to our rescue. Either way, the current value
610			 * of tf->tf_pc is irrelevant.
611			 */
612			tf->tf_spsr |= PSR_SVC32_MODE;
613			if (pcb->pcb_onfault == NULL)
614				printf("\nKernel mode double abort!\n");
615		} else {
616			/*
617			 * User mode. We've lost the program counter at the
618			 * time of the fault (not that it was accurate anyway;
619			 * it's not called an imprecise fault for nothing).
620			 * About all we can do is copy r14_usr to tf_pc and
621			 * hope for the best. The process is about to get a
622			 * SIGBUS, so it's probably history anyway.
623			 */
624			tf->tf_spsr |= PSR_USR32_MODE;
625			tf->tf_pc = tf->tf_usr_lr;
626		}
627	}
628
629	/* FAR is invalid for imprecise exceptions */
630	if ((fsr & FAULT_IMPRECISE) != 0)
631		far = 0;
632#endif /* __XSCALE__ */
633
634	if (pcb->pcb_onfault) {
635		tf->tf_r0 = EFAULT;
636		tf->tf_pc = (register_t)(intptr_t) pcb->pcb_onfault;
637		return (0);
638	}
639
640	/* See if the cpu state needs to be fixed up */
641	(void) data_abort_fixup(tf, fsr, far, td, ksig);
642
643	/*
644	 * At this point, if the fault happened in kernel mode, we're toast
645	 */
646	if (!TRAP_USERMODE(tf))
647		dab_fatal(tf, fsr, far, td, ksig);
648
649	/* Deliver a bus error signal to the process */
650	ksig->signb = SIGBUS;
651	ksig->code = 0;
652	td->td_frame = tf;
653
654	return (1);
655}
656
657static __inline int
658prefetch_abort_fixup(trapframe_t *tf, struct ksig *ksig)
659{
660#ifdef CPU_ABORT_FIXUP_REQUIRED
661	int error;
662
663	/* Call the cpu specific prefetch abort fixup routine */
664	error = cpu_prefetchabt_fixup(tf);
665	if (__predict_true(error != ABORT_FIXUP_FAILED))
666		return (error);
667
668	/*
669	 * Oops, couldn't fix up the instruction
670	 */
671	printf(
672	    "prefetch_abort_fixup: fixup for %s mode prefetch abort failed.\n",
673	    TRAP_USERMODE(tf) ? "user" : "kernel");
674	printf("pc = 0x%08x, opcode 0x%08x, insn = ", tf->tf_pc,
675	    *((u_int *)tf->tf_pc));
676	disassemble(tf->tf_pc);
677
678	/* Die now if this happened in kernel mode */
679	if (!TRAP_USERMODE(tf))
680		dab_fatal(tf, 0, tf->tf_pc, NULL, ksig);
681
682	return (error);
683#else
684	return (ABORT_FIXUP_OK);
685#endif /* CPU_ABORT_FIXUP_REQUIRED */
686}
687
688/*
689 * void prefetch_abort_handler(trapframe_t *tf)
690 *
691 * Abort handler called when instruction execution occurs at
692 * a non existent or restricted (access permissions) memory page.
693 * If the address is invalid and we were in SVC mode then panic as
694 * the kernel should never prefetch abort.
695 * If the address is invalid and the page is mapped then the user process
696 * does no have read permission so send it a signal.
697 * Otherwise fault the page in and try again.
698 */
699void
700prefetch_abort_handler(trapframe_t *tf)
701{
702	struct thread *td;
703	struct proc * p;
704	struct vm_map *map;
705	vm_offset_t fault_pc, va;
706	int error = 0;
707	u_int sticks = 0;
708	struct ksig ksig;
709
710
711#if 0
712	/* Update vmmeter statistics */
713	uvmexp.traps++;
714#endif
715#if 0
716	printf("prefetch abort handler: %p %p\n", (void*)tf->tf_pc,
717	    (void*)tf->tf_usr_lr);
718#endif
719
720 	td = curthread;
721	p = td->td_proc;
722	PCPU_LAZY_INC(cnt.v_trap);
723
724	if (TRAP_USERMODE(tf)) {
725		td->td_frame = tf;
726		if (td->td_ucred != td->td_proc->p_ucred)
727			cred_update_thread(td);
728		if (td->td_proc->p_flag & P_SA)
729			thread_user_enter(td);
730	}
731	fault_pc = tf->tf_pc;
732	if (td->td_critnest == 0 &&
733	    __predict_true((tf->tf_spsr & I32_bit) == 0))
734		enable_interrupts(I32_bit);
735
736
737	/* See if the cpu state needs to be fixed up */
738	switch (prefetch_abort_fixup(tf, &ksig)) {
739	case ABORT_FIXUP_RETURN:
740		return;
741	case ABORT_FIXUP_FAILED:
742		/* Deliver a SIGILL to the process */
743		ksig.signb = SIGILL;
744		ksig.code = 0;
745		td->td_frame = tf;
746		goto do_trapsignal;
747	default:
748		break;
749	}
750
751	/* Prefetch aborts cannot happen in kernel mode */
752	if (__predict_false(!TRAP_USERMODE(tf)))
753		dab_fatal(tf, 0, tf->tf_pc, NULL, &ksig);
754	sticks = td->td_sticks;
755
756
757	/* Ok validate the address, can only execute in USER space */
758	if (__predict_false(fault_pc >= VM_MAXUSER_ADDRESS ||
759	    (fault_pc < VM_MIN_ADDRESS && vector_page == ARM_VECTORS_LOW))) {
760		ksig.signb = SIGSEGV;
761		ksig.code = 0;
762		goto do_trapsignal;
763	}
764
765	map = &td->td_proc->p_vmspace->vm_map;
766	va = trunc_page(fault_pc);
767
768	/*
769	 * See if the pmap can handle this fault on its own...
770	 */
771#ifdef DEBUG
772	last_fault_code = -1;
773#endif
774	if (pmap_fault_fixup(map->pmap, va, VM_PROT_READ, 1))
775		goto out;
776
777	if (map != kernel_map) {
778		PROC_LOCK(p);
779		p->p_lock++;
780		PROC_UNLOCK(p);
781	}
782
783	error = vm_fault(map, va, VM_PROT_READ | VM_PROT_EXECUTE,
784	    VM_FAULT_NORMAL);
785	if (map != kernel_map) {
786		PROC_LOCK(p);
787		p->p_lock--;
788		PROC_UNLOCK(p);
789	}
790
791	if (__predict_true(error == 0))
792		goto out;
793
794	if (error == ENOMEM) {
795		printf("VM: pid %d (%s), uid %d killed: "
796		    "out of swap\n", td->td_proc->p_pid, td->td_proc->p_comm,
797		    (td->td_proc->p_ucred) ?
798		     td->td_proc->p_ucred->cr_uid : -1);
799		ksig.signb = SIGKILL;
800	} else {
801		ksig.signb = SIGSEGV;
802	}
803	ksig.code = 0;
804
805do_trapsignal:
806	call_trapsignal(td, ksig.signb, ksig.code);
807
808out:
809	userret(td, tf, sticks);
810
811}
812
813extern int badaddr_read_1(const uint8_t *, uint8_t *);
814extern int badaddr_read_2(const uint16_t *, uint16_t *);
815extern int badaddr_read_4(const uint32_t *, uint32_t *);
816/*
817 * Tentatively read an 8, 16, or 32-bit value from 'addr'.
818 * If the read succeeds, the value is written to 'rptr' and zero is returned.
819 * Else, return EFAULT.
820 */
821int
822badaddr_read(void *addr, size_t size, void *rptr)
823{
824	union {
825		uint8_t v1;
826		uint16_t v2;
827		uint32_t v4;
828	} u;
829	int rv;
830
831	cpu_drain_writebuf();
832
833	/* Read from the test address. */
834	switch (size) {
835	case sizeof(uint8_t):
836		rv = badaddr_read_1(addr, &u.v1);
837		if (rv == 0 && rptr)
838			*(uint8_t *) rptr = u.v1;
839		break;
840
841	case sizeof(uint16_t):
842		rv = badaddr_read_2(addr, &u.v2);
843		if (rv == 0 && rptr)
844			*(uint16_t *) rptr = u.v2;
845		break;
846
847	case sizeof(uint32_t):
848		rv = badaddr_read_4(addr, &u.v4);
849		if (rv == 0 && rptr)
850			*(uint32_t *) rptr = u.v4;
851		break;
852
853	default:
854		panic("badaddr: invalid size (%lu)", (u_long) size);
855	}
856
857	/* Return EFAULT if the address was invalid, else zero */
858	return (rv);
859}
860
861#define MAXARGS	8
862static void
863syscall(struct thread *td, trapframe_t *frame, u_int32_t insn)
864{
865	struct proc *p = td->td_proc;
866	int code, error;
867	u_int nap, nargs;
868	register_t *ap, *args, copyargs[MAXARGS];
869	struct sysent *callp;
870	int locked = 0;
871	u_int sticks = 0;
872
873	PCPU_LAZY_INC(cnt.v_syscall);
874	sticks = td->td_sticks;
875	if (td->td_ucred != td->td_proc->p_ucred)
876		cred_update_thread(td);
877	switch (insn & SWI_OS_MASK) {
878	case 0: /* XXX: we need our own one. */
879		nap = 4;
880		break;
881	default:
882		call_trapsignal(td, SIGILL, 0);
883		userret(td, frame, td->td_sticks);
884		return;
885	}
886	code = insn & 0x000fffff;
887	sticks = td->td_sticks;
888	ap = &frame->tf_r0;
889	if (code == SYS_syscall) {
890		code = *ap++;
891
892		nap--;
893	} else if (code == SYS___syscall) {
894		code = *ap++;
895		nap -= 2;
896		ap++;
897	}
898	if (p->p_sysent->sv_mask)
899		code &= p->p_sysent->sv_mask;
900	if (code >= p->p_sysent->sv_size)
901		callp = &p->p_sysent->sv_table[0];
902	else
903		callp = &p->p_sysent->sv_table[code];
904	nargs = callp->sy_narg & SYF_ARGMASK;
905	memcpy(copyargs, ap, nap * sizeof(register_t));
906	if (nargs > nap) {
907		error = copyin((void *)frame->tf_usr_sp, copyargs + nap,
908		    (nargs - nap) * sizeof(register_t));
909		if (error)
910			goto bad;
911	}
912	args = copyargs;
913	error = 0;
914#ifdef KTRACE
915	if (KTRPOINT(td, KTR_SYSCALL))
916		ktrsyscall(code, nargs, args);
917#endif
918
919	CTR4(KTR_SYSC, "syscall enter thread %p pid %d proc %s code %d", td,
920	    td->td_proc->p_pid, td->td_proc->p_comm, code);
921	if ((callp->sy_narg & SYF_MPSAFE) == 0)
922		mtx_lock(&Giant);
923	locked = 1;
924	if (error == 0) {
925		td->td_retval[0] = 0;
926		td->td_retval[1] = 0;
927		STOPEVENT(p, S_SCE, (callp->sy_narg & SYF_ARGMASK));
928		PTRACESTOP_SC(p, td, S_PT_SCE);
929		error = (*callp->sy_call)(td, args);
930	}
931	switch (error) {
932	case 0:
933#ifdef __ARMEB__
934		if ((insn & 0x000fffff) &&
935		    (code != SYS_lseek)) {
936			/*
937			 * 64-bit return, 32-bit syscall. Fixup byte order
938			 */
939			frame->tf_r0 = 0;
940			frame->tf_r1 = td->td_retval[0];
941		} else {
942                        frame->tf_r0 = td->td_retval[0];
943                        frame->tf_r1 = td->td_retval[1];
944		}
945#else
946      		frame->tf_r0 = td->td_retval[0];
947		frame->tf_r1 = td->td_retval[1];
948#endif
949		frame->tf_spsr &= ~PSR_C_bit;   /* carry bit */
950		break;
951
952	case ERESTART:
953		/*
954		 * Reconstruct the pc to point at the swi.
955		 */
956		frame->tf_pc -= INSN_SIZE;
957		break;
958	case EJUSTRETURN:
959		/* nothing to do */
960		break;
961	default:
962bad:
963		frame->tf_r0 = error;
964		frame->tf_spsr |= PSR_C_bit;    /* carry bit */
965		break;
966	}
967	if (locked && (callp->sy_narg & SYF_MPSAFE) == 0)
968		mtx_unlock(&Giant);
969
970
971	userret(td, frame, sticks);
972	CTR4(KTR_SYSC, "syscall exit thread %p pid %d proc %s code %d", td,
973	    td->td_proc->p_pid, td->td_proc->p_comm, code);
974
975	STOPEVENT(p, S_SCX, code);
976	PTRACESTOP_SC(p, td, S_PT_SCX);
977#ifdef KTRACE
978      	if (KTRPOINT(td, KTR_SYSRET))
979		ktrsysret(code, error, td->td_retval[0]);
980#endif
981	mtx_assert(&sched_lock, MA_NOTOWNED);
982	mtx_assert(&Giant, MA_NOTOWNED);
983}
984
985void
986swi_handler(trapframe_t *frame)
987{
988	struct thread *td = curthread;
989	uint32_t insn;
990
991	td->td_frame = frame;
992
993	if (td->td_proc->p_flag & P_SA)
994		thread_user_enter(td);
995	/*
996      	 * Make sure the program counter is correctly aligned so we
997	 * don't take an alignment fault trying to read the opcode.
998	 */
999	if (__predict_false(((frame->tf_pc - INSN_SIZE) & 3) != 0)) {
1000		call_trapsignal(td, SIGILL, 0);
1001		userret(td, frame, td->td_sticks);
1002		return;
1003	}
1004	insn = *(u_int32_t *)(frame->tf_pc - INSN_SIZE);
1005	/*
1006	 * Enable interrupts if they were enabled before the exception.
1007	 * Since all syscalls *should* come from user mode it will always
1008	 * be safe to enable them, but check anyway.
1009	 */
1010	if (td->td_critnest == 0 && !(frame->tf_spsr & I32_bit))
1011		enable_interrupts(I32_bit);
1012
1013	syscall(td, frame, insn);
1014}
1015
1016