trap.c revision 1.57
1/*	$NetBSD: trap.c,v 1.57 2010/12/20 00:25:41 matt Exp $	*/
2
3/*
4 * Copyright 2001 Wasabi Systems, Inc.
5 * All rights reserved.
6 *
7 * Written by Eduardo Horvath and Simon Burge for Wasabi Systems, Inc.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 *    notice, this list of conditions and the following disclaimer in the
16 *    documentation and/or other materials provided with the distribution.
17 * 3. All advertising materials mentioning features or use of this software
18 *    must display the following acknowledgement:
19 *      This product includes software developed for the NetBSD Project by
20 *      Wasabi Systems, Inc.
21 * 4. The name of Wasabi Systems, Inc. may not be used to endorse
22 *    or promote products derived from this software without specific prior
23 *    written permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
27 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
28 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
29 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
30 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
31 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
32 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
33 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
34 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
35 * POSSIBILITY OF SUCH DAMAGE.
36 */
37
38/*
39 * Copyright (C) 1995, 1996 Wolfgang Solfrank.
40 * Copyright (C) 1995, 1996 TooLs GmbH.
41 * All rights reserved.
42 *
43 * Redistribution and use in source and binary forms, with or without
44 * modification, are permitted provided that the following conditions
45 * are met:
46 * 1. Redistributions of source code must retain the above copyright
47 *    notice, this list of conditions and the following disclaimer.
48 * 2. Redistributions in binary form must reproduce the above copyright
49 *    notice, this list of conditions and the following disclaimer in the
50 *    documentation and/or other materials provided with the distribution.
51 * 3. All advertising materials mentioning features or use of this software
52 *    must display the following acknowledgement:
53 *	This product includes software developed by TooLs GmbH.
54 * 4. The name of TooLs GmbH may not be used to endorse or promote products
55 *    derived from this software without specific prior written permission.
56 *
57 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
58 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
59 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
60 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
61 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
62 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
63 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
64 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
65 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
66 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
67 */
68
69#include <sys/cdefs.h>
70__KERNEL_RCSID(0, "$NetBSD: trap.c,v 1.57 2010/12/20 00:25:41 matt Exp $");
71
72#include "opt_altivec.h"
73#include "opt_ddb.h"
74#include "opt_kgdb.h"
75
76#include <sys/param.h>
77#include <sys/proc.h>
78#include <sys/reboot.h>
79#include <sys/syscall.h>
80#include <sys/systm.h>
81#include <sys/sa.h>
82#include <sys/savar.h>
83#include <sys/userret.h>
84#include <sys/kauth.h>
85#include <sys/kmem.h>
86
87#if defined(KGDB)
88#include <sys/kgdb.h>
89#endif
90
91#include <uvm/uvm_extern.h>
92
93#include <dev/cons.h>
94
95#include <machine/cpu.h>
96#include <machine/db_machdep.h>
97#include <machine/fpu.h>
98#include <machine/frame.h>
99#include <machine/pcb.h>
100#include <machine/psl.h>
101#include <machine/trap.h>
102
103#include <powerpc/spr.h>
104#include <powerpc/ibm4xx/spr.h>
105#include <powerpc/ibm4xx/pmap.h>
106#include <powerpc/ibm4xx/tlb.h>
107#include <powerpc/fpu/fpu_extern.h>
108
109/* These definitions should probably be somewhere else			XXX */
110#define	FIRSTARG	3		/* first argument is in reg 3 */
111#define	NARGREG		8		/* 8 args are in registers */
112#define	MOREARGS(sp)	((void *)((int)(sp) + 8)) /* more args go here */
113
114static int fix_unaligned(struct lwp *l, struct trapframe *frame);
115
116void trap(struct trapframe *);	/* Called from locore / trap_subr */
117/* Why are these not defined in a header? */
118int badaddr(void *, size_t);
119int badaddr_read(void *, size_t, int *);
120int ctx_setup(int, int);
121
122#ifdef DEBUG
123#define TDB_ALL	0x1
124int trapdebug = /* TDB_ALL */ 0;
125#define	DBPRINTF(x, y)	if (trapdebug & (x)) printf y
126#else
127#define DBPRINTF(x, y)
128#endif
129
130void
131trap(struct trapframe *frame)
132{
133	struct lwp *l = curlwp;
134	struct proc *p = l->l_proc;
135	struct pcb *pcb;
136	int type = frame->exc;
137	int ftype, rv;
138	ksiginfo_t ksi;
139
140	KASSERT(l->l_stat == LSONPROC);
141
142	if (frame->srr1 & PSL_PR) {
143		LWP_CACHE_CREDS(l, p);
144		type |= EXC_USER;
145	}
146
147	ftype = VM_PROT_READ;
148
149	DBPRINTF(TDB_ALL, ("trap(%x) at %lx from frame %p &frame %p\n",
150	    type, frame->srr0, frame, &frame));
151
152	switch (type) {
153	case EXC_DEBUG|EXC_USER:
154		{
155			int srr2, srr3;
156
157			__asm volatile("mfspr %0,0x3f0" :
158			    "=r" (rv), "=r" (srr2), "=r" (srr3) :);
159			printf("debug reg is %x srr2 %x srr3 %x\n", rv, srr2,
160			    srr3);
161			/* XXX fall through or break here?! */
162		}
163		/*
164		 * DEBUG intr -- probably single-step.
165		 */
166	case EXC_TRC|EXC_USER:
167		frame->srr1 &= ~PSL_SE;
168		KSI_INIT_TRAP(&ksi);
169		ksi.ksi_signo = SIGTRAP;
170		ksi.ksi_trap = EXC_TRC;
171		ksi.ksi_addr = (void *)frame->srr0;
172		trapsignal(l, &ksi);
173		break;
174
175	/*
176	 * If we could not find and install appropriate TLB entry, fall through.
177	 */
178
179	case EXC_DSI:
180		/* FALLTHROUGH */
181	case EXC_DTMISS:
182		{
183			struct vm_map *map;
184			vaddr_t va;
185			struct faultbuf *fb = NULL;
186
187			va = frame->dar;
188			if (frame->tf_xtra[TF_PID] == KERNEL_PID) {
189				map = kernel_map;
190			} else {
191				map = &p->p_vmspace->vm_map;
192				if ((l->l_flag & LW_SA)
193				    && (~l->l_pflag & LP_SA_NOBLOCK)) {
194					l->l_savp->savp_faultaddr = va;
195					l->l_pflag |= LP_SA_PAGEFAULT;
196				}
197			}
198
199			if (frame->tf_xtra[TF_ESR] & (ESR_DST|ESR_DIZ))
200				ftype = VM_PROT_WRITE;
201
202			DBPRINTF(TDB_ALL,
203			    ("trap(EXC_DSI) at %lx %s fault on %p esr %x\n",
204			    frame->srr0,
205			    (ftype & VM_PROT_WRITE) ? "write" : "read",
206			    (void *)va, frame->tf_xtra[TF_ESR]));
207
208			pcb = lwp_getpcb(l);
209			fb = pcb->pcb_onfault;
210			pcb->pcb_onfault = NULL;
211			rv = uvm_fault(map, trunc_page(va), ftype);
212			pcb->pcb_onfault = fb;
213			if (map != kernel_map) {
214				l->l_pflag &= ~LP_SA_PAGEFAULT;
215			}
216			if (rv == 0)
217				goto done;
218			if (fb != NULL) {
219				frame->tf_xtra[TF_PID] = KERNEL_PID;
220				frame->srr0 = fb->fb_pc;
221				frame->srr1 |= PSL_IR; /* Re-enable IMMU */
222				frame->fixreg[1] = fb->fb_sp;
223				frame->fixreg[2] = fb->fb_r2;
224				frame->fixreg[3] = rv;
225				frame->cr = fb->fb_cr;
226				memcpy(&frame->fixreg[13], fb->fb_fixreg,
227				    sizeof(fb->fb_fixreg));
228				goto done;
229			}
230		}
231		goto brain_damage;
232
233	case EXC_DSI|EXC_USER:
234		/* FALLTHROUGH */
235	case EXC_DTMISS|EXC_USER:
236		if (frame->tf_xtra[TF_ESR] & (ESR_DST|ESR_DIZ))
237			ftype = VM_PROT_WRITE;
238
239		DBPRINTF(TDB_ALL,
240		    ("trap(EXC_DSI|EXC_USER) at %lx %s fault on %lx %x\n",
241		    frame->srr0, (ftype & VM_PROT_WRITE) ? "write" : "read",
242		    frame->dar, frame->tf_xtra[TF_ESR]));
243		KASSERT(l == curlwp && (l->l_stat == LSONPROC));
244		if (l->l_flag & LW_SA) {
245			l->l_savp->savp_faultaddr = (vaddr_t)frame->dar;
246			l->l_pflag |= LP_SA_PAGEFAULT;
247		}
248//		KASSERT(curpcb->pcb_onfault == NULL);
249		rv = uvm_fault(&p->p_vmspace->vm_map, trunc_page(frame->dar),
250		    ftype);
251		if (rv == 0) {
252			l->l_pflag &= ~LP_SA_PAGEFAULT;
253			break;
254		}
255		KSI_INIT_TRAP(&ksi);
256		ksi.ksi_signo = SIGSEGV;
257		ksi.ksi_trap = EXC_DSI;
258		ksi.ksi_addr = (void *)frame->dar;
259		if (rv == ENOMEM) {
260			printf("UVM: pid %d (%s) lid %d, uid %d killed: "
261			    "out of swap\n",
262			    p->p_pid, p->p_comm, l->l_lid,
263			    l->l_cred ?
264			    kauth_cred_geteuid(l->l_cred) : -1);
265			ksi.ksi_signo = SIGKILL;
266		}
267		trapsignal(l, &ksi);
268		l->l_pflag &= ~LP_SA_PAGEFAULT;
269		break;
270
271	case EXC_ITMISS|EXC_USER:
272	case EXC_ISI|EXC_USER:
273		if (l->l_flag & LW_SA) {
274			l->l_savp->savp_faultaddr = (vaddr_t)frame->srr0;
275			l->l_pflag |= LP_SA_PAGEFAULT;
276		}
277		ftype = VM_PROT_EXECUTE;
278		DBPRINTF(TDB_ALL,
279		    ("trap(EXC_ISI|EXC_USER) at %lx execute fault tf %p\n",
280		    frame->srr0, frame));
281//		KASSERT(curpcb->pcb_onfault == NULL);
282		rv = uvm_fault(&p->p_vmspace->vm_map, trunc_page(frame->srr0),
283		    ftype);
284		if (rv == 0) {
285			l->l_pflag &= ~LP_SA_PAGEFAULT;
286			break;
287		}
288		KSI_INIT_TRAP(&ksi);
289		ksi.ksi_signo = SIGSEGV;
290		ksi.ksi_trap = EXC_ISI;
291		ksi.ksi_addr = (void *)frame->srr0;
292		ksi.ksi_code = (rv == EACCES ? SEGV_ACCERR : SEGV_MAPERR);
293		trapsignal(l, &ksi);
294		l->l_pflag &= ~LP_SA_PAGEFAULT;
295		break;
296
297	case EXC_AST|EXC_USER:
298		curcpu()->ci_astpending = 0;	/* we are about to do it */
299		//curcpu()->ci_data.cpu_nast++;
300		if (l->l_pflag & LP_OWEUPC) {
301			l->l_pflag &= ~LP_OWEUPC;
302			ADDUPROF(l);
303		}
304		/* Check whether we are being preempted. */
305		if (curcpu()->ci_want_resched)
306			preempt();
307		break;
308
309
310	case EXC_ALI|EXC_USER:
311		if (fix_unaligned(l, frame) != 0) {
312			KSI_INIT_TRAP(&ksi);
313			ksi.ksi_signo = SIGBUS;
314			ksi.ksi_trap = EXC_ALI;
315			ksi.ksi_addr = (void *)frame->dar;
316			trapsignal(l, &ksi);
317		} else
318			frame->srr0 += 4;
319		break;
320
321	case EXC_PGM|EXC_USER:
322		/*
323		 * Illegal insn:
324		 *
325		 * let's try to see if it's FPU and can be emulated.
326		 */
327		curcpu()->ci_data.cpu_ntrap++;
328		pcb = lwp_getpcb(l);
329
330		if (!(pcb->pcb_flags & PCB_FPU)) {
331			memset(&pcb->pcb_fpu, 0, sizeof(pcb->pcb_fpu));
332			pcb->pcb_flags |= PCB_FPU;
333		}
334
335		if ((rv = fpu_emulate(frame, (struct fpreg *)&pcb->pcb_fpu))) {
336			KSI_INIT_TRAP(&ksi);
337			ksi.ksi_signo = rv;
338			ksi.ksi_trap = EXC_PGM;
339			ksi.ksi_addr = (void *)frame->srr0;
340			trapsignal(l, &ksi);
341		}
342		break;
343
344	case EXC_MCHK:
345		{
346			struct faultbuf *fb;
347
348			pcb = lwp_getpcb(l);
349			if ((fb = pcb->pcb_onfault) != NULL) {
350				frame->tf_xtra[TF_PID] = KERNEL_PID;
351				frame->srr0 = fb->fb_pc;
352				frame->srr1 |= PSL_IR; /* Re-enable IMMU */
353				frame->fixreg[1] = fb->fb_sp;
354				frame->fixreg[2] = fb->fb_r2;
355				frame->fixreg[3] = EFAULT;
356				frame->cr = fb->fb_cr;
357				memcpy(&frame->fixreg[13], fb->fb_fixreg,
358				    sizeof(fb->fb_fixreg));
359				goto done;
360			}
361		}
362		goto brain_damage;
363	default:
364 brain_damage:
365		printf("trap type 0x%x at 0x%lx\n", type, frame->srr0);
366#if defined(DDB) || defined(KGDB)
367		if (kdb_trap(type, frame))
368			goto done;
369#endif
370#ifdef TRAP_PANICWAIT
371		printf("Press a key to panic.\n");
372		cngetc();
373#endif
374		panic("trap");
375	}
376
377	/* Invoke MI userret code */
378	mi_userret(l);
379 done:
380	return;
381}
382
383int
384ctx_setup(int ctx, int srr1)
385{
386	volatile struct pmap *pm;
387
388	/* Update PID if we're returning to user mode. */
389	if (srr1 & PSL_PR) {
390		pm = curproc->p_vmspace->vm_map.pmap;
391		if (!pm->pm_ctx) {
392			ctx_alloc(__UNVOLATILE(pm));
393		}
394		ctx = pm->pm_ctx;
395		if (srr1 & PSL_SE) {
396			int dbreg, mask = 0x48000000;
397				/*
398				 * Set the Internal Debug and
399				 * Instruction Completion bits of
400				 * the DBCR0 register.
401				 *
402				 * XXX this is also used by jtag debuggers...
403				 */
404			__asm volatile("mfspr %0,0x3f2;"
405			    "or %0,%0,%1;"
406			    "mtspr 0x3f2,%0;" :
407			    "=&r" (dbreg) : "r" (mask));
408		}
409	}
410	else if (!ctx) {
411		ctx = KERNEL_PID;
412	}
413	return (ctx);
414}
415
416/*
417 * Used by copyin()/copyout()
418 */
419extern vaddr_t vmaprange(struct proc *, vaddr_t, vsize_t, int);
420extern void vunmaprange(vaddr_t, vsize_t);
421static int bigcopyin(const void *, void *, size_t );
422static int bigcopyout(const void *, void *, size_t );
423
424int
425copyin(const void *udaddr, void *kaddr, size_t len)
426{
427	struct pmap *pm = curproc->p_vmspace->vm_map.pmap;
428	int rv, msr, pid, tmp, ctx, count = 0;
429	struct faultbuf env;
430
431	/* For bigger buffers use the faster copy */
432	if (len > 1024)
433		return (bigcopyin(udaddr, kaddr, len));
434
435	if ((rv = setfault(&env))) {
436		curpcb->pcb_onfault = NULL;
437		return rv;
438	}
439
440	if (!(ctx = pm->pm_ctx)) {
441		/* No context -- assign it one */
442		ctx_alloc(pm);
443		ctx = pm->pm_ctx;
444	}
445
446	__asm volatile(
447		"   mfmsr %[msr];"          /* Save MSR */
448		"   li %[pid],0x20; "
449		"   andc %[pid],%[msr],%[pid]; mtmsr %[pid];"   /* Disable IMMU */
450		"   mfpid %[pid];"          /* Save old PID */
451		"   sync; isync;"
452
453		"   srwi. %[count],%[len],0x2;"     /* How many words? */
454		"   beq-  2f;"              /* No words. Go do bytes */
455		"   mtctr %[count];"
456		"1: mtpid %[ctx]; sync;"
457		"   lswi %[tmp],%[udaddr],4;"       /* Load user word */
458		"   addi %[udaddr],%[udaddr],0x4;"  /* next udaddr word */
459		"   sync; isync;"
460		"   mtpid %[pid];sync;"
461		"   stswi %[tmp],%[kaddr],4;"        /* Store kernel word */
462		"   dcbf 0,%[kaddr];"           /* flush cache */
463		"   addi %[kaddr],%[kaddr],0x4;"    /* next udaddr word */
464		"   sync; isync;"
465		"   bdnz 1b;"               /* repeat */
466
467		"2: andi. %[count],%[len],0x3;"     /* How many remaining bytes? */
468		"   addi %[count],%[count],0x1;"
469		"   mtctr %[count];"
470		"3: bdz 10f;"               /* while count */
471		"   mtpid %[ctx];sync;"
472		"   lbz %[tmp],0(%[udaddr]);"       /* Load user byte */
473		"   addi %[udaddr],%[udaddr],0x1;"  /* next udaddr byte */
474		"   sync; isync;"
475		"   mtpid %[pid]; sync;"
476		"   stb %[tmp],0(%[kaddr]);"        /* Store kernel byte */
477		"   dcbf 0,%[kaddr];"           /* flush cache */
478		"   addi %[kaddr],%[kaddr],0x1;"
479		"   sync; isync;"
480		"   b 3b;"
481		"10:mtpid %[pid]; mtmsr %[msr]; sync; isync;" /* Restore PID and MSR */
482		: [msr] "=&r" (msr), [pid] "=&r" (pid), [tmp] "=&r" (tmp)
483		: [udaddr] "b" (udaddr), [ctx] "b" (ctx), [kaddr] "b" (kaddr), [len] "b" (len), [count] "b" (count));
484
485	curpcb->pcb_onfault = NULL;
486	return 0;
487}
488
489static int
490bigcopyin(const void *udaddr, void *kaddr, size_t len)
491{
492	const char *up;
493	char *kp = kaddr;
494	struct lwp *l = curlwp;
495	struct proc *p;
496	struct faultbuf env;
497	int error;
498
499	p = l->l_proc;
500
501	/*
502	 * Stolen from physio():
503	 */
504	error = uvm_vslock(p->p_vmspace, __UNCONST(udaddr), len, VM_PROT_READ);
505	if (error) {
506		return error;
507	}
508	up = (char *)vmaprange(p, (vaddr_t)udaddr, len, VM_PROT_READ);
509
510	if ((error = setfault(&env)) == 0) {
511		memcpy(kp, up, len);
512	}
513
514	curpcb->pcb_onfault = NULL;
515	vunmaprange((vaddr_t)up, len);
516	uvm_vsunlock(p->p_vmspace, __UNCONST(udaddr), len);
517
518	return error;
519}
520
521int
522copyout(const void *kaddr, void *udaddr, size_t len)
523{
524	struct pmap *pm = curproc->p_vmspace->vm_map.pmap;
525	int rv, msr, pid, tmp, ctx, count = 0;
526	struct faultbuf env;
527
528	/* For big copies use more efficient routine */
529	if (len > 1024)
530		return (bigcopyout(kaddr, udaddr, len));
531
532	if ((rv = setfault(&env))) {
533		curpcb->pcb_onfault = NULL;
534		return rv;
535	}
536
537	if (!(ctx = pm->pm_ctx)) {
538		/* No context -- assign it one */
539		ctx_alloc(pm);
540		ctx = pm->pm_ctx;
541	}
542
543	__asm volatile(
544		"   mfmsr %[msr];"          /* Save MSR */ \
545		"   li %[pid],0x20; " \
546		"   andc %[pid],%[msr],%[pid]; mtmsr %[pid];"   /* Disable IMMU */ \
547		"   mfpid %[pid];"          /* Save old PID */ \
548		"   sync; isync;"
549
550		"   srwi. %[count],%[len],0x2;"     /* How many words? */
551		"   beq-  2f;"              /* No words. Go do bytes */
552		"   mtctr %[count];"
553		"1: mtpid %[pid];sync;"
554		"   lswi %[tmp],%[kaddr],4;"        /* Load kernel word */
555		"   addi %[kaddr],%[kaddr],0x4;"    /* next kaddr word */
556		"   sync; isync;"
557		"   mtpid %[ctx]; sync;"
558		"   stswi %[tmp],%[udaddr],4;"       /* Store user word */
559		"   dcbf 0,%[udaddr];"          /* flush cache */
560		"   addi %[udaddr],%[udaddr],0x4;"  /* next udaddr word */
561		"   sync; isync;"
562		"   bdnz 1b;"               /* repeat */
563
564		"2: andi. %[count],%[len],0x3;"     /* How many remaining bytes? */
565		"   addi %[count],%[count],0x1;"
566		"   mtctr %[count];"
567		"3: bdz  10f;"              /* while count */
568		"   mtpid %[pid];sync;"
569		"   lbz %[tmp],0(%[kaddr]);"        /* Load kernel byte */
570		"   addi %[kaddr],%[kaddr],0x1;"    /* next kaddr byte */
571		"   sync; isync;"
572		"   mtpid %[ctx]; sync;"
573		"   stb %[tmp],0(%[udaddr]);"       /* Store user byte */
574		"   dcbf 0,%[udaddr];"          /* flush cache */
575		"   addi %[udaddr],%[udaddr],0x1;"
576		"   sync; isync;"
577		"   b 3b;"
578		"10:mtpid %[pid]; mtmsr %[msr]; sync; isync;" /* Restore PID and MSR */
579		: [msr] "=&r" (msr), [pid] "=&r" (pid), [tmp] "=&r" (tmp)
580		: [udaddr] "b" (udaddr), [ctx] "b" (ctx), [kaddr] "b" (kaddr), [len] "b" (len), [count] "b" (count));
581
582	curpcb->pcb_onfault = NULL;
583	return 0;
584}
585
586static int
587bigcopyout(const void *kaddr, void *udaddr, size_t len)
588{
589	char *up;
590	const char *kp = (const char *)kaddr;
591	struct lwp *l = curlwp;
592	struct proc *p;
593	struct faultbuf env;
594	int error;
595
596	p = l->l_proc;
597
598	/*
599	 * Stolen from physio():
600	 */
601	error = uvm_vslock(p->p_vmspace, udaddr, len, VM_PROT_WRITE);
602	if (error) {
603		return error;
604	}
605	up = (char *)vmaprange(p, (vaddr_t)udaddr, len,
606	    VM_PROT_READ | VM_PROT_WRITE);
607
608	if ((error = setfault(&env)) == 0) {
609		memcpy(up, kp, len);
610	}
611
612	curpcb->pcb_onfault = NULL;
613	vunmaprange((vaddr_t)up, len);
614	uvm_vsunlock(p->p_vmspace, udaddr, len);
615
616	return error;
617}
618
619/*
620 * kcopy(const void *src, void *dst, size_t len);
621 *
622 * Copy len bytes from src to dst, aborting if we encounter a fatal
623 * page fault.
624 *
625 * kcopy() _must_ save and restore the old fault handler since it is
626 * called by uiomove(), which may be in the path of servicing a non-fatal
627 * page fault.
628 */
629int
630kcopy(const void *src, void *dst, size_t len)
631{
632	struct faultbuf env, *oldfault;
633	int rv;
634
635	oldfault = curpcb->pcb_onfault;
636	if ((rv = setfault(&env))) {
637		curpcb->pcb_onfault = oldfault;
638		return rv;
639	}
640
641	memcpy(dst, src, len);
642
643	curpcb->pcb_onfault = oldfault;
644	return 0;
645}
646
647int
648badaddr(void *addr, size_t size)
649{
650
651	return badaddr_read(addr, size, NULL);
652}
653
654int
655badaddr_read(void *addr, size_t size, int *rptr)
656{
657	struct faultbuf env;
658	int x;
659
660	/* Get rid of any stale machine checks that have been waiting.  */
661	__asm volatile ("sync; isync");
662
663	if (setfault(&env)) {
664		curpcb->pcb_onfault = NULL;
665		__asm volatile ("sync");
666		return 1;
667	}
668
669	__asm volatile ("sync");
670
671	switch (size) {
672	case 1:
673		x = *(volatile int8_t *)addr;
674		break;
675	case 2:
676		x = *(volatile int16_t *)addr;
677		break;
678	case 4:
679		x = *(volatile int32_t *)addr;
680		break;
681	default:
682		panic("badaddr: invalid size (%d)", size);
683	}
684
685	/* Make sure we took the machine check, if we caused one. */
686	__asm volatile ("sync; isync");
687
688	curpcb->pcb_onfault = NULL;
689	__asm volatile ("sync");	/* To be sure. */
690
691	/* Use the value to avoid reorder. */
692	if (rptr)
693		*rptr = x;
694
695	return 0;
696}
697
698/*
699 * For now, this only deals with the particular unaligned access case
700 * that gcc tends to generate.  Eventually it should handle all of the
701 * possibilities that can happen on a 32-bit PowerPC in big-endian mode.
702 */
703
704static int
705fix_unaligned(struct lwp *l, struct trapframe *frame)
706{
707
708	return -1;
709}
710
711/*
712 * Start a new LWP
713 */
714void
715startlwp(void *arg)
716{
717	ucontext_t *uc = arg;
718	lwp_t *l = curlwp;
719	int error;
720
721	error = cpu_setmcontext(l, &uc->uc_mcontext, uc->uc_flags);
722	KASSERT(error == 0);
723
724	kmem_free(uc, sizeof(ucontext_t));
725	upcallret(l);
726}
727
728/*
729 * XXX This is a terrible name.
730 */
731void
732upcallret(struct lwp *l)
733{
734
735	/* Invoke MI userret code */
736	mi_userret(l);
737}
738