xref: /illumos-gate/usr/src/uts/i86pc/os/trap.c (revision 54719d5edb7521bb3a321ad8bd9ebc251d61a1a0)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 
23 /*
24  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
25  * Use is subject to license terms.
26  */
27 
28 /*	Copyright (c) 1990, 1991 UNIX System Laboratories, Inc. */
29 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989, 1990 AT&T   */
30 /*		All Rights Reserved   				*/
31 /*								*/
32 /*	Copyright (c) 1987, 1988 Microsoft Corporation  	*/
33 /*		All Rights Reserved   				*/
34 /*								*/
35 
36 #pragma ident	"%Z%%M%	%I%	%E% SMI"
37 
38 #include <sys/types.h>
39 #include <sys/sysmacros.h>
40 #include <sys/param.h>
41 #include <sys/signal.h>
42 #include <sys/systm.h>
43 #include <sys/user.h>
44 #include <sys/proc.h>
45 #include <sys/disp.h>
46 #include <sys/class.h>
47 #include <sys/core.h>
48 #include <sys/syscall.h>
49 #include <sys/cpuvar.h>
50 #include <sys/vm.h>
51 #include <sys/sysinfo.h>
52 #include <sys/fault.h>
53 #include <sys/stack.h>
54 #include <sys/mmu.h>
55 #include <sys/psw.h>
56 #include <sys/regset.h>
57 #include <sys/fp.h>
58 #include <sys/trap.h>
59 #include <sys/kmem.h>
60 #include <sys/vtrace.h>
61 #include <sys/cmn_err.h>
62 #include <sys/prsystm.h>
63 #include <sys/mutex_impl.h>
64 #include <sys/machsystm.h>
65 #include <sys/archsystm.h>
66 #include <sys/sdt.h>
67 #include <sys/avintr.h>
68 #include <sys/kobj.h>
69 
70 #include <vm/hat.h>
71 
72 #include <vm/seg_kmem.h>
73 #include <vm/as.h>
74 #include <vm/seg.h>
75 #include <vm/hat_pte.h>
76 
77 #include <sys/procfs.h>
78 
79 #include <sys/reboot.h>
80 #include <sys/debug.h>
81 #include <sys/debugreg.h>
82 #include <sys/modctl.h>
83 #include <sys/aio_impl.h>
84 #include <sys/tnf.h>
85 #include <sys/tnf_probe.h>
86 #include <sys/cred.h>
87 #include <sys/mman.h>
88 #include <sys/x86_archext.h>
89 #include <sys/copyops.h>
90 #include <c2/audit.h>
91 #include <sys/ftrace.h>
92 #include <sys/panic.h>
93 #include <sys/traptrace.h>
94 #include <sys/ontrap.h>
95 #include <sys/cpc_impl.h>
96 
97 #define	USER	0x10000		/* user-mode flag added to trap type */
98 
99 static const char *trap_type_mnemonic[] = {
100 	"de",	"db",	"2",	"bp",
101 	"of",	"br",	"ud",	"nm",
102 	"df",	"9",	"ts",	"np",
103 	"ss",	"gp",	"pf",	"15",
104 	"mf",	"ac",	"mc",	"xf"
105 };
106 
107 static const char *trap_type[] = {
108 	"Divide error",				/* trap id 0 	*/
109 	"Debug",				/* trap id 1	*/
110 	"NMI interrupt",			/* trap id 2	*/
111 	"Breakpoint",				/* trap id 3 	*/
112 	"Overflow",				/* trap id 4 	*/
113 	"BOUND range exceeded",			/* trap id 5 	*/
114 	"Invalid opcode",			/* trap id 6 	*/
115 	"Device not available",			/* trap id 7 	*/
116 	"Double fault",				/* trap id 8 	*/
117 	"Coprocessor segment overrun",		/* trap id 9 	*/
118 	"Invalid TSS",				/* trap id 10 	*/
119 	"Segment not present",			/* trap id 11 	*/
120 	"Stack segment fault",			/* trap id 12 	*/
121 	"General protection",			/* trap id 13 	*/
122 	"Page fault",				/* trap id 14 	*/
123 	"Reserved",				/* trap id 15 	*/
124 	"x87 floating point error",		/* trap id 16 	*/
125 	"Alignment check",			/* trap id 17 	*/
126 	"Machine check",			/* trap id 18	*/
127 	"SIMD floating point exception",	/* trap id 19	*/
128 };
129 
130 #define	TRAP_TYPES	(sizeof (trap_type) / sizeof (trap_type[0]))
131 
132 int tudebug = 0;
133 int tudebugbpt = 0;
134 int tudebugfpe = 0;
135 int tudebugsse = 0;
136 
137 #if defined(TRAPDEBUG) || defined(lint)
138 int tdebug = 0;
139 int lodebug = 0;
140 int faultdebug = 0;
141 #else
142 #define	tdebug	0
143 #define	lodebug	0
144 #define	faultdebug	0
145 #endif /* defined(TRAPDEBUG) || defined(lint) */
146 
147 #if defined(TRAPTRACE)
148 static void dump_ttrace(void);
149 #endif	/* TRAPTRACE */
150 static void dumpregs(struct regs *);
151 static void showregs(uint_t, struct regs *, caddr_t);
152 static void dump_tss(void);
153 static int kern_gpfault(struct regs *);
154 
155 struct trap_info {
156 	struct regs *trap_regs;
157 	uint_t trap_type;
158 	caddr_t trap_addr;
159 };
160 
161 /*ARGSUSED*/
162 static int
163 die(uint_t type, struct regs *rp, caddr_t addr, processorid_t cpuid)
164 {
165 	struct trap_info ti;
166 	const char *trap_name, *trap_mnemonic;
167 
168 	if (type < TRAP_TYPES) {
169 		trap_name = trap_type[type];
170 		trap_mnemonic = trap_type_mnemonic[type];
171 	} else {
172 		trap_name = "trap";
173 		trap_mnemonic = "-";
174 	}
175 
176 #ifdef TRAPTRACE
177 	TRAPTRACE_FREEZE;
178 #endif
179 
180 	ti.trap_regs = rp;
181 	ti.trap_type = type & ~USER;
182 	ti.trap_addr = addr;
183 
184 	curthread->t_panic_trap = &ti;
185 
186 	if (type == T_PGFLT && addr < (caddr_t)KERNELBASE) {
187 		panic("BAD TRAP: type=%x (#%s %s) rp=%p addr=%p "
188 		    "occurred in module \"%s\" due to %s",
189 		    type, trap_mnemonic, trap_name, (void *)rp, (void *)addr,
190 		    mod_containing_pc((caddr_t)rp->r_pc),
191 		    addr < (caddr_t)PAGESIZE ?
192 		    "a NULL pointer dereference" :
193 		    "an illegal access to a user address");
194 	} else
195 		panic("BAD TRAP: type=%x (#%s %s) rp=%p addr=%p",
196 		    type, trap_mnemonic, trap_name, (void *)rp, (void *)addr);
197 	return (0);
198 }
199 
200 /*
201  * Rewrite the instruction at pc to be an int $T_SYSCALLINT instruction.
202  *
203  * int <vector> is two bytes: 0xCD <vector>
204  */
205 
206 #define	SLOW_SCALL_SIZE	2
207 
208 static int
209 rewrite_syscall(caddr_t pc)
210 {
211 	uchar_t instr[SLOW_SCALL_SIZE] = { 0xCD, T_SYSCALLINT };
212 
213 	if (uwrite(curthread->t_procp, instr, SLOW_SCALL_SIZE,
214 	    (uintptr_t)pc) != 0)
215 		return (1);
216 
217 	return (0);
218 }
219 
220 /*
221  * Test to see if the instruction at pc is sysenter or syscall. The second
222  * argument should be the x86 feature flag corresponding to the expected
223  * instruction.
224  *
225  * sysenter is two bytes: 0x0F 0x34
226  * syscall is two bytes:  0x0F 0x05
227  */
228 
229 #define	FAST_SCALL_SIZE	2
230 
231 static int
232 instr_is_fast_syscall(caddr_t pc, int which)
233 {
234 	uchar_t instr[FAST_SCALL_SIZE];
235 
236 	ASSERT(which == X86_SEP || which == X86_ASYSC);
237 
238 	if (copyin_nowatch(pc, (caddr_t)instr, FAST_SCALL_SIZE) != 0 ||
239 	    instr[0] != 0x0F)
240 		return (0);
241 
242 	if ((which == X86_SEP && instr[1] == 0x34) ||
243 	    (which == X86_ASYSC && instr[1] == 0x05))
244 		return (1);
245 
246 	return (0);
247 }
248 
249 /*
250  * Test to see if the instruction at pc is a system call instruction.
251  *
252  * The bytes of an lcall instruction used for the syscall trap.
253  * static uchar_t lcall[7] = { 0x9a, 0, 0, 0, 0, 0x7, 0 };
254  * static uchar_t lcallalt[7] = { 0x9a, 0, 0, 0, 0, 0x27, 0 };
255  */
256 
257 #define	LCALLSIZE	7
258 
259 static int
260 instr_is_syscall(caddr_t pc)
261 {
262 	uchar_t instr[LCALLSIZE];
263 
264 	if (copyin_nowatch(pc, (caddr_t)instr, LCALLSIZE) == 0 &&
265 	    instr[0] == 0x9a &&
266 	    instr[1] == 0 &&
267 	    instr[2] == 0 &&
268 	    instr[3] == 0 &&
269 	    instr[4] == 0 &&
270 	    (instr[5] == 0x7 || instr[5] == 0x27) &&
271 	    instr[6] == 0)
272 		return (1);
273 
274 	return (0);
275 }
276 
277 #ifdef OPTERON_ERRATUM_91
278 
279 /*
280  * Test to see if the instruction at pc is a prefetch instruction.
281  *
282  * The first byte of prefetch instructions is always 0x0F.
283  * The second byte is 0x18 for regular prefetch or 0x0D for AMD 3dnow prefetch.
284  * The third byte is between 0 and 3 inclusive.
285  */
286 
287 #define	PREFETCHSIZE 3
288 
289 static int
290 cmp_to_prefetch(uchar_t *p)
291 {
292 	if (*p == 0x0F && (*(p+1) == 0x18 || *(p+1) == 0x0D) && *(p+2) <= 3)
293 		return (1);
294 	return (0);
295 }
296 
297 static int
298 instr_is_prefetch(caddr_t pc)
299 {
300 	uchar_t instr[PREFETCHSIZE];
301 	int	error;
302 
303 	error = copyin_nowatch(pc, (caddr_t)instr, PREFETCHSIZE);
304 
305 	if (error == 0 && cmp_to_prefetch(instr))
306 		return (1);
307 	return (0);
308 }
309 
310 #endif /* OPTERON_ERRATUM_91 */
311 
312 /*
313  * Called from the trap handler when a processor trap occurs.
314  *
315  * Note: All user-level traps that might call stop() must exit
316  * trap() by 'goto out' or by falling through.
317  */
318 void
319 trap(struct regs *rp, caddr_t addr, processorid_t cpuid)
320 {
321 	kthread_t *cur_thread = curthread;
322 	enum seg_rw rw;
323 	unsigned type;
324 	extern int stop_on_fault(uint_t, k_siginfo_t *);
325 	proc_t *p = ttoproc(cur_thread);
326 	klwp_t *lwp = ttolwp(cur_thread);
327 	uintptr_t lofault;
328 	faultcode_t pagefault(), res, errcode;
329 	enum fault_type fault_type;
330 	k_siginfo_t siginfo;
331 	uint_t fault = 0;
332 	int mstate;
333 	int sicode = 0;
334 	int watchcode;
335 	int watchpage;
336 	caddr_t vaddr;
337 	size_t sz;
338 	int ta;
339 
340 	ASSERT_STACK_ALIGNED();
341 
342 	type = rp->r_trapno;
343 	CPU_STATS_ADDQ(CPU, sys, trap, 1);
344 
345 	ASSERT(cur_thread->t_schedflag & TS_DONT_SWAP);
346 
347 	if (type == T_PGFLT) {
348 
349 		errcode = rp->r_err;
350 		if (errcode & PF_ERR_WRITE)
351 			rw = S_WRITE;
352 		else if ((caddr_t)rp->r_pc == addr ||
353 		    (mmu.pt_nx != 0 && (errcode & PF_ERR_EXEC)))
354 			rw = S_EXEC;
355 		else
356 			rw = S_READ;
357 
358 #if defined(__i386)
359 		/*
360 		 * Pentium Pro work-around
361 		 */
362 		if ((errcode & PF_ERR_PROT) && pentiumpro_bug4046376) {
363 			uint_t	attr;
364 			uint_t	priv_violation;
365 			uint_t	access_violation;
366 
367 			if (hat_getattr(addr < (caddr_t)kernelbase ?
368 			    curproc->p_as->a_hat : kas.a_hat, addr, &attr)
369 			    == -1) {
370 				errcode &= ~PF_ERR_PROT;
371 			} else {
372 				priv_violation = (errcode & PF_ERR_USER) &&
373 					!(attr & PROT_USER);
374 				access_violation = (errcode & PF_ERR_WRITE) &&
375 					!(attr & PROT_WRITE);
376 				if (!priv_violation && !access_violation)
377 					goto cleanup;
378 			}
379 		}
380 #endif /* __i386 */
381 
382 	}
383 
384 	if (tdebug)
385 		showregs(type, rp, addr);
386 
387 	if (USERMODE(rp->r_cs)) {
388 		/*
389 		 * Set up the current cred to use during this trap. u_cred
390 		 * no longer exists.  t_cred is used instead.
391 		 * The current process credential applies to the thread for
392 		 * the entire trap.  If trapping from the kernel, this
393 		 * should already be set up.
394 		 */
395 		if (cur_thread->t_cred != p->p_cred) {
396 			cred_t *oldcred = cur_thread->t_cred;
397 			/*
398 			 * DTrace accesses t_cred in probe context.  t_cred
399 			 * must always be either NULL, or point to a valid,
400 			 * allocated cred structure.
401 			 */
402 			cur_thread->t_cred = crgetcred();
403 			crfree(oldcred);
404 		}
405 		ASSERT(lwp != NULL);
406 		type |= USER;
407 		ASSERT(lwptoregs(lwp) == rp);
408 		lwp->lwp_state = LWP_SYS;
409 
410 		switch (type) {
411 		case T_PGFLT + USER:
412 			if ((caddr_t)rp->r_pc == addr)
413 				mstate = LMS_TFAULT;
414 			else
415 				mstate = LMS_DFAULT;
416 			break;
417 		default:
418 			mstate = LMS_TRAP;
419 			break;
420 		}
421 		/* Kernel probe */
422 		TNF_PROBE_1(thread_state, "thread", /* CSTYLED */,
423 		    tnf_microstate, state, mstate);
424 		mstate = new_mstate(cur_thread, mstate);
425 
426 		bzero(&siginfo, sizeof (siginfo));
427 	}
428 
429 	switch (type) {
430 	case T_PGFLT + USER:
431 	case T_SGLSTP:
432 	case T_SGLSTP + USER:
433 	case T_BPTFLT + USER:
434 		break;
435 
436 	default:
437 		FTRACE_2("trap(): type=0x%lx, regs=0x%lx",
438 		    (ulong_t)type, (ulong_t)rp);
439 		break;
440 	}
441 
442 	switch (type) {
443 
444 	case T_MCE:	/* Machine check exception */
445 	case T_MCE + USER:
446 		if (x86_feature & X86_MCA) {
447 			if (mca_exception(rp))
448 				(void) die(type, rp, addr, cpuid);
449 			type &= ~USER;
450 			goto cleanup;
451 		}
452 	default:
453 		if (type & USER) {
454 			if (tudebug)
455 				showregs(type, rp, (caddr_t)0);
456 			printf("trap: Unknown trap type %d in user mode\n",
457 			    type & ~USER);
458 			siginfo.si_signo = SIGILL;
459 			siginfo.si_code  = ILL_ILLTRP;
460 			siginfo.si_addr  = (caddr_t)rp->r_pc;
461 			siginfo.si_trapno = type & ~USER;
462 			fault = FLTILL;
463 			break;
464 		} else {
465 			(void) die(type, rp, addr, cpuid);
466 			/*NOTREACHED*/
467 		}
468 
469 	case T_PGFLT:		/* system page fault */
470 		/*
471 		 * If we're under on_trap() protection (see <sys/ontrap.h>),
472 		 * set ot_trap and longjmp back to the on_trap() call site.
473 		 */
474 		if ((cur_thread->t_ontrap != NULL) &&
475 		    (cur_thread->t_ontrap->ot_prot & OT_DATA_ACCESS)) {
476 			curthread->t_ontrap->ot_trap |= OT_DATA_ACCESS;
477 			longjmp(&curthread->t_ontrap->ot_jmpbuf);
478 		}
479 
480 		/*
481 		 * See if we can handle as pagefault. Save lofault
482 		 * across this. Here we assume that an address
483 		 * less than KERNELBASE is a user fault.
484 		 * We can do this as copy.s routines verify that the
485 		 * starting address is less than KERNELBASE before
486 		 * starting and because we know that we always have
487 		 * KERNELBASE mapped as invalid to serve as a "barrier".
488 		 */
489 		lofault = cur_thread->t_lofault;
490 		cur_thread->t_lofault = 0;
491 
492 		mstate = new_mstate(cur_thread, LMS_KFAULT);
493 
494 		if (addr < (caddr_t)kernelbase) {
495 			res = pagefault(addr,
496 			    (errcode & PF_ERR_PROT)? F_PROT: F_INVAL, rw, 0);
497 			if (res == FC_NOMAP &&
498 			    addr < p->p_usrstack &&
499 			    grow(addr))
500 				res = 0;
501 		} else {
502 			res = pagefault(addr,
503 			    (errcode & PF_ERR_PROT)? F_PROT: F_INVAL, rw, 1);
504 		}
505 		(void) new_mstate(cur_thread, mstate);
506 
507 		/*
508 		 * Restore lofault. If we resolved the fault, exit.
509 		 * If we didn't and lofault wasn't set, die.
510 		 */
511 		cur_thread->t_lofault = lofault;
512 		if (res == 0)
513 			goto cleanup;
514 
515 #if defined(OPTERON_ERRATUM_93) && defined(_LP64)
516 		if (lofault == 0 && opteron_erratum_93) {
517 			/*
518 			 * Workaround for Opteron Erratum 93. On return from
519 			 * a System Managment Interrupt at a HLT instruction
520 			 * the %rip might be truncated to a 32 bit value.
521 			 * BIOS is supposed to fix this, but some don't.
522 			 * If this occurs we simply restore the high order bits.
523 			 * The HLT instruction is 1 byte of 0xf4.
524 			 */
525 			uintptr_t	rip = rp->r_pc;
526 
527 			if ((rip & 0xfffffffful) == rip) {
528 				rip |= 0xfffffffful << 32;
529 				if (hat_getpfnum(kas.a_hat, (caddr_t)rip) !=
530 				    PFN_INVALID &&
531 				    (*(uchar_t *)rip == 0xf4 ||
532 				    *(uchar_t *)(rip - 1) == 0xf4)) {
533 					rp->r_pc = rip;
534 					goto cleanup;
535 				}
536 			}
537 		}
538 #endif /* OPTERON_ERRATUM_93 && _LP64 */
539 
540 #ifdef OPTERON_ERRATUM_91
541 		if (lofault == 0 && opteron_erratum_91) {
542 			/*
543 			 * Workaround for Opteron Erratum 91. Prefetches may
544 			 * generate a page fault (they're not supposed to do
545 			 * that!). If this occurs we simply return back to the
546 			 * instruction.
547 			 */
548 			caddr_t		pc = (caddr_t)rp->r_pc;
549 
550 			/*
551 			 * If the faulting PC is not mapped, this is a
552 			 * legitimate kernel page fault that must result in a
553 			 * panic. If the faulting PC is mapped, it could contain
554 			 * a prefetch instruction. Check for that here.
555 			 */
556 			if (hat_getpfnum(kas.a_hat, pc) != PFN_INVALID) {
557 				if (cmp_to_prefetch((uchar_t *)pc)) {
558 #ifdef DEBUG
559 					cmn_err(CE_WARN, "Opteron erratum 91 "
560 					    "occurred: kernel prefetch"
561 					    " at %p generated a page fault!",
562 					    (void *)rp->r_pc);
563 #endif /* DEBUG */
564 					goto cleanup;
565 				}
566 			}
567 			(void) die(type, rp, addr, cpuid);
568 		}
569 #endif /* OPTERON_ERRATUM_91 */
570 
571 		if (lofault == 0)
572 			(void) die(type, rp, addr, cpuid);
573 
574 		/*
575 		 * Cannot resolve fault.  Return to lofault.
576 		 */
577 		if (lodebug) {
578 			showregs(type, rp, addr);
579 			traceregs(rp);
580 		}
581 		if (FC_CODE(res) == FC_OBJERR)
582 			res = FC_ERRNO(res);
583 		else
584 			res = EFAULT;
585 		rp->r_r0 = res;
586 		rp->r_pc = cur_thread->t_lofault;
587 		goto cleanup;
588 
589 	case T_PGFLT + USER:	/* user page fault */
590 		if (faultdebug) {
591 			char *fault_str;
592 
593 			switch (rw) {
594 			case S_READ:
595 				fault_str = "read";
596 				break;
597 			case S_WRITE:
598 				fault_str = "write";
599 				break;
600 			case S_EXEC:
601 				fault_str = "exec";
602 				break;
603 			default:
604 				fault_str = "";
605 				break;
606 			}
607 			printf("user %s fault:  addr=0x%lx errcode=0x%x\n",
608 			    fault_str, (uintptr_t)addr, errcode);
609 		}
610 
611 #if defined(OPTERON_ERRATUM_100) && defined(_LP64)
612 		/*
613 		 * Workaround for AMD erratum 100
614 		 *
615 		 * A 32-bit process may receive a page fault on a non
616 		 * 32-bit address by mistake. The range of the faulting
617 		 * address will be
618 		 *
619 		 *	0xffffffff80000000 .. 0xffffffffffffffff or
620 		 *	0x0000000100000000 .. 0x000000017fffffff
621 		 *
622 		 * The fault is always due to an instruction fetch, however
623 		 * the value of r_pc should be correct (in 32 bit range),
624 		 * so we ignore the page fault on the bogus address.
625 		 */
626 		if (p->p_model == DATAMODEL_ILP32 &&
627 		    (0xffffffff80000000 <= (uintptr_t)addr ||
628 		    (0x100000000 <= (uintptr_t)addr &&
629 		    (uintptr_t)addr <= 0x17fffffff))) {
630 			if (!opteron_erratum_100)
631 				panic("unexpected erratum #100");
632 			if (rp->r_pc <= 0xffffffff)
633 				goto out;
634 		}
635 #endif /* OPTERON_ERRATUM_100 && _LP64 */
636 
637 		ASSERT(!(curthread->t_flag & T_WATCHPT));
638 		watchpage = (pr_watch_active(p) && pr_is_watchpage(addr, rw));
639 #ifdef __i386
640 		/*
641 		 * In 32-bit mode, the lcall (system call) instruction fetches
642 		 * one word from the stack, at the stack pointer, because of the
643 		 * way the call gate is constructed.  This is a bogus
644 		 * read and should not be counted as a read watchpoint.
645 		 * We work around the problem here by testing to see if
646 		 * this situation applies and, if so, simply jumping to
647 		 * the code in locore.s that fields the system call trap.
648 		 * The registers on the stack are already set up properly
649 		 * due to the match between the call gate sequence and the
650 		 * trap gate sequence.  We just have to adjust the pc.
651 		 */
652 		if (watchpage && addr == (caddr_t)rp->r_sp &&
653 		    rw == S_READ && instr_is_syscall((caddr_t)rp->r_pc)) {
654 			extern void watch_syscall(void);
655 
656 			rp->r_pc += LCALLSIZE;
657 			watch_syscall();	/* never returns */
658 			/* NOTREACHED */
659 		}
660 #endif /* __i386 */
661 		vaddr = addr;
662 		if (!watchpage || (sz = instr_size(rp, &vaddr, rw)) <= 0)
663 			fault_type = (errcode & PF_ERR_PROT)? F_PROT: F_INVAL;
664 		else if ((watchcode = pr_is_watchpoint(&vaddr, &ta,
665 		    sz, NULL, rw)) != 0) {
666 			if (ta) {
667 				do_watch_step(vaddr, sz, rw,
668 					watchcode, rp->r_pc);
669 				fault_type = F_INVAL;
670 			} else {
671 				bzero(&siginfo, sizeof (siginfo));
672 				siginfo.si_signo = SIGTRAP;
673 				siginfo.si_code = watchcode;
674 				siginfo.si_addr = vaddr;
675 				siginfo.si_trapafter = 0;
676 				siginfo.si_pc = (caddr_t)rp->r_pc;
677 				fault = FLTWATCH;
678 				break;
679 			}
680 		} else {
681 			/* XXX pr_watch_emul() never succeeds (for now) */
682 			if (rw != S_EXEC && pr_watch_emul(rp, vaddr, rw))
683 				goto out;
684 			do_watch_step(vaddr, sz, rw, 0, 0);
685 			fault_type = F_INVAL;
686 		}
687 
688 		res = pagefault(addr, fault_type, rw, 0);
689 
690 		/*
691 		 * If pagefault() succeeded, ok.
692 		 * Otherwise attempt to grow the stack.
693 		 */
694 		if (res == 0 ||
695 		    (res == FC_NOMAP &&
696 		    addr < p->p_usrstack &&
697 		    grow(addr))) {
698 			lwp->lwp_lastfault = FLTPAGE;
699 			lwp->lwp_lastfaddr = addr;
700 			if (prismember(&p->p_fltmask, FLTPAGE)) {
701 				bzero(&siginfo, sizeof (siginfo));
702 				siginfo.si_addr = addr;
703 				(void) stop_on_fault(FLTPAGE, &siginfo);
704 			}
705 			goto out;
706 		} else if (res == FC_PROT && addr < p->p_usrstack &&
707 		    (mmu.pt_nx != 0 && (errcode & PF_ERR_EXEC))) {
708 			report_stack_exec(p, addr);
709 		}
710 
711 #ifdef OPTERON_ERRATUM_91
712 		/*
713 		 * Workaround for Opteron Erratum 91. Prefetches may generate a
714 		 * page fault (they're not supposed to do that!). If this
715 		 * occurs we simply return back to the instruction.
716 		 *
717 		 * We rely on copyin to properly fault in the page with r_pc.
718 		 */
719 		if (opteron_erratum_91 &&
720 		    addr != (caddr_t)rp->r_pc &&
721 		    instr_is_prefetch((caddr_t)rp->r_pc)) {
722 #ifdef DEBUG
723 			cmn_err(CE_WARN, "Opteron erratum 91 occurred: "
724 			    "prefetch at %p in pid %d generated a trap!",
725 			    (void *)rp->r_pc, p->p_pid);
726 #endif /* DEBUG */
727 			goto out;
728 		}
729 #endif /* OPTERON_ERRATUM_91 */
730 
731 		if (tudebug)
732 			showregs(type, rp, addr);
733 		/*
734 		 * In the case where both pagefault and grow fail,
735 		 * set the code to the value provided by pagefault.
736 		 * We map all errors returned from pagefault() to SIGSEGV.
737 		 */
738 		bzero(&siginfo, sizeof (siginfo));
739 		siginfo.si_addr = addr;
740 		switch (FC_CODE(res)) {
741 		case FC_HWERR:
742 		case FC_NOSUPPORT:
743 			siginfo.si_signo = SIGBUS;
744 			siginfo.si_code = BUS_ADRERR;
745 			fault = FLTACCESS;
746 			break;
747 		case FC_ALIGN:
748 			siginfo.si_signo = SIGBUS;
749 			siginfo.si_code = BUS_ADRALN;
750 			fault = FLTACCESS;
751 			break;
752 		case FC_OBJERR:
753 			if ((siginfo.si_errno = FC_ERRNO(res)) != EINTR) {
754 				siginfo.si_signo = SIGBUS;
755 				siginfo.si_code = BUS_OBJERR;
756 				fault = FLTACCESS;
757 			}
758 			break;
759 		default:	/* FC_NOMAP or FC_PROT */
760 			siginfo.si_signo = SIGSEGV;
761 			siginfo.si_code =
762 			    (res == FC_NOMAP)? SEGV_MAPERR : SEGV_ACCERR;
763 			fault = FLTBOUNDS;
764 			break;
765 		}
766 		break;
767 
768 	case T_ILLINST + USER:	/* invalid opcode fault */
769 		/*
770 		 * If the syscall instruction is disabled due to LDT usage, a
771 		 * user program that attempts to execute it will trigger a #ud
772 		 * trap. Check for that case here. If this occurs on a CPU which
773 		 * doesn't even support syscall, the result of all of this will
774 		 * be to emulate that particular instruction.
775 		 */
776 		if (p->p_ldt != NULL &&
777 		    instr_is_fast_syscall((caddr_t)rp->r_pc, X86_ASYSC)) {
778 			if (rewrite_syscall((caddr_t)rp->r_pc) == 0)
779 				goto out;
780 #ifdef DEBUG
781 			else
782 				cmn_err(CE_WARN, "failed to rewrite syscall "
783 				    "instruction in process %d",
784 				    curthread->t_procp->p_pid);
785 #endif /* DEBUG */
786 		}
787 		/*FALLTHROUGH*/
788 
789 		if (tudebug)
790 			showregs(type, rp, (caddr_t)0);
791 		siginfo.si_signo = SIGILL;
792 		siginfo.si_code  = ILL_ILLOPC;
793 		siginfo.si_addr  = (caddr_t)rp->r_pc;
794 		fault = FLTILL;
795 		break;
796 
797 	case T_ZERODIV + USER:		/* integer divide by zero */
798 		if (tudebug && tudebugfpe)
799 			showregs(type, rp, (caddr_t)0);
800 		siginfo.si_signo = SIGFPE;
801 		siginfo.si_code  = FPE_INTDIV;
802 		siginfo.si_addr  = (caddr_t)rp->r_pc;
803 		fault = FLTIZDIV;
804 		break;
805 
806 	case T_OVFLW + USER:	/* integer overflow */
807 		if (tudebug && tudebugfpe)
808 			showregs(type, rp, (caddr_t)0);
809 		siginfo.si_signo = SIGFPE;
810 		siginfo.si_code  = FPE_INTOVF;
811 		siginfo.si_addr  = (caddr_t)rp->r_pc;
812 		fault = FLTIOVF;
813 		break;
814 
815 	case T_NOEXTFLT + USER:	/* math coprocessor not available */
816 		if (tudebug && tudebugfpe)
817 			showregs(type, rp, addr);
818 		if (fpnoextflt(rp)) {
819 			siginfo.si_signo = SIGFPE;
820 			siginfo.si_code  = ILL_ILLOPC;
821 			siginfo.si_addr  = (caddr_t)rp->r_pc;
822 			fault = FLTFPE;
823 		}
824 		break;
825 
826 	case T_EXTOVRFLT:	/* extension overrun fault */
827 		/* check if we took a kernel trap on behalf of user */
828 		{
829 			extern  void ndptrap_frstor(void);
830 			if (rp->r_pc != (uintptr_t)ndptrap_frstor)
831 				(void) die(type, rp, addr, cpuid);
832 			type |= USER;
833 		}
834 		/*FALLTHROUGH*/
835 	case T_EXTOVRFLT + USER:	/* extension overrun fault */
836 		if (tudebug && tudebugfpe)
837 			showregs(type, rp, addr);
838 		if (fpextovrflt(rp)) {
839 			siginfo.si_signo = SIGSEGV;
840 			siginfo.si_code  = SEGV_MAPERR;
841 			siginfo.si_addr  = (caddr_t)rp->r_pc;
842 			fault = FLTBOUNDS;
843 		}
844 		break;
845 
846 	case T_EXTERRFLT:	/* x87 floating point exception pending */
847 		/* check if we took a kernel trap on behalf of user */
848 		{
849 			extern  void ndptrap_frstor(void);
850 			if (rp->r_pc != (uintptr_t)ndptrap_frstor)
851 				(void) die(type, rp, addr, cpuid);
852 			type |= USER;
853 		}
854 		/*FALLTHROUGH*/
855 
856 	case T_EXTERRFLT + USER: /* x87 floating point exception pending */
857 		if (tudebug && tudebugfpe)
858 			showregs(type, rp, addr);
859 		if (sicode = fpexterrflt(rp)) {
860 			siginfo.si_signo = SIGFPE;
861 			siginfo.si_code  = sicode;
862 			siginfo.si_addr  = (caddr_t)rp->r_pc;
863 			fault = FLTFPE;
864 		}
865 		break;
866 
867 	case T_SIMDFPE + USER:		/* SSE and SSE2 exceptions */
868 		if (tudebug && tudebugsse)
869 			showregs(type, rp, addr);
870 		if ((x86_feature & (X86_SSE|X86_SSE2)) == 0) {
871 			/*
872 			 * There are rumours that some user instructions
873 			 * on older CPUs can cause this trap to occur; in
874 			 * which case send a SIGILL instead of a SIGFPE.
875 			 */
876 			siginfo.si_signo = SIGILL;
877 			siginfo.si_code  = ILL_ILLTRP;
878 			siginfo.si_addr  = (caddr_t)rp->r_pc;
879 			siginfo.si_trapno = type & ~USER;
880 			fault = FLTILL;
881 		} else if ((sicode = fpsimderrflt(rp)) != 0) {
882 			siginfo.si_signo = SIGFPE;
883 			siginfo.si_code = sicode;
884 			siginfo.si_addr = (caddr_t)rp->r_pc;
885 			fault = FLTFPE;
886 		}
887 		break;
888 
889 	case T_BPTFLT:	/* breakpoint trap */
890 		/*
891 		 * Kernel breakpoint traps should only happen when kmdb is
892 		 * active, and even then, it'll have interposed on the IDT, so
893 		 * control won't get here.  If it does, we've hit a breakpoint
894 		 * without the debugger, which is very strange, and very
895 		 * fatal.
896 		 */
897 		if (tudebug && tudebugbpt)
898 			showregs(type, rp, (caddr_t)0);
899 
900 		(void) die(type, rp, addr, cpuid);
901 		break;
902 
903 	case T_SGLSTP: /* single step/hw breakpoint exception */
904 		if (tudebug && tudebugbpt)
905 			showregs(type, rp, (caddr_t)0);
906 
907 		/* Now evaluate how we got here */
908 		if (lwp != NULL && (lwp->lwp_pcb.pcb_drstat & DR_SINGLESTEP)) {
909 			/*
910 			 * i386 single-steps even through lcalls which
911 			 * change the privilege level. So we take a trap at
912 			 * the first instruction in privileged mode.
913 			 *
914 			 * Set a flag to indicate that upon completion of
915 			 * the system call, deal with the single-step trap.
916 			 *
917 			 * The same thing happens for sysenter, too.
918 			 */
919 #if defined(__amd64)
920 			if (rp->r_pc == (uintptr_t)sys_sysenter) {
921 				/*
922 				 * Adjust the pc so that we don't execute the
923 				 * swapgs instruction at the head of the
924 				 * handler and completely confuse things.
925 				 */
926 				rp->r_pc = (uintptr_t)
927 				    _sys_sysenter_post_swapgs;
928 #elif defined(__i386)
929 			if (rp->r_pc == (uintptr_t)sys_call ||
930 			    rp->r_pc == (uintptr_t)sys_sysenter) {
931 #endif
932 				rp->r_ps &= ~PS_T; /* turn off trace */
933 				lwp->lwp_pcb.pcb_flags |= DEBUG_PENDING;
934 				cur_thread->t_post_sys = 1;
935 				goto cleanup;
936 			}
937 		}
938 		/* XXX - needs review on debugger interface? */
939 		if (boothowto & RB_DEBUG)
940 			debug_enter((char *)NULL);
941 		else
942 			(void) die(type, rp, addr, cpuid);
943 		break;
944 
945 	case T_NMIFLT:	/* NMI interrupt */
946 		printf("Unexpected NMI in system mode\n");
947 		goto cleanup;
948 
949 	case T_NMIFLT + USER:	/* NMI interrupt */
950 		printf("Unexpected NMI in user mode\n");
951 		break;
952 
953 	case T_GPFLT:	/* general protection violation */
954 #if defined(__amd64)
955 		/*
956 		 * On amd64, we can get a #gp from referencing addresses
957 		 * in the virtual address hole e.g. from a copyin.
958 		 */
959 
960 		/*
961 		 * If we're under on_trap() protection (see <sys/ontrap.h>),
962 		 * set ot_trap and longjmp back to the on_trap() call site.
963 		 */
964 		if ((cur_thread->t_ontrap != NULL) &&
965 		    (cur_thread->t_ontrap->ot_prot & OT_DATA_ACCESS)) {
966 			curthread->t_ontrap->ot_trap |= OT_DATA_ACCESS;
967 			longjmp(&curthread->t_ontrap->ot_jmpbuf);
968 		}
969 
970 		/*
971 		 * If we're under lofault protection (copyin etc.),
972 		 * longjmp back to lofault with an EFAULT.
973 		 */
974 		if (cur_thread->t_lofault) {
975 			/*
976 			 * Fault is not resolvable, so just return to lofault
977 			 */
978 			if (lodebug) {
979 				showregs(type, rp, addr);
980 				traceregs(rp);
981 			}
982 			rp->r_r0 = EFAULT;
983 			rp->r_pc = cur_thread->t_lofault;
984 			goto cleanup;
985 		}
986 		/*FALLTHROUGH*/
987 #endif
988 	case T_STKFLT:	/* stack fault */
989 	case T_TSSFLT:	/* invalid TSS fault */
990 	case T_SEGFLT:	/* segment not present fault */
991 		if (tudebug)
992 			showregs(type, rp, (caddr_t)0);
993 		if (kern_gpfault(rp))
994 			(void) die(type, rp, addr, cpuid);
995 		goto cleanup;
996 
997 	case T_SEGFLT + USER:	/* segment not present fault */
998 #ifdef _SYSCALL32_IMPL
999 		if (instr_is_syscall((caddr_t)rp->r_pc)) {
1000 			/*
1001 			 * System calls via the call gate come in through
1002 			 * not-present traps.
1003 			 *
1004 			 * Since this is a not-present trap, rp->r_pc points to
1005 			 * the trapping lcall instruction. We need to bump it
1006 			 * to the next insn so the app can continue on.
1007 			 */
1008 			rp->r_pc += LCALLSIZE;
1009 			lwp->lwp_regs = rp;
1010 
1011 			/*
1012 			 * Normally the microstate of the LWP is forced back to
1013 			 * LMS_USER by the syscall handlers. Emulate that
1014 			 * behavior here.
1015 			 */
1016 			mstate = LMS_USER;
1017 
1018 			dosyscall();
1019 			goto out;
1020 		}
1021 #endif /* _SYSCALL32_IMPL */
1022 		/*FALLTHROUGH*/
1023 
1024 	case T_GPFLT + USER:	/* general protection violation */
1025 		/*
1026 		 * If the current process is using a private LDT and the
1027 		 * trapping instruction is sysenter, the sysenter instruction
1028 		 * has been disabled on the CPU because it destroys segment
1029 		 * registers. If this is the case, rewrite the instruction to
1030 		 * be a safe system call and retry it. If this occurs on a CPU
1031 		 * which doesn't even support sysenter, the result of all of
1032 		 * this will be to emulate that particular instruction.
1033 		 */
1034 		if (p->p_ldt != NULL &&
1035 		    instr_is_fast_syscall((caddr_t)rp->r_pc, X86_SEP)) {
1036 			if (rewrite_syscall((caddr_t)rp->r_pc) == 0)
1037 				goto out;
1038 #ifdef DEBUG
1039 			else
1040 				cmn_err(CE_WARN, "failed to rewrite sysenter "
1041 				    "instruction in process %d",
1042 				    curthread->t_procp->p_pid);
1043 #endif /* DEBUG */
1044 		}
1045 		/*FALLTHROUGH*/
1046 
1047 	case T_BOUNDFLT + USER:	/* bound fault */
1048 	case T_STKFLT + USER:	/* stack fault */
1049 	case T_TSSFLT + USER:	/* invalid TSS fault */
1050 		if (tudebug)
1051 			showregs(type, rp, (caddr_t)0);
1052 		siginfo.si_signo = SIGSEGV;
1053 		siginfo.si_code  = SEGV_MAPERR;
1054 		siginfo.si_addr  = (caddr_t)rp->r_pc;
1055 		fault = FLTBOUNDS;
1056 		break;
1057 
1058 	case T_ALIGNMENT + USER:	/* user alignment error (486) */
1059 		if (tudebug)
1060 			showregs(type, rp, (caddr_t)0);
1061 		bzero(&siginfo, sizeof (siginfo));
1062 		siginfo.si_signo = SIGBUS;
1063 		siginfo.si_code = BUS_ADRALN;
1064 		siginfo.si_addr = (caddr_t)rp->r_pc;
1065 		fault = FLTACCESS;
1066 		break;
1067 
1068 	case T_SGLSTP + USER: /* single step/hw breakpoint exception */
1069 		if (tudebug && tudebugbpt)
1070 			showregs(type, rp, (caddr_t)0);
1071 
1072 		/* Was it single-stepping? */
1073 		if (lwp->lwp_pcb.pcb_drstat & DR_SINGLESTEP) {
1074 			pcb_t *pcb = &lwp->lwp_pcb;
1075 
1076 			rp->r_ps &= ~PS_T;
1077 			/*
1078 			 * If both NORMAL_STEP and WATCH_STEP are in effect,
1079 			 * give precedence to NORMAL_STEP.  If neither is set,
1080 			 * user must have set the PS_T bit in %efl; treat this
1081 			 * as NORMAL_STEP.
1082 			 */
1083 			if ((pcb->pcb_flags & NORMAL_STEP) ||
1084 			    !(pcb->pcb_flags & WATCH_STEP)) {
1085 				siginfo.si_signo = SIGTRAP;
1086 				siginfo.si_code = TRAP_TRACE;
1087 				siginfo.si_addr = (caddr_t)rp->r_pc;
1088 				fault = FLTTRACE;
1089 				if (pcb->pcb_flags & WATCH_STEP)
1090 					(void) undo_watch_step(NULL);
1091 			} else {
1092 				fault = undo_watch_step(&siginfo);
1093 			}
1094 			pcb->pcb_flags &= ~(NORMAL_STEP|WATCH_STEP);
1095 		} else {
1096 			cmn_err(CE_WARN,
1097 			    "Unexpected INT 1 in user mode, dr6=%lx",
1098 			    lwp->lwp_pcb.pcb_drstat);
1099 		}
1100 		break;
1101 
1102 	case T_BPTFLT + USER:	/* breakpoint trap */
1103 		if (tudebug && tudebugbpt)
1104 			showregs(type, rp, (caddr_t)0);
1105 		/*
1106 		 * int 3 (the breakpoint instruction) leaves the pc referring
1107 		 * to the address one byte after the breakpointed address.
1108 		 * If the P_PR_BPTADJ flag has been set via /proc, We adjust
1109 		 * it back so it refers to the breakpointed address.
1110 		 */
1111 		if (p->p_proc_flag & P_PR_BPTADJ)
1112 			rp->r_pc--;
1113 		siginfo.si_signo = SIGTRAP;
1114 		siginfo.si_code  = TRAP_BRKPT;
1115 		siginfo.si_addr  = (caddr_t)rp->r_pc;
1116 		fault = FLTBPT;
1117 		break;
1118 
1119 	case T_AST:
1120 		/*
1121 		 * This occurs only after the cs register has been made to
1122 		 * look like a kernel selector, either through debugging or
1123 		 * possibly by functions like setcontext().  The thread is
1124 		 * about to cause a general protection fault at common_iret()
1125 		 * in locore.  We let that happen immediately instead of
1126 		 * doing the T_AST processing.
1127 		 */
1128 		goto cleanup;
1129 
1130 	case T_AST + USER:		/* profiling or resched pseudo trap */
1131 		if (lwp->lwp_pcb.pcb_flags & CPC_OVERFLOW) {
1132 			lwp->lwp_pcb.pcb_flags &= ~CPC_OVERFLOW;
1133 			if (kcpc_overflow_ast()) {
1134 				/*
1135 				 * Signal performance counter overflow
1136 				 */
1137 				if (tudebug)
1138 					showregs(type, rp, (caddr_t)0);
1139 				bzero(&siginfo, sizeof (siginfo));
1140 				siginfo.si_signo = SIGEMT;
1141 				siginfo.si_code = EMT_CPCOVF;
1142 				siginfo.si_addr = (caddr_t)rp->r_pc;
1143 				fault = FLTCPCOVF;
1144 			}
1145 		}
1146 		break;
1147 	}
1148 
1149 	/*
1150 	 * We can't get here from a system trap
1151 	 */
1152 	ASSERT(type & USER);
1153 
1154 	if (fault) {
1155 		/*
1156 		 * Remember the fault and fault adddress
1157 		 * for real-time (SIGPROF) profiling.
1158 		 */
1159 		lwp->lwp_lastfault = fault;
1160 		lwp->lwp_lastfaddr = siginfo.si_addr;
1161 
1162 		DTRACE_PROC2(fault, int, fault, ksiginfo_t *, &siginfo);
1163 
1164 		/*
1165 		 * If a debugger has declared this fault to be an
1166 		 * event of interest, stop the lwp.  Otherwise just
1167 		 * deliver the associated signal.
1168 		 */
1169 		if (siginfo.si_signo != SIGKILL &&
1170 		    prismember(&p->p_fltmask, fault) &&
1171 		    stop_on_fault(fault, &siginfo) == 0)
1172 			siginfo.si_signo = 0;
1173 	}
1174 
1175 	if (siginfo.si_signo)
1176 		trapsig(&siginfo, (fault == FLTCPCOVF)? 0 : 1);
1177 
1178 	if (lwp->lwp_oweupc)
1179 		profil_tick(rp->r_pc);
1180 
1181 	if (cur_thread->t_astflag | cur_thread->t_sig_check) {
1182 		/*
1183 		 * Turn off the AST flag before checking all the conditions that
1184 		 * may have caused an AST.  This flag is on whenever a signal or
1185 		 * unusual condition should be handled after the next trap or
1186 		 * syscall.
1187 		 */
1188 		astoff(cur_thread);
1189 		cur_thread->t_sig_check = 0;
1190 
1191 		mutex_enter(&p->p_lock);
1192 		if (curthread->t_proc_flag & TP_CHANGEBIND) {
1193 			timer_lwpbind();
1194 			curthread->t_proc_flag &= ~TP_CHANGEBIND;
1195 		}
1196 		mutex_exit(&p->p_lock);
1197 
1198 		/*
1199 		 * for kaio requests that are on the per-process poll queue,
1200 		 * aiop->aio_pollq, they're AIO_POLL bit is set, the kernel
1201 		 * should copyout their result_t to user memory. by copying
1202 		 * out the result_t, the user can poll on memory waiting
1203 		 * for the kaio request to complete.
1204 		 */
1205 		if (p->p_aio)
1206 			aio_cleanup(0);
1207 		/*
1208 		 * If this LWP was asked to hold, call holdlwp(), which will
1209 		 * stop.  holdlwps() sets this up and calls pokelwps() which
1210 		 * sets the AST flag.
1211 		 *
1212 		 * Also check TP_EXITLWP, since this is used by fresh new LWPs
1213 		 * through lwp_rtt().  That flag is set if the lwp_create(2)
1214 		 * syscall failed after creating the LWP.
1215 		 */
1216 		if (ISHOLD(p))
1217 			holdlwp();
1218 
1219 		/*
1220 		 * All code that sets signals and makes ISSIG evaluate true must
1221 		 * set t_astflag afterwards.
1222 		 */
1223 		if (ISSIG_PENDING(cur_thread, lwp, p)) {
1224 			if (issig(FORREAL))
1225 				psig();
1226 			cur_thread->t_sig_check = 1;
1227 		}
1228 
1229 		if (cur_thread->t_rprof != NULL) {
1230 			realsigprof(0, 0);
1231 			cur_thread->t_sig_check = 1;
1232 		}
1233 	}
1234 
1235 out:	/* We can't get here from a system trap */
1236 	ASSERT(type & USER);
1237 
1238 	if (ISHOLD(p))
1239 		holdlwp();
1240 
1241 	/*
1242 	 * Set state to LWP_USER here so preempt won't give us a kernel
1243 	 * priority if it occurs after this point.  Call CL_TRAPRET() to
1244 	 * restore the user-level priority.
1245 	 *
1246 	 * It is important that no locks (other than spinlocks) be entered
1247 	 * after this point before returning to user mode (unless lwp_state
1248 	 * is set back to LWP_SYS).
1249 	 */
1250 	lwp->lwp_state = LWP_USER;
1251 
1252 	if (cur_thread->t_trapret) {
1253 		cur_thread->t_trapret = 0;
1254 		thread_lock(cur_thread);
1255 		CL_TRAPRET(cur_thread);
1256 		thread_unlock(cur_thread);
1257 	}
1258 	if (CPU->cpu_runrun)
1259 		preempt();
1260 	(void) new_mstate(cur_thread, mstate);
1261 
1262 	/* Kernel probe */
1263 	TNF_PROBE_1(thread_state, "thread", /* CSTYLED */,
1264 	    tnf_microstate, state, LMS_USER);
1265 
1266 	return;
1267 
1268 cleanup:	/* system traps end up here */
1269 	ASSERT(!(type & USER));
1270 }
1271 
1272 /*
1273  * Patch non-zero to disable preemption of threads in the kernel.
1274  */
1275 int IGNORE_KERNEL_PREEMPTION = 0;	/* XXX - delete this someday */
1276 
1277 struct kpreempt_cnts {		/* kernel preemption statistics */
1278 	int	kpc_idle;	/* executing idle thread */
1279 	int	kpc_intr;	/* executing interrupt thread */
1280 	int	kpc_clock;	/* executing clock thread */
1281 	int	kpc_blocked;	/* thread has blocked preemption (t_preempt) */
1282 	int	kpc_notonproc;	/* thread is surrendering processor */
1283 	int	kpc_inswtch;	/* thread has ratified scheduling decision */
1284 	int	kpc_prilevel;	/* processor interrupt level is too high */
1285 	int	kpc_apreempt;	/* asynchronous preemption */
1286 	int	kpc_spreempt;	/* synchronous preemption */
1287 } kpreempt_cnts;
1288 
1289 /*
1290  * kernel preemption: forced rescheduling, preempt the running kernel thread.
1291  *	the argument is old PIL for an interrupt,
1292  *	or the distingished value KPREEMPT_SYNC.
1293  */
1294 void
1295 kpreempt(int asyncspl)
1296 {
1297 	kthread_t *cur_thread = curthread;
1298 
1299 	if (IGNORE_KERNEL_PREEMPTION) {
1300 		aston(CPU->cpu_dispthread);
1301 		return;
1302 	}
1303 
1304 	/*
1305 	 * Check that conditions are right for kernel preemption
1306 	 */
1307 	do {
1308 		if (cur_thread->t_preempt) {
1309 			/*
1310 			 * either a privileged thread (idle, panic, interrupt)
1311 			 *	or will check when t_preempt is lowered
1312 			 */
1313 			if (cur_thread->t_pri < 0)
1314 				kpreempt_cnts.kpc_idle++;
1315 			else if (cur_thread->t_flag & T_INTR_THREAD) {
1316 				kpreempt_cnts.kpc_intr++;
1317 				if (cur_thread->t_pil == CLOCK_LEVEL)
1318 					kpreempt_cnts.kpc_clock++;
1319 			} else
1320 				kpreempt_cnts.kpc_blocked++;
1321 			aston(CPU->cpu_dispthread);
1322 			return;
1323 		}
1324 		if (cur_thread->t_state != TS_ONPROC ||
1325 		    cur_thread->t_disp_queue != CPU->cpu_disp) {
1326 			/* this thread will be calling swtch() shortly */
1327 			kpreempt_cnts.kpc_notonproc++;
1328 			if (CPU->cpu_thread != CPU->cpu_dispthread) {
1329 				/* already in swtch(), force another */
1330 				kpreempt_cnts.kpc_inswtch++;
1331 				siron();
1332 			}
1333 			return;
1334 		}
1335 		if (getpil() >= DISP_LEVEL) {
1336 			/*
1337 			 * We can't preempt this thread if it is at
1338 			 * a PIL >= DISP_LEVEL since it may be holding
1339 			 * a spin lock (like sched_lock).
1340 			 */
1341 			siron();	/* check back later */
1342 			kpreempt_cnts.kpc_prilevel++;
1343 			return;
1344 		}
1345 
1346 		if (asyncspl != KPREEMPT_SYNC)
1347 			kpreempt_cnts.kpc_apreempt++;
1348 		else
1349 			kpreempt_cnts.kpc_spreempt++;
1350 
1351 		cur_thread->t_preempt++;
1352 		preempt();
1353 		cur_thread->t_preempt--;
1354 	} while (CPU->cpu_kprunrun);
1355 }
1356 
1357 /*
1358  * Print out debugging info.
1359  */
1360 static void
1361 showregs(uint_t type, struct regs *rp, caddr_t addr)
1362 {
1363 	int s;
1364 
1365 	s = spl7();
1366 	type &= ~USER;
1367 	if (u.u_comm[0])
1368 		printf("%s: ", u.u_comm);
1369 	if (type < TRAP_TYPES)
1370 		printf("#%s %s\n", trap_type_mnemonic[type], trap_type[type]);
1371 	else
1372 		switch (type) {
1373 		case T_SYSCALL:
1374 			printf("Syscall Trap:\n");
1375 			break;
1376 		case T_AST:
1377 			printf("AST\n");
1378 			break;
1379 		default:
1380 			printf("Bad Trap = %d\n", type);
1381 			break;
1382 		}
1383 	if (type == T_PGFLT) {
1384 		printf("Bad %s fault at addr=0x%lx\n",
1385 		    USERMODE(rp->r_cs) ? "user": "kernel", (uintptr_t)addr);
1386 	} else if (addr) {
1387 		printf("addr=0x%lx\n", (uintptr_t)addr);
1388 	}
1389 
1390 	printf("pid=%d, pc=0x%lx, sp=0x%lx, eflags=0x%lx\n",
1391 	    (ttoproc(curthread) && ttoproc(curthread)->p_pidp) ?
1392 	    ttoproc(curthread)->p_pid : 0, rp->r_pc, rp->r_sp, rp->r_ps);
1393 
1394 #if defined(__lint)
1395 	/*
1396 	 * this clause can be deleted when lint bug 4870403 is fixed
1397 	 * (lint thinks that bit 32 is illegal in a %b format string)
1398 	 */
1399 	printf("cr0: %x cr4: %b\n",
1400 	    (uint_t)getcr0(), (uint_t)getcr4(), FMT_CR4);
1401 #else
1402 	printf("cr0: %b cr4: %b\n",
1403 	    (uint_t)getcr0(), FMT_CR0, (uint_t)getcr4(), FMT_CR4);
1404 #endif
1405 
1406 #if defined(__amd64)
1407 	printf("cr2: %lx cr3: %lx cr8: %lx\n", getcr2(), getcr3(), getcr8());
1408 #elif defined(__i386)
1409 	printf("cr2: %lx cr3: %lx\n", getcr2(), getcr3());
1410 #endif
1411 
1412 	dumpregs(rp);
1413 	splx(s);
1414 }
1415 
1416 static void
1417 dumpregs(struct regs *rp)
1418 {
1419 #if defined(__amd64)
1420 	const char fmt[] = "\t%3s: %16lx %3s: %16lx %3s: %16lx\n";
1421 
1422 	printf(fmt, "rdi", rp->r_rdi, "rsi", rp->r_rsi, "rdx", rp->r_rdx);
1423 	printf(fmt, "rcx", rp->r_rcx, " r8", rp->r_r8, " r9", rp->r_r9);
1424 	printf(fmt, "rax", rp->r_rax, "rbx", rp->r_rbx, "rbp", rp->r_rbp);
1425 	printf(fmt, "r10", rp->r_r10, "r11", rp->r_r11, "r12", rp->r_r12);
1426 	printf(fmt, "r13", rp->r_r13, "r14", rp->r_r14, "r15", rp->r_r15);
1427 
1428 	printf(fmt, "fsb", rp->r_fsbase, "gsb", rp->r_gsbase, " ds", rp->r_ds);
1429 	printf(fmt, " es", rp->r_es, " fs", rp->r_fs, " gs", rp->r_gs);
1430 
1431 	printf(fmt, "trp", rp->r_trapno, "err", rp->r_err, "rip", rp->r_rip);
1432 	printf(fmt, " cs", rp->r_cs, "rfl", rp->r_rfl, "rsp", rp->r_rsp);
1433 
1434 	printf("\t%3s: %16lx\n", " ss", rp->r_ss);
1435 
1436 #elif defined(__i386)
1437 	const char fmt[] = "\t%3s: %8lx %3s: %8lx %3s: %8lx %3s: %8lx\n";
1438 
1439 	printf(fmt, " gs", rp->r_gs, " fs", rp->r_fs,
1440 	    " es", rp->r_es, " ds", rp->r_ds);
1441 	printf(fmt, "edi", rp->r_edi, "esi", rp->r_esi,
1442 	    "ebp", rp->r_ebp, "esp", rp->r_esp);
1443 	printf(fmt, "ebx", rp->r_ebx, "edx", rp->r_edx,
1444 	    "ecx", rp->r_ecx, "eax", rp->r_eax);
1445 	printf(fmt, "trp", rp->r_trapno, "err", rp->r_err,
1446 	    "eip", rp->r_eip, " cs", rp->r_cs);
1447 	printf("\t%3s: %8lx %3s: %8lx %3s: %8lx\n",
1448 	    "efl", rp->r_efl, "usp", rp->r_uesp, " ss", rp->r_ss);
1449 
1450 #endif	/* __i386 */
1451 }
1452 
1453 /*
1454  * Handle #gp faults in kernel mode.
1455  *
1456  * One legitimate way this can happen is if we attempt to update segment
1457  * registers to naughty values on the way out of the kernel.
1458  *
1459  * This can happen in a couple of ways: someone - either accidentally or
1460  * on purpose - creates (setcontext(2), lwp_create(2)) or modifies
1461  * (signal(2)) a ucontext that contains silly segment register values.
1462  * Or someone - either accidentally or on purpose - modifies the prgregset_t
1463  * of a subject process via /proc to contain silly segment register values.
1464  *
1465  * (The unfortunate part is that we can end up discovering the bad segment
1466  * register value in the middle of an 'iret' after we've popped most of the
1467  * stack.  So it becomes quite difficult to associate an accurate ucontext
1468  * with the lwp, because the act of taking the #gp trap overwrites most of
1469  * what we were going to send the lwp.)
1470  *
1471  * OTOH if it turns out that's -not- the problem, and we're -not- an lwp
1472  * trying to return to user mode and we get a #gp fault, then we need
1473  * to die() -- which will happen if we return non-zero from this routine.
1474  */
1475 static int
1476 kern_gpfault(struct regs *rp)
1477 {
1478 	kthread_t *t = curthread;
1479 	proc_t *p = ttoproc(t);
1480 	klwp_t *lwp = ttolwp(t);
1481 	struct regs tmpregs, *trp = NULL;
1482 	caddr_t pc = (caddr_t)rp->r_pc;
1483 	int v;
1484 
1485 	extern void _sys_rtt(), sr_sup();
1486 
1487 #if defined(__amd64)
1488 	extern void _update_sregs(), _update_sregs_done();
1489 	static const uint8_t iretq_insn[2] = { 0x48, 0xcf };
1490 
1491 #elif defined(__i386)
1492 	static const uint8_t iret_insn[1] = { 0xcf };
1493 
1494 	/*
1495 	 * Note carefully the appallingly awful dependency between
1496 	 * the instruction sequence used in __SEGREGS_POP and these
1497 	 * instructions encoded here.
1498 	 *
1499 	 * XX64	Add some commentary to locore.s/privregs.h to document this.
1500 	 */
1501 	static const uint8_t movw_0_esp_gs[4] = { 0x8e, 0x6c, 0x24, 0x0 };
1502 	static const uint8_t movw_4_esp_fs[4] = { 0x8e, 0x64, 0x24, 0x4 };
1503 	static const uint8_t movw_8_esp_es[4] = { 0x8e, 0x44, 0x24, 0x8 };
1504 	static const uint8_t movw_c_esp_ds[4] = { 0x8e, 0x5c, 0x24, 0xc };
1505 #endif
1506 	/*
1507 	 * if we're not an lwp, or the pc range is outside _sys_rtt, then
1508 	 * we should immediately be die()ing horribly
1509 	 */
1510 	if (lwp == NULL ||
1511 	    (uintptr_t)pc < (uintptr_t)_sys_rtt ||
1512 	    (uintptr_t)pc > (uintptr_t)sr_sup)
1513 		return (1);
1514 
1515 	/*
1516 	 * So at least we're in the right part of the kernel.
1517 	 *
1518 	 * Disassemble the instruction at the faulting pc.
1519 	 * Once we know what it is, we carefully reconstruct the stack
1520 	 * based on the order in which the stack is deconstructed in
1521 	 * _sys_rtt. Ew.
1522 	 */
1523 
1524 #if defined(__amd64)
1525 
1526 	if (bcmp(pc, iretq_insn, sizeof (iretq_insn)) == 0) {
1527 		/*
1528 		 * We took the #gp while trying to perform the iretq.
1529 		 * This means that either %cs or %ss are bad.
1530 		 * All we know for sure is that most of the general
1531 		 * registers have been restored, including the
1532 		 * segment registers, and all we have left on the
1533 		 * topmost part of the lwp's stack are the
1534 		 * registers that the iretq was unable to consume.
1535 		 *
1536 		 * All the rest of the state was crushed by the #gp
1537 		 * which pushed -its- registers atop our old save area
1538 		 * (because we had to decrement the stack pointer, sigh) so
1539 		 * all that we can try and do is to reconstruct the
1540 		 * crushed frame from the #gp trap frame itself.
1541 		 */
1542 		trp = &tmpregs;
1543 		trp->r_ss = lwptoregs(lwp)->r_ss;
1544 		trp->r_sp = lwptoregs(lwp)->r_sp;
1545 		trp->r_ps = lwptoregs(lwp)->r_ps;
1546 		trp->r_cs = lwptoregs(lwp)->r_cs;
1547 		trp->r_pc = lwptoregs(lwp)->r_pc;
1548 		bcopy(rp, trp, offsetof(struct regs, r_pc));
1549 
1550 		/*
1551 		 * Validate simple math
1552 		 */
1553 		ASSERT(trp->r_pc == lwptoregs(lwp)->r_pc);
1554 		ASSERT(trp->r_err == rp->r_err);
1555 
1556 	} else if ((lwp->lwp_pcb.pcb_flags & RUPDATE_PENDING) != 0 &&
1557 	    pc >= (caddr_t)_update_sregs &&
1558 	    pc < (caddr_t)_update_sregs_done) {
1559 		/*
1560 		 * This is the common case -- we're trying to load
1561 		 * a bad segment register value in the only section
1562 		 * of kernel code that ever loads segment registers.
1563 		 *
1564 		 * We don't need to do anything at this point because
1565 		 * the pcb contains all the pending segment register
1566 		 * state, and the regs are still intact because we
1567 		 * didn't adjust the stack pointer yet.  Given the fidelity
1568 		 * of all this, we could conceivably send a signal
1569 		 * to the lwp, rather than core-ing.
1570 		 */
1571 		trp = lwptoregs(lwp);
1572 		ASSERT((caddr_t)trp == (caddr_t)rp->r_sp);
1573 	}
1574 
1575 #elif defined(__i386)
1576 
1577 	if (bcmp(pc, iret_insn, sizeof (iret_insn)) == 0) {
1578 		/*
1579 		 * We took the #gp while trying to perform the iret.
1580 		 * This means that either %cs or %ss are bad.
1581 		 * All we know for sure is that most of the general
1582 		 * registers have been restored, including the
1583 		 * segment registers, and all we have left on the
1584 		 * topmost part of the lwp's stack are the registers that
1585 		 * the iret was unable to consume.
1586 		 *
1587 		 * All the rest of the state was crushed by the #gp
1588 		 * which pushed -its- registers atop our old save area
1589 		 * (because we had to decrement the stack pointer, sigh) so
1590 		 * all that we can try and do is to reconstruct the
1591 		 * crushed frame from the #gp trap frame itself.
1592 		 */
1593 		trp = &tmpregs;
1594 		trp->r_ss = lwptoregs(lwp)->r_ss;
1595 		trp->r_sp = lwptoregs(lwp)->r_sp;
1596 		trp->r_ps = lwptoregs(lwp)->r_ps;
1597 		trp->r_cs = lwptoregs(lwp)->r_cs;
1598 		trp->r_pc = lwptoregs(lwp)->r_pc;
1599 		bcopy(rp, trp, offsetof(struct regs, r_pc));
1600 
1601 		ASSERT(trp->r_pc == lwptoregs(lwp)->r_pc);
1602 		ASSERT(trp->r_err == rp->r_err);
1603 
1604 	} else {
1605 		/*
1606 		 * Segment registers are reloaded in _sys_rtt
1607 		 * via the following sequence:
1608 		 *
1609 		 *	movw	0(%esp), %gs
1610 		 *	movw	4(%esp), %fs
1611 		 *	movw	8(%esp), %es
1612 		 *	movw	12(%esp), %ds
1613 		 *	addl	$16, %esp
1614 		 *
1615 		 * Thus if any of them fault, we know the user
1616 		 * registers are left unharmed on the stack.
1617 		 */
1618 		if (bcmp(pc, movw_0_esp_gs, sizeof (movw_0_esp_gs)) == 0 ||
1619 		    bcmp(pc, movw_4_esp_fs, sizeof (movw_4_esp_fs)) == 0 ||
1620 		    bcmp(pc, movw_8_esp_es, sizeof (movw_8_esp_es)) == 0 ||
1621 		    bcmp(pc, movw_c_esp_ds, sizeof (movw_c_esp_ds)) == 0)
1622 			trp = lwptoregs(lwp);
1623 	}
1624 #endif	/* __amd64 */
1625 
1626 	if (trp == NULL)
1627 		return (1);
1628 
1629 	/*
1630 	 * If we get to here, we're reasonably confident that we've
1631 	 * correctly decoded what happened on the way out of the kernel.
1632 	 * Rewrite the lwp's registers so that we can create a core dump
1633 	 * the (at least vaguely) represents the mcontext we were
1634 	 * being asked to restore when things went so terribly wrong.
1635 	 */
1636 
1637 	/*
1638 	 * Make sure that we have a meaningful %trapno and %err.
1639 	 */
1640 	trp->r_trapno = rp->r_trapno;
1641 	trp->r_err = rp->r_err;
1642 
1643 	if ((caddr_t)trp != (caddr_t)lwptoregs(lwp))
1644 		bcopy(trp, lwptoregs(lwp), sizeof (*trp));
1645 
1646 	mutex_enter(&p->p_lock);
1647 	lwp->lwp_cursig = SIGSEGV;
1648 	mutex_exit(&p->p_lock);
1649 
1650 	/*
1651 	 * Terminate all LWPs but don't discard them.  If another lwp beat us to
1652 	 * the punch by calling exit(), evaporate now.
1653 	 */
1654 	proc_is_exiting(p);
1655 	if (exitlwps(1) != 0) {
1656 		mutex_enter(&p->p_lock);
1657 		lwp_exit();
1658 	}
1659 
1660 #ifdef C2_AUDIT
1661 	if (audit_active)		/* audit core dump */
1662 		audit_core_start(SIGSEGV);
1663 #endif
1664 	v = core(SIGSEGV, B_FALSE);
1665 #ifdef C2_AUDIT
1666 	if (audit_active)		/* audit core dump */
1667 		audit_core_finish(v ? CLD_KILLED : CLD_DUMPED);
1668 #endif
1669 	exit(v ? CLD_KILLED : CLD_DUMPED, SIGSEGV);
1670 	return (0);
1671 }
1672 
1673 /*
1674  * dump_tss() - Display the TSS structure
1675  */
1676 
1677 #if defined(__amd64)
1678 
1679 static void
1680 dump_tss(void)
1681 {
1682 	const char tss_fmt[] = "tss.%s:\t0x%p\n";  /* Format string */
1683 	struct tss *tss = CPU->cpu_tss;
1684 
1685 	printf(tss_fmt, "tss_rsp0", (void *)tss->tss_rsp0);
1686 	printf(tss_fmt, "tss_rsp1", (void *)tss->tss_rsp1);
1687 	printf(tss_fmt, "tss_rsp2", (void *)tss->tss_rsp2);
1688 
1689 	printf(tss_fmt, "tss_ist1", (void *)tss->tss_ist1);
1690 	printf(tss_fmt, "tss_ist2", (void *)tss->tss_ist2);
1691 	printf(tss_fmt, "tss_ist3", (void *)tss->tss_ist3);
1692 	printf(tss_fmt, "tss_ist4", (void *)tss->tss_ist4);
1693 	printf(tss_fmt, "tss_ist5", (void *)tss->tss_ist5);
1694 	printf(tss_fmt, "tss_ist6", (void *)tss->tss_ist6);
1695 	printf(tss_fmt, "tss_ist7", (void *)tss->tss_ist7);
1696 }
1697 
1698 #elif defined(__i386)
1699 
1700 static void
1701 dump_tss(void)
1702 {
1703 	const char tss_fmt[] = "tss.%s:\t0x%p\n";  /* Format string */
1704 	struct tss *tss = CPU->cpu_tss;
1705 
1706 	printf(tss_fmt, "tss_link", (void *)(uintptr_t)tss->tss_link);
1707 	printf(tss_fmt, "tss_esp0", (void *)(uintptr_t)tss->tss_esp0);
1708 	printf(tss_fmt, "tss_ss0", (void *)(uintptr_t)tss->tss_ss0);
1709 	printf(tss_fmt, "tss_esp1", (void *)(uintptr_t)tss->tss_esp1);
1710 	printf(tss_fmt, "tss_ss1", (void *)(uintptr_t)tss->tss_ss1);
1711 	printf(tss_fmt, "tss_esp2", (void *)(uintptr_t)tss->tss_esp2);
1712 	printf(tss_fmt, "tss_ss2", (void *)(uintptr_t)tss->tss_ss2);
1713 	printf(tss_fmt, "tss_cr3", (void *)(uintptr_t)tss->tss_cr3);
1714 	printf(tss_fmt, "tss_eip", (void *)(uintptr_t)tss->tss_eip);
1715 	printf(tss_fmt, "tss_eflags", (void *)(uintptr_t)tss->tss_eflags);
1716 	printf(tss_fmt, "tss_eax", (void *)(uintptr_t)tss->tss_eax);
1717 	printf(tss_fmt, "tss_ebx", (void *)(uintptr_t)tss->tss_ebx);
1718 	printf(tss_fmt, "tss_ecx", (void *)(uintptr_t)tss->tss_ecx);
1719 	printf(tss_fmt, "tss_edx", (void *)(uintptr_t)tss->tss_edx);
1720 	printf(tss_fmt, "tss_esp", (void *)(uintptr_t)tss->tss_esp);
1721 }
1722 
1723 #endif	/* __amd64 */
1724 
1725 #if defined(TRAPTRACE)
1726 
1727 int ttrace_nrec = 0;		/* number of records to dump out */
1728 int ttrace_dump_nregs = 5;	/* dump out this many records with regs too */
1729 
1730 /*
1731  * Dump out the last ttrace_nrec traptrace records on each CPU
1732  */
1733 static void
1734 dump_ttrace(void)
1735 {
1736 	trap_trace_ctl_t *ttc;
1737 	trap_trace_rec_t *rec;
1738 	uintptr_t current;
1739 	int i, j, k;
1740 	int n = NCPU;
1741 #if defined(__amd64)
1742 	const char banner[] =
1743 		"\ncpu          address    timestamp "
1744 		"type  vc  handler   pc\n";
1745 	const char fmt1[] = "%3d %016lx %12llx ";
1746 #elif defined(__i386)
1747 	const char banner[] =
1748 		"\ncpu address     timestamp type  vc  handler   pc\n";
1749 	const char fmt1[] = "%3d %08lx %12llx ";
1750 #endif
1751 	const char fmt2[] = "%4s %3x ";
1752 	const char fmt3[] = "%8s ";
1753 
1754 	if (ttrace_nrec == 0)
1755 		return;
1756 
1757 	printf(banner);
1758 
1759 	for (i = 0; i < n; i++) {
1760 		ttc = &trap_trace_ctl[i];
1761 		if (ttc->ttc_first == NULL)
1762 			continue;
1763 
1764 		current = ttc->ttc_next - sizeof (trap_trace_rec_t);
1765 		for (j = 0; j < ttrace_nrec; j++) {
1766 			struct sysent	*sys;
1767 			struct autovec	*vec;
1768 			extern struct av_head autovect[];
1769 			int type;
1770 			ulong_t	off;
1771 			char *sym, *stype;
1772 
1773 			if (current < ttc->ttc_first)
1774 				current =
1775 				    ttc->ttc_limit - sizeof (trap_trace_rec_t);
1776 
1777 			if (current == NULL)
1778 				continue;
1779 
1780 			rec = (trap_trace_rec_t *)current;
1781 
1782 			if (rec->ttr_stamp == 0)
1783 				break;
1784 
1785 			printf(fmt1, i, (uintptr_t)rec, rec->ttr_stamp);
1786 
1787 			switch (rec->ttr_marker) {
1788 			case TT_SYSCALL:
1789 			case TT_SYSENTER:
1790 			case TT_SYSC:
1791 			case TT_SYSC64:
1792 #if defined(__amd64)
1793 				sys = &sysent32[rec->ttr_sysnum];
1794 				switch (rec->ttr_marker) {
1795 				case TT_SYSC64:
1796 					sys = &sysent[rec->ttr_sysnum];
1797 					/*FALLTHROUGH*/
1798 #elif defined(__i386)
1799 				sys = &sysent[rec->ttr_sysnum];
1800 				switch (rec->ttr_marker) {
1801 				case TT_SYSC64:
1802 #endif
1803 				case TT_SYSC:
1804 					stype = "sysc";	/* syscall */
1805 					break;
1806 				case TT_SYSCALL:
1807 					stype = "lcal";	/* lcall */
1808 					break;
1809 				case TT_SYSENTER:
1810 					stype = "syse";	/* sysenter */
1811 					break;
1812 				default:
1813 					break;
1814 				}
1815 				printf(fmt2, "sysc", rec->ttr_sysnum);
1816 				if (sys != NULL) {
1817 					sym = kobj_getsymname(
1818 					    (uintptr_t)sys->sy_callc,
1819 					    &off);
1820 					if (sym != NULL)
1821 						printf("%s ", sym);
1822 					else
1823 						printf("%p ", sys->sy_callc);
1824 				} else {
1825 					printf("unknown ");
1826 				}
1827 				break;
1828 
1829 			case TT_INTERRUPT:
1830 				printf(fmt2, "intr", rec->ttr_vector);
1831 				vec = (&autovect[rec->ttr_vector])->avh_link;
1832 				if (vec != NULL) {
1833 					sym = kobj_getsymname(
1834 					    (uintptr_t)vec->av_vector, &off);
1835 					if (sym != NULL)
1836 						printf("%s ", sym);
1837 					else
1838 						printf("%p ", vec->av_vector);
1839 				} else {
1840 					printf("unknown ");
1841 				}
1842 				break;
1843 
1844 			case TT_TRAP:
1845 				type = rec->ttr_regs.r_trapno;
1846 				printf(fmt2, "trap", type);
1847 				printf("#%s ", type < TRAP_TYPES ?
1848 				    trap_type_mnemonic[type] : "trap");
1849 				break;
1850 
1851 			default:
1852 				break;
1853 			}
1854 
1855 			sym = kobj_getsymname(rec->ttr_regs.r_pc, &off);
1856 			if (sym != NULL)
1857 				printf("%s+%lx\n", sym, off);
1858 			else
1859 				printf("%lx\n", rec->ttr_regs.r_pc);
1860 
1861 			if (ttrace_dump_nregs-- > 0) {
1862 				int s;
1863 
1864 				if (rec->ttr_marker == TT_INTERRUPT)
1865 					printf(
1866 					    "\t\tipl %x spl %x pri %x\n",
1867 					    rec->ttr_ipl,
1868 					    rec->ttr_spl,
1869 					    rec->ttr_pri);
1870 
1871 				dumpregs(&rec->ttr_regs);
1872 
1873 				printf("\t%3s: %p\n\n", " ct",
1874 				    (void *)rec->ttr_curthread);
1875 
1876 				/*
1877 				 * print out the pc stack that we recorded
1878 				 * at trap time (if any)
1879 				 */
1880 				for (s = 0; s < rec->ttr_sdepth; s++) {
1881 					uintptr_t fullpc;
1882 
1883 					if (s >= TTR_STACK_DEPTH) {
1884 						printf("ttr_sdepth corrupt\n");
1885 						break;
1886 					}
1887 
1888 					fullpc = (uintptr_t)rec->ttr_stack[s];
1889 
1890 					sym = kobj_getsymname(fullpc, &off);
1891 					if (sym != NULL)
1892 						printf("-> %s+0x%lx()\n",
1893 						    sym, off);
1894 					else
1895 						printf("-> 0x%lx()\n", fullpc);
1896 				}
1897 				printf("\n");
1898 			}
1899 			current -= sizeof (trap_trace_rec_t);
1900 		}
1901 	}
1902 }
1903 
1904 #endif	/* TRAPTRACE */
1905 
1906 void
1907 panic_showtrap(struct trap_info *tip)
1908 {
1909 	showregs(tip->trap_type, tip->trap_regs, tip->trap_addr);
1910 
1911 #if defined(TRAPTRACE)
1912 	dump_ttrace();
1913 #endif	/* TRAPTRACE */
1914 
1915 	if (tip->trap_type == T_DBLFLT)
1916 		dump_tss();
1917 }
1918 
1919 void
1920 panic_savetrap(panic_data_t *pdp, struct trap_info *tip)
1921 {
1922 	panic_saveregs(pdp, tip->trap_regs);
1923 }
1924