trap.c revision 97347 
1/*
2 * Copyright (C) 1995, 1996 Wolfgang Solfrank.
3 * Copyright (C) 1995, 1996 TooLs GmbH.
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 *    notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 * 3. All advertising materials mentioning features or use of this software
15 *    must display the following acknowledgement:
16 *	This product includes software developed by TooLs GmbH.
17 * 4. The name of TooLs GmbH may not be used to endorse or promote products
18 *    derived from this software without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
21 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
22 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
23 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
24 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
25 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
26 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
27 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
28 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
29 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30 *
31 * $NetBSD: trap.c,v 1.58 2002/03/04 04:07:35 dbj Exp $
32 */
33
34#ifndef lint
35static const char rcsid[] =
36  "$FreeBSD: head/sys/powerpc/aim/trap.c 97347 2002-05-27 11:20:19Z benno $";
37#endif /* not lint */
38
39#include "opt_ddb.h"
40#include "opt_ktrace.h"
41
42#include <sys/param.h>
43#include <sys/proc.h>
44#include <sys/ktr.h>
45#include <sys/lock.h>
46#include <sys/mutex.h>
47#include <sys/pioctl.h>
48#include <sys/reboot.h>
49#include <sys/syscall.h>
50#include <sys/sysent.h>
51#include <sys/systm.h>
52#include <sys/uio.h>
53#include <sys/user.h>
54#ifdef KTRACE
55#include <sys/ktrace.h>
56#endif
57#include <sys/vmmeter.h>
58
59#include <vm/vm.h>
60#include <vm/pmap.h>
61#include <vm/vm_extern.h>
62#include <vm/vm_param.h>
63#include <vm/vm_kern.h>
64#include <vm/vm_map.h>
65#include <vm/vm_page.h>
66
67#include <machine/cpu.h>
68#include <machine/db_machdep.h>
69#include <machine/fpu.h>
70#include <machine/frame.h>
71#include <machine/pcb.h>
72#include <machine/pmap.h>
73#include <machine/psl.h>
74#include <machine/trap.h>
75#include <machine/spr.h>
76#include <machine/sr.h>
77
78/* These definitions should probably be somewhere else			XXX */
79#define	FIRSTARG	3		/* first argument is in reg 3 */
80#define	NARGREG		8		/* 8 args are in registers */
81#define	MOREARGS(sp)	((caddr_t)((int)(sp) + 8)) /* more args go here */
82
83#ifndef MULTIPROCESSOR
84extern int intr_depth;
85#endif
86
87void		trap(struct trapframe *);
88
89static void	trap_fatal(struct trapframe *frame);
90static void	printtrap(u_int vector, struct trapframe *frame, int isfatal,
91		    int user);
92static int	trap_pfault(struct trapframe *frame, int user);
93static int	fix_unaligned(struct thread *td, struct trapframe *frame);
94static int	handle_onfault(struct trapframe *frame);
95static void	syscall(struct trapframe *frame);
96
97static __inline void	setusr(u_int);
98
99int	setfault(faultbuf);		/* defined in locore.S */
100
101/* Why are these not defined in a header? */
102int	badaddr(void *, size_t);
103int	badaddr_read(void *, size_t, int *);
104int	kcopy(const void *, void *, size_t);
105
106#ifdef	WITNESS
107extern char	*syscallnames[];
108#endif
109
110struct powerpc_exception {
111	u_int	vector;
112	char	*name;
113};
114
115static struct powerpc_exception powerpc_exceptions[] = {
116	{ 0x0100, "system reset" },
117	{ 0x0200, "machine check" },
118	{ 0x0300, "data storage interrupt" },
119	{ 0x0400, "instruction storage interrupt" },
120	{ 0x0500, "external interrupt" },
121	{ 0x0600, "alignment" },
122	{ 0x0700, "program" },
123	{ 0x0800, "floating-point unavailable" },
124	{ 0x0900, "decrementer" },
125	{ 0x0c00, "system call" },
126	{ 0x0d00, "trace" },
127	{ 0x0e00, "floating-point assist" },
128	{ 0x0f00, "performance monitoring" },
129	{ 0x0f20, "altivec unavailable" },
130	{ 0x1000, "instruction tlb miss" },
131	{ 0x1100, "data load tlb miss" },
132	{ 0x1200, "data store tlb miss" },
133	{ 0x1300, "instruction breakpoint" },
134	{ 0x1400, "system management" },
135	{ 0x1600, "altivec assist" },
136	{ 0x1700, "thermal management" },
137	{ 0x2000, "run mode/trace" },
138	{ 0x3000, NULL }
139};
140
141static const char *
142trapname(u_int vector)
143{
144	struct	powerpc_exception *pe;
145
146	for (pe = powerpc_exceptions; pe->vector != 0x3000; pe++) {
147		if (pe->vector == vector)
148			return (pe->name);
149	}
150
151	return ("unknown");
152}
153
154void
155trap(struct trapframe *frame)
156{
157	struct thread	*td, *fputhread;
158	struct proc	*p;
159	int		sig, type, user;
160	u_int		sticks, ucode;
161
162	atomic_add_int(&cnt.v_trap, 1);
163
164	td = PCPU_GET(curthread);
165	p = td->td_proc;
166
167	type = ucode = frame->exc;
168	sig = 0;
169	user = frame->srr1 & PSL_PR;
170	sticks = 0;
171
172	CTR3(KTR_TRAP, "trap: %s type=%s (%s)", p->p_comm,
173	    trapname(type), user ? "user" : "kernel");
174
175	if (user) {
176		sticks = td->td_kse->ke_sticks;
177		td->td_frame = frame;
178		if (td->td_ucred != p->p_ucred)
179			cred_update_thread(td);
180
181		/* User Mode Traps */
182		switch (type) {
183		case EXC_RUNMODETRC:
184		case EXC_TRC:
185			frame->srr1 &= ~PSL_SE;
186			sig = SIGTRAP;
187			break;
188
189		case EXC_DSI:
190		case EXC_ISI:
191			sig = trap_pfault(frame, 1);
192			break;
193
194		case EXC_SC:
195			syscall(frame);
196			break;
197
198		case EXC_FPU:
199			if ((fputhread = PCPU_GET(fputhread)) != NULL) {
200				KASSERT(fputhread != td,
201				    ("floating-point already enabled"));
202				save_fpu(fputhread);
203			}
204			PCPU_SET(fputhread, td);
205			td->td_pcb->pcb_fpcpu = PCPU_GET(cpuid);
206			enable_fpu(td);
207			frame->srr1 |= PSL_FP;
208			break;
209
210#ifdef	ALTIVEC
211		case EXC_VEC:
212			if ((vecthread = PCPU_GET(vecthread)) != NULL) {
213				KASSERT(vecthread != td,
214				    ("altivec already enabled"));
215				save_vec(vecthread);
216			}
217			PCPU_SET(vecthread, td);
218			td->td_pcb->pcb_veccpu = PCPU_GET(cpuid);
219			enable_vec(td);
220			frame->srr1 |= PSL_VEC;
221			break;
222#endif /* ALTIVEC */
223
224		case EXC_ALI:
225			if (fix_unaligned(td, frame) != 0)
226				sig = SIGBUS;
227			else
228				frame->srr0 += 4;
229			break;
230
231		case EXC_PGM:
232			/* XXX temporarily */
233			/* XXX: Magic Number? */
234			if (frame->srr1 & 0x0002000)
235				sig = SIGTRAP;
236 			else
237				sig = SIGILL;
238			break;
239
240		default:
241			trap_fatal(frame);
242		}
243	} else {
244		/* Kernel Mode Traps */
245
246		KASSERT(cold || td->td_ucred != NULL,
247		    ("kernel trap doesn't have ucred"));
248		switch (type) {
249		case EXC_DSI:
250			if (trap_pfault(frame, 0) == 0)
251 				return;
252			break;
253		case EXC_MCHK:
254			if (handle_onfault(frame))
255 				return;
256			break;
257		default:
258			trap_fatal(frame);
259		}
260	}
261
262	if (td != PCPU_GET(fputhread) ||
263	    td->td_pcb->pcb_fpcpu != PCPU_GET(cpuid))
264		frame->srr1 &= ~PSL_FP;
265
266#ifdef	ALTIVEC
267	if (td != PCPU_GET(vecthread) ||
268	    td->td_pcb->pcb_veccpu != PCPU_GET(cpuid))
269		frame->srr1 &= ~PSL_VEC;
270#endif /* ALTIVEC */
271
272	if (sig != 0) {
273		if (p->p_sysent->sv_transtrap != NULL)
274			sig = (p->p_sysent->sv_transtrap)(sig, type);
275		trapsignal(p, sig, ucode);
276	}
277
278	userret(td, frame, sticks);
279	mtx_assert(&Giant, MA_NOTOWNED);
280#ifdef	DIAGNOSTIC
281	cred_free_thread(td);
282#endif	/* DIAGNOSTIC */
283}
284
285static void
286trap_fatal(struct trapframe *frame)
287{
288
289	printtrap(frame->exc, frame, 1, (frame->srr1 & PSL_PR));
290#ifdef DDB
291	if ((debugger_on_panic || db_active) && kdb_trap(frame->exc, 0, frame))
292		return;
293#endif
294	panic("%s trap", trapname(frame->exc));
295}
296
297static void
298printtrap(u_int vector, struct trapframe *frame, int isfatal, int user)
299{
300
301	printf("\n");
302	printf("%s %s trap:\n", isfatal ? "fatal" : "handled",
303	    user ? "user" : "kernel");
304	printf("\n");
305	printf("   exception       = 0x%x (%s)\n", vector >> 8,
306	    trapname(vector));
307	switch (vector) {
308	case EXC_DSI:
309		printf("   virtual address = 0x%x\n", frame->dar);
310		break;
311	case EXC_ISI:
312		printf("   virtual address = 0x%x\n", frame->srr0);
313		break;
314	}
315	printf("   srr0            = 0x%x\n", frame->srr0);
316	printf("   srr1            = 0x%x\n", frame->srr1);
317	printf("   curthread       = %p\n", curthread);
318	if (curthread != NULL)
319		printf("          pid = %d, comm = %s\n",
320		    curthread->td_proc->p_pid, curthread->td_proc->p_comm);
321	printf("\n");
322}
323
324/*
325 * Handles a fatal fault when we have onfault state to recover.  Returns
326 * non-zero if there was onfault recovery state available.
327 */
328static int
329handle_onfault(struct trapframe *frame)
330{
331	struct		thread *td;
332	faultbuf	*fb;
333
334	td = curthread;
335	fb = td->td_pcb->pcb_onfault;
336	if (fb != NULL) {
337		frame->srr0 = (*fb)[0];
338		frame->fixreg[1] = (*fb)[1];
339		frame->fixreg[2] = (*fb)[2];
340		frame->cr = (*fb)[3];
341		bcopy(&(*fb)[4], &frame->fixreg[13],
342		    19 * sizeof(register_t));
343		return (1);
344	}
345	return (0);
346}
347
348void
349syscall(struct trapframe *frame)
350{
351	caddr_t		params;
352	struct		sysent *callp;
353	struct		thread *td;
354	struct		proc *p;
355	int		error, n;
356	size_t		narg;
357	register_t	args[10];
358	u_int		code;
359
360	td = PCPU_GET(curthread);
361	p = td->td_proc;
362
363	atomic_add_int(&cnt.v_syscall, 1);
364
365	code = frame->fixreg[0];
366	params = (caddr_t)(frame->fixreg + FIRSTARG);
367	n = NARGREG;
368
369	if (p->p_sysent->sv_prepsyscall) {
370		/*
371		 * The prep code is MP aware.
372		 */
373		(*p->p_sysent->sv_prepsyscall)(frame, args, &code, &params);
374	} else if (code == SYS_syscall) {
375		/*
376		 * code is first argument,
377		 * followed by actual args.
378		 */
379		code = *params++;
380		n -= 1;
381	} else if (code == SYS___syscall) {
382		/*
383		 * Like syscall, but code is a quad,
384		 * so as to maintain quad alignment
385		 * for the rest of the args.
386		 */
387		params++;
388		code = *params++;
389		n -= 2;
390	}
391
392 	if (p->p_sysent->sv_mask)
393 		code &= p->p_sysent->sv_mask;
394
395 	if (code >= p->p_sysent->sv_size)
396 		callp = &p->p_sysent->sv_table[0];
397  	else
398 		callp = &p->p_sysent->sv_table[code];
399
400	narg = callp->sy_narg & SYF_ARGMASK;
401
402	if (narg > n * sizeof(register_t)) {
403		bcopy(params, args, n * sizeof(register_t));
404		error = copyin(MOREARGS(frame->fixreg[1]), args + n,
405			narg - n * sizeof(register_t));
406		if (error) {
407#ifdef	KTRACE
408			/* Can't get all the arguments! */
409			if (KTRPOINT(p, KTR_SYSCALL))
410				ktrsyscall(p->p_tracep, code, narg, args);
411#endif
412			goto bad;
413		}
414		params = (caddr_t)args;
415	}
416
417	/*
418	 * Try to run the syscall without Giant if the syscall is MP safe.
419	 */
420	if ((callp->sy_narg & SYF_MPSAFE) == 0)
421		mtx_lock(&Giant);
422
423#ifdef	KTRACE
424	if (KTRPOINT(p, KTR_SYSCALL))
425		ktrsyscall(p->p_tracep, code, narg, params);
426#endif
427	td->td_retval[0] = 0;
428	td->td_retval[1] = frame->fixreg[FIRSTARG + 1];
429
430	STOPEVENT(p, S_SCE, narg);
431
432	error = (*callp->sy_call)(td, params);
433	switch (error) {
434	case 0:
435		frame->fixreg[FIRSTARG] = td->td_retval[0];
436		frame->fixreg[FIRSTARG + 1] = td->td_retval[1];
437		/* XXX: Magic number */
438		frame->cr &= ~0x10000000;
439		break;
440	case ERESTART:
441		/*
442		 * Set user's pc back to redo the system call.
443		 */
444		frame->srr0 -= 4;
445		break;
446	case EJUSTRETURN:
447		/* nothing to do */
448		break;
449	default:
450bad:
451		if (p->p_sysent->sv_errsize) {
452			if (error >= p->p_sysent->sv_errsize)
453				error = -1;	/* XXX */
454			else
455				error = p->p_sysent->sv_errtbl[error];
456		}
457		frame->fixreg[FIRSTARG] = error;
458		/* XXX: Magic number: Carry Flag Equivalent? */
459		frame->cr |= 0x10000000;
460		break;
461	}
462
463
464#ifdef	KTRACE
465	if (KTRPOINT(p, KTR_SYSRET))
466		ktrsysret(p->p_tracep, code, error, td->td_retval[0]);
467#endif
468
469	if ((callp->sy_narg & SYF_MPSAFE) == 0)
470		mtx_unlock(&Giant);
471
472	/*
473	 * Does the comment in the i386 code about errno apply here?
474	 */
475	STOPEVENT(p, S_SCX, code);
476
477#ifdef WITNESS
478	if (witness_list(td)) {
479		panic("system call %s returning with mutex(s) held\n",
480		    syscallnames[code]);
481	}
482#endif
483	mtx_assert(&sched_lock, MA_NOTOWNED);
484	mtx_assert(&Giant, MA_NOTOWNED);
485}
486
487static int
488trap_pfault(struct trapframe *frame, int user)
489{
490	vm_offset_t	eva, va;
491	struct		thread *td;
492	struct		proc *p;
493	vm_map_t	map;
494	vm_prot_t	ftype;
495	int		rv;
496	u_int		user_sr;
497
498	td = curthread;
499	p = td->td_proc;
500	if (frame->exc == EXC_ISI) {
501		eva = frame->srr0;
502		ftype = VM_PROT_READ | VM_PROT_EXECUTE;
503	} else {
504		eva = frame->dar;
505		if (frame->dsisr & DSISR_STORE)
506			ftype = VM_PROT_READ | VM_PROT_WRITE;
507		else
508			ftype = VM_PROT_READ;
509	}
510
511	if (user) {
512		map = &p->p_vmspace->vm_map;
513	} else {
514		if ((eva >> ADDR_SR_SHFT) == USER_SR) {
515			if (p->p_vmspace == NULL)
516				return (SIGSEGV);
517
518			__asm ("mfsr %0, %1"
519			    : "=r"(user_sr)
520			    : "K"(USER_SR));
521			eva &= ADDR_PIDX | ADDR_POFF;
522			eva |= user_sr << ADDR_SR_SHFT;
523			map = &p->p_vmspace->vm_map;
524		} else {
525			map = kernel_map;
526		}
527	}
528	va = trunc_page(eva);
529
530	mtx_lock(&Giant);
531	if (map != kernel_map) {
532		/*
533		 * Keep swapout from messing with us during this
534		 *	critical time.
535		 */
536		PROC_LOCK(p);
537		++p->p_lock;
538		PROC_UNLOCK(p);
539
540		/* Fault in the user page: */
541		rv = vm_fault(map, va, ftype,
542		      (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
543					      : VM_FAULT_NORMAL);
544
545		PROC_LOCK(p);
546		--p->p_lock;
547		PROC_UNLOCK(p);
548	} else {
549		/*
550		 * Don't have to worry about process locking or stacks in the
551		 * kernel.
552		 */
553		rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
554	}
555	mtx_unlock(&Giant);
556
557	if (rv == KERN_SUCCESS)
558		return (0);
559
560	if (!user && handle_onfault(frame))
561		return (0);
562
563	return (SIGSEGV);
564}
565
566static __inline void
567setusr(u_int content)
568{
569	__asm __volatile ("isync; mtsr %0,%1; isync"
570		      :: "n"(USER_SR), "r"(content));
571}
572
573/*
574 * kcopy(const void *src, void *dst, size_t len);
575 *
576 * Copy len bytes from src to dst, aborting if we encounter a fatal
577 * page fault.
578 *
579 * kcopy() _must_ save and restore the old fault handler since it is
580 * called by uiomove(), which may be in the path of servicing a non-fatal
581 * page fault.
582 */
583int
584kcopy(const void *src, void *dst, size_t len)
585{
586	struct thread	*td;
587	faultbuf	env, *oldfault;
588	int		rv;
589
590	td = PCPU_GET(curthread);
591	oldfault = td->td_pcb->pcb_onfault;
592	if ((rv = setfault(env)) != 0) {
593		td->td_pcb->pcb_onfault = oldfault;
594		return rv;
595	}
596
597	memcpy(dst, src, len);
598
599	td->td_pcb->pcb_onfault = oldfault;
600	return (0);
601}
602
603int
604badaddr(void *addr, size_t size)
605{
606	return (badaddr_read(addr, size, NULL));
607}
608
609int
610badaddr_read(void *addr, size_t size, int *rptr)
611{
612	struct thread	*td;
613	faultbuf	env;
614	int		x;
615
616	/* Get rid of any stale machine checks that have been waiting.  */
617	__asm __volatile ("sync; isync");
618
619	td = PCPU_GET(curthread);
620
621	if (setfault(env)) {
622		td->td_pcb->pcb_onfault = 0;
623		__asm __volatile ("sync");
624		return 1;
625	}
626
627	__asm __volatile ("sync");
628
629	switch (size) {
630	case 1:
631		x = *(volatile int8_t *)addr;
632		break;
633	case 2:
634		x = *(volatile int16_t *)addr;
635		break;
636	case 4:
637		x = *(volatile int32_t *)addr;
638		break;
639	default:
640		panic("badaddr: invalid size (%d)", size);
641	}
642
643	/* Make sure we took the machine check, if we caused one. */
644	__asm __volatile ("sync; isync");
645
646	td->td_pcb->pcb_onfault = 0;
647	__asm __volatile ("sync");	/* To be sure. */
648
649	/* Use the value to avoid reorder. */
650	if (rptr)
651		*rptr = x;
652
653	return (0);
654}
655
656/*
657 * For now, this only deals with the particular unaligned access case
658 * that gcc tends to generate.  Eventually it should handle all of the
659 * possibilities that can happen on a 32-bit PowerPC in big-endian mode.
660 */
661
662static int
663fix_unaligned(struct thread *td, struct trapframe *frame)
664{
665	struct thread	*fputhread;
666	int		indicator, reg;
667	double		*fpr;
668
669	indicator = EXC_ALI_OPCODE_INDICATOR(frame->dsisr);
670
671	switch (indicator) {
672	case EXC_ALI_LFD:
673	case EXC_ALI_STFD:
674		reg = EXC_ALI_RST(frame->dsisr);
675		fpr = &td->td_pcb->pcb_fpu.fpr[reg];
676		fputhread = PCPU_GET(fputhread);
677
678		/* Juggle the FPU to ensure that we've initialized
679		 * the FPRs, and that their current state is in
680		 * the PCB.
681		 */
682		if (fputhread != td) {
683			if (fputhread)
684				save_fpu(fputhread);
685			enable_fpu(td);
686		}
687		save_fpu(td);
688
689		if (indicator == EXC_ALI_LFD) {
690			if (copyin((void *)frame->dar, fpr,
691			    sizeof(double)) != 0)
692				return -1;
693			enable_fpu(td);
694		} else {
695			if (copyout(fpr, (void *)frame->dar,
696			    sizeof(double)) != 0)
697				return -1;
698		}
699		return 0;
700		break;
701	}
702
703	return -1;
704}
705