xref: /illumos-gate/usr/src/uts/i86pc/os/trap.c (revision 9dd828891378a0a6a509ab601b4c5c20ca5562ec)
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 (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*	Copyright (c) 1990, 1991 UNIX System Laboratories, Inc. */
28 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989, 1990 AT&T   */
29 /*		All Rights Reserved   				*/
30 /*								*/
31 /*	Copyright (c) 1987, 1988 Microsoft Corporation  	*/
32 /*		All Rights Reserved   				*/
33 /*								*/
34 
35 #pragma ident	"%Z%%M%	%I%	%E% SMI"
36 
37 #include <sys/types.h>
38 #include <sys/sysmacros.h>
39 #include <sys/param.h>
40 #include <sys/signal.h>
41 #include <sys/systm.h>
42 #include <sys/user.h>
43 #include <sys/proc.h>
44 #include <sys/disp.h>
45 #include <sys/class.h>
46 #include <sys/core.h>
47 #include <sys/syscall.h>
48 #include <sys/cpuvar.h>
49 #include <sys/vm.h>
50 #include <sys/sysinfo.h>
51 #include <sys/fault.h>
52 #include <sys/stack.h>
53 #include <sys/mmu.h>
54 #include <sys/psw.h>
55 #include <sys/regset.h>
56 #include <sys/fp.h>
57 #include <sys/trap.h>
58 #include <sys/kmem.h>
59 #include <sys/vtrace.h>
60 #include <sys/cmn_err.h>
61 #include <sys/prsystm.h>
62 #include <sys/mutex_impl.h>
63 #include <sys/machsystm.h>
64 #include <sys/archsystm.h>
65 #include <sys/sdt.h>
66 #include <sys/avintr.h>
67 #include <sys/kobj.h>
68 
69 #include <vm/hat.h>
70 
71 #include <vm/seg_kmem.h>
72 #include <vm/as.h>
73 #include <vm/seg.h>
74 #include <vm/hat_pte.h>
75 
76 #include <sys/procfs.h>
77 
78 #include <sys/reboot.h>
79 #include <sys/debug.h>
80 #include <sys/debugreg.h>
81 #include <sys/modctl.h>
82 #include <sys/aio_impl.h>
83 #include <sys/tnf.h>
84 #include <sys/tnf_probe.h>
85 #include <sys/cred.h>
86 #include <sys/mman.h>
87 #include <sys/x86_archext.h>
88 #include <sys/copyops.h>
89 #include <c2/audit.h>
90 #include <sys/ftrace.h>
91 #include <sys/panic.h>
92 #include <sys/traptrace.h>
93 #include <sys/ontrap.h>
94 #include <sys/cpc_impl.h>
95 
96 #define	USER	0x10000		/* user-mode flag added to trap type */
97 
98 static const char *trap_type_mnemonic[] = {
99 	"de",	"db",	"2",	"bp",
100 	"of",	"br",	"ud",	"nm",
101 	"df",	"9",	"ts",	"np",
102 	"ss",	"gp",	"pf",	"15",
103 	"mf",	"ac",	"mc",	"xf"
104 };
105 
106 static const char *trap_type[] = {
107 	"Divide error",				/* trap id 0 	*/
108 	"Debug",				/* trap id 1	*/
109 	"NMI interrupt",			/* trap id 2	*/
110 	"Breakpoint",				/* trap id 3 	*/
111 	"Overflow",				/* trap id 4 	*/
112 	"BOUND range exceeded",			/* trap id 5 	*/
113 	"Invalid opcode",			/* trap id 6 	*/
114 	"Device not available",			/* trap id 7 	*/
115 	"Double fault",				/* trap id 8 	*/
116 	"Coprocessor segment overrun",		/* trap id 9 	*/
117 	"Invalid TSS",				/* trap id 10 	*/
118 	"Segment not present",			/* trap id 11 	*/
119 	"Stack segment fault",			/* trap id 12 	*/
120 	"General protection",			/* trap id 13 	*/
121 	"Page fault",				/* trap id 14 	*/
122 	"Reserved",				/* trap id 15 	*/
123 	"x87 floating point error",		/* trap id 16 	*/
124 	"Alignment check",			/* trap id 17 	*/
125 	"Machine check",			/* trap id 18	*/
126 	"SIMD floating point exception",	/* trap id 19	*/
127 };
128 
129 #define	TRAP_TYPES	(sizeof (trap_type) / sizeof (trap_type[0]))
130 
131 int tudebug = 0;
132 int tudebugbpt = 0;
133 int tudebugfpe = 0;
134 int tudebugsse = 0;
135 
136 #if defined(TRAPDEBUG) || defined(lint)
137 int tdebug = 0;
138 int lodebug = 0;
139 int faultdebug = 0;
140 #else
141 #define	tdebug	0
142 #define	lodebug	0
143 #define	faultdebug	0
144 #endif /* defined(TRAPDEBUG) || defined(lint) */
145 
146 #if defined(TRAPTRACE)
147 static void dump_ttrace(void);
148 #endif	/* TRAPTRACE */
149 static void dumpregs(struct regs *);
150 static void showregs(uint_t, struct regs *, caddr_t);
151 static void dump_tss(void);
152 static int kern_gpfault(struct regs *);
153 
154 struct trap_info {
155 	struct regs *trap_regs;
156 	uint_t trap_type;
157 	caddr_t trap_addr;
158 };
159 
160 /*ARGSUSED*/
161 static int
162 die(uint_t type, struct regs *rp, caddr_t addr, processorid_t cpuid)
163 {
164 	struct trap_info ti;
165 	const char *trap_name, *trap_mnemonic;
166 
167 	if (type < TRAP_TYPES) {
168 		trap_name = trap_type[type];
169 		trap_mnemonic = trap_type_mnemonic[type];
170 	} else {
171 		trap_name = "trap";
172 		trap_mnemonic = "-";
173 	}
174 
175 #ifdef TRAPTRACE
176 	TRAPTRACE_FREEZE;
177 #endif
178 
179 	ti.trap_regs = rp;
180 	ti.trap_type = type & ~USER;
181 	ti.trap_addr = addr;
182 
183 	curthread->t_panic_trap = &ti;
184 
185 	if (type == T_PGFLT && addr < (caddr_t)KERNELBASE) {
186 		panic("BAD TRAP: type=%x (#%s %s) rp=%p addr=%p "
187 		    "occurred in module \"%s\" due to %s",
188 		    type, trap_mnemonic, trap_name, (void *)rp, (void *)addr,
189 		    mod_containing_pc((caddr_t)rp->r_pc),
190 		    addr < (caddr_t)PAGESIZE ?
191 		    "a NULL pointer dereference" :
192 		    "an illegal access to a user address");
193 	} else
194 		panic("BAD TRAP: type=%x (#%s %s) rp=%p addr=%p",
195 		    type, trap_mnemonic, trap_name, (void *)rp, (void *)addr);
196 	return (0);
197 }
198 
199 /*
200  * Rewrite the instruction at pc to be an int $T_SYSCALLINT instruction.
201  *
202  * int <vector> is two bytes: 0xCD <vector>
203  */
204 
205 #define	SLOW_SCALL_SIZE	2
206 
207 static int
208 rewrite_syscall(caddr_t pc)
209 {
210 	uchar_t instr[SLOW_SCALL_SIZE] = { 0xCD, T_SYSCALLINT };
211 
212 	if (uwrite(curthread->t_procp, instr, SLOW_SCALL_SIZE,
213 	    (uintptr_t)pc) != 0)
214 		return (1);
215 
216 	return (0);
217 }
218 
219 /*
220  * Test to see if the instruction at pc is sysenter or syscall. The second
221  * argument should be the x86 feature flag corresponding to the expected
222  * instruction.
223  *
224  * sysenter is two bytes: 0x0F 0x34
225  * syscall is two bytes:  0x0F 0x05
226  */
227 
228 #define	FAST_SCALL_SIZE	2
229 
230 static int
231 instr_is_fast_syscall(caddr_t pc, int which)
232 {
233 	uchar_t instr[FAST_SCALL_SIZE];
234 
235 	ASSERT(which == X86_SEP || which == X86_ASYSC);
236 
237 	if (copyin_nowatch(pc, (caddr_t)instr, FAST_SCALL_SIZE) != 0 ||
238 	    instr[0] != 0x0F)
239 		return (0);
240 
241 	if ((which == X86_SEP && instr[1] == 0x34) ||
242 	    (which == X86_ASYSC && instr[1] == 0x05))
243 		return (1);
244 
245 	return (0);
246 }
247 
248 /*
249  * Test to see if the instruction at pc is a system call instruction.
250  *
251  * The bytes of an lcall instruction used for the syscall trap.
252  * static uchar_t lcall[7] = { 0x9a, 0, 0, 0, 0, 0x7, 0 };
253  * static uchar_t lcallalt[7] = { 0x9a, 0, 0, 0, 0, 0x27, 0 };
254  */
255 
256 #define	LCALLSIZE	7
257 
258 static int
259 instr_is_syscall(caddr_t pc)
260 {
261 	uchar_t instr[LCALLSIZE];
262 
263 	if (copyin_nowatch(pc, (caddr_t)instr, LCALLSIZE) == 0 &&
264 	    instr[0] == 0x9a &&
265 	    instr[1] == 0 &&
266 	    instr[2] == 0 &&
267 	    instr[3] == 0 &&
268 	    instr[4] == 0 &&
269 	    (instr[5] == 0x7 || instr[5] == 0x27) &&
270 	    instr[6] == 0)
271 		return (1);
272 
273 	return (0);
274 }
275 
276 #ifdef __amd64
277 
278 /*
279  * In the first revisions of AMD64 CPUs produced by AMD, the LAHF and
280  * SAHF instructions were not implemented in 64bit mode. Later revisions
281  * did implement these instructions. An extension to the cpuid instruction
282  * was added to check for the capability of executing these instructions
283  * in 64bit mode.
284  *
285  * Intel originally did not implement these instructions in EM64T either,
286  * but added them in later revisions.
287  *
288  * So, there are different chip revisions by both vendors out there that
289  * may or may not implement these instructions. The easy solution is to
290  * just always emulate these instructions on demand.
291  *
292  * SAHF == store %ah in the lower 8 bits of %rflags (opcode 0x9e)
293  * LAHF == load the lower 8 bits of %rflags into %ah (opcode 0x9f)
294  */
295 
296 #define	LSAHFSIZE 1
297 
298 static int
299 instr_is_lsahf(caddr_t pc, uchar_t *instr)
300 {
301 	if (copyin_nowatch(pc, (caddr_t)instr, LSAHFSIZE) == 0 &&
302 	    (*instr == 0x9e || *instr == 0x9f))
303 		return (1);
304 	return (0);
305 }
306 
307 /*
308  * Emulate the LAHF and SAHF instructions. The reference manuals define
309  * these instructions to always load/store bit 1 as a 1, and bits 3 and 5
310  * as a 0. The other, defined, bits are copied (the PS_ICC bits and PS_P).
311  *
312  * Note that %ah is bits 8-15 of %rax.
313  */
314 static void
315 emulate_lsahf(struct regs *rp, uchar_t instr)
316 {
317 	if (instr == 0x9e) {
318 		/* sahf. Copy bits from %ah to flags. */
319 		rp->r_ps = (rp->r_ps & ~0xff) |
320 		    ((rp->r_rax >> 8) & PSL_LSAHFMASK) | PS_MB1;
321 	} else {
322 		/* lahf. Copy bits from flags to %ah. */
323 		rp->r_rax = (rp->r_rax & ~0xff00) |
324 		    (((rp->r_ps & PSL_LSAHFMASK) | PS_MB1) << 8);
325 	}
326 	rp->r_pc += LSAHFSIZE;
327 }
328 #endif /* __amd64 */
329 
330 #ifdef OPTERON_ERRATUM_91
331 
332 /*
333  * Test to see if the instruction at pc is a prefetch instruction.
334  *
335  * The first byte of prefetch instructions is always 0x0F.
336  * The second byte is 0x18 for regular prefetch or 0x0D for AMD 3dnow prefetch.
337  * The third byte is between 0 and 3 inclusive.
338  */
339 
340 #define	PREFETCHSIZE 3
341 
342 static int
343 cmp_to_prefetch(uchar_t *p)
344 {
345 	if (*p == 0x0F && (*(p+1) == 0x18 || *(p+1) == 0x0D) && *(p+2) <= 3)
346 		return (1);
347 	return (0);
348 }
349 
350 static int
351 instr_is_prefetch(caddr_t pc)
352 {
353 	uchar_t instr[PREFETCHSIZE];
354 	int	error;
355 
356 	error = copyin_nowatch(pc, (caddr_t)instr, PREFETCHSIZE);
357 
358 	if (error == 0 && cmp_to_prefetch(instr))
359 		return (1);
360 	return (0);
361 }
362 
363 #endif /* OPTERON_ERRATUM_91 */
364 
365 /*
366  * Called from the trap handler when a processor trap occurs.
367  *
368  * Note: All user-level traps that might call stop() must exit
369  * trap() by 'goto out' or by falling through.
370  */
371 void
372 trap(struct regs *rp, caddr_t addr, processorid_t cpuid)
373 {
374 	kthread_t *cur_thread = curthread;
375 	enum seg_rw rw;
376 	unsigned type;
377 	proc_t *p = ttoproc(cur_thread);
378 	klwp_t *lwp = ttolwp(cur_thread);
379 	uintptr_t lofault;
380 	faultcode_t pagefault(), res, errcode;
381 	enum fault_type fault_type;
382 	k_siginfo_t siginfo;
383 	uint_t fault = 0;
384 	int mstate;
385 	int sicode = 0;
386 	int watchcode;
387 	int watchpage;
388 	caddr_t vaddr;
389 	size_t sz;
390 	int ta;
391 #ifdef __amd64
392 	uchar_t instr;
393 #endif
394 
395 	ASSERT_STACK_ALIGNED();
396 
397 	type = rp->r_trapno;
398 	CPU_STATS_ADDQ(CPU, sys, trap, 1);
399 
400 	ASSERT(cur_thread->t_schedflag & TS_DONT_SWAP);
401 
402 	if (type == T_PGFLT) {
403 
404 		errcode = rp->r_err;
405 		if (errcode & PF_ERR_WRITE)
406 			rw = S_WRITE;
407 		else if ((caddr_t)rp->r_pc == addr ||
408 		    (mmu.pt_nx != 0 && (errcode & PF_ERR_EXEC)))
409 			rw = S_EXEC;
410 		else
411 			rw = S_READ;
412 
413 #if defined(__i386)
414 		/*
415 		 * Pentium Pro work-around
416 		 */
417 		if ((errcode & PF_ERR_PROT) && pentiumpro_bug4046376) {
418 			uint_t	attr;
419 			uint_t	priv_violation;
420 			uint_t	access_violation;
421 
422 			if (hat_getattr(addr < (caddr_t)kernelbase ?
423 			    curproc->p_as->a_hat : kas.a_hat, addr, &attr)
424 			    == -1) {
425 				errcode &= ~PF_ERR_PROT;
426 			} else {
427 				priv_violation = (errcode & PF_ERR_USER) &&
428 					!(attr & PROT_USER);
429 				access_violation = (errcode & PF_ERR_WRITE) &&
430 					!(attr & PROT_WRITE);
431 				if (!priv_violation && !access_violation)
432 					goto cleanup;
433 			}
434 		}
435 #endif /* __i386 */
436 
437 	}
438 
439 	if (tdebug)
440 		showregs(type, rp, addr);
441 
442 	if (USERMODE(rp->r_cs)) {
443 		/*
444 		 * Set up the current cred to use during this trap. u_cred
445 		 * no longer exists.  t_cred is used instead.
446 		 * The current process credential applies to the thread for
447 		 * the entire trap.  If trapping from the kernel, this
448 		 * should already be set up.
449 		 */
450 		if (cur_thread->t_cred != p->p_cred) {
451 			cred_t *oldcred = cur_thread->t_cred;
452 			/*
453 			 * DTrace accesses t_cred in probe context.  t_cred
454 			 * must always be either NULL, or point to a valid,
455 			 * allocated cred structure.
456 			 */
457 			cur_thread->t_cred = crgetcred();
458 			crfree(oldcred);
459 		}
460 		ASSERT(lwp != NULL);
461 		type |= USER;
462 		ASSERT(lwptoregs(lwp) == rp);
463 		lwp->lwp_state = LWP_SYS;
464 
465 		switch (type) {
466 		case T_PGFLT + USER:
467 			if ((caddr_t)rp->r_pc == addr)
468 				mstate = LMS_TFAULT;
469 			else
470 				mstate = LMS_DFAULT;
471 			break;
472 		default:
473 			mstate = LMS_TRAP;
474 			break;
475 		}
476 		/* Kernel probe */
477 		TNF_PROBE_1(thread_state, "thread", /* CSTYLED */,
478 		    tnf_microstate, state, mstate);
479 		mstate = new_mstate(cur_thread, mstate);
480 
481 		bzero(&siginfo, sizeof (siginfo));
482 	}
483 
484 	switch (type) {
485 	case T_PGFLT + USER:
486 	case T_SGLSTP:
487 	case T_SGLSTP + USER:
488 	case T_BPTFLT + USER:
489 		break;
490 
491 	default:
492 		FTRACE_2("trap(): type=0x%lx, regs=0x%lx",
493 		    (ulong_t)type, (ulong_t)rp);
494 		break;
495 	}
496 
497 	switch (type) {
498 	default:
499 		if (type & USER) {
500 			if (tudebug)
501 				showregs(type, rp, (caddr_t)0);
502 			printf("trap: Unknown trap type %d in user mode\n",
503 			    type & ~USER);
504 			siginfo.si_signo = SIGILL;
505 			siginfo.si_code  = ILL_ILLTRP;
506 			siginfo.si_addr  = (caddr_t)rp->r_pc;
507 			siginfo.si_trapno = type & ~USER;
508 			fault = FLTILL;
509 			break;
510 		} else {
511 			(void) die(type, rp, addr, cpuid);
512 			/*NOTREACHED*/
513 		}
514 
515 	case T_PGFLT:		/* system page fault */
516 		/*
517 		 * If we're under on_trap() protection (see <sys/ontrap.h>),
518 		 * set ot_trap and longjmp back to the on_trap() call site.
519 		 */
520 		if ((cur_thread->t_ontrap != NULL) &&
521 		    (cur_thread->t_ontrap->ot_prot & OT_DATA_ACCESS)) {
522 			curthread->t_ontrap->ot_trap |= OT_DATA_ACCESS;
523 			longjmp(&curthread->t_ontrap->ot_jmpbuf);
524 		}
525 
526 		/*
527 		 * See if we can handle as pagefault. Save lofault
528 		 * across this. Here we assume that an address
529 		 * less than KERNELBASE is a user fault.
530 		 * We can do this as copy.s routines verify that the
531 		 * starting address is less than KERNELBASE before
532 		 * starting and because we know that we always have
533 		 * KERNELBASE mapped as invalid to serve as a "barrier".
534 		 */
535 		lofault = cur_thread->t_lofault;
536 		cur_thread->t_lofault = 0;
537 
538 		mstate = new_mstate(cur_thread, LMS_KFAULT);
539 
540 		if (addr < (caddr_t)kernelbase) {
541 			res = pagefault(addr,
542 			    (errcode & PF_ERR_PROT)? F_PROT: F_INVAL, rw, 0);
543 			if (res == FC_NOMAP &&
544 			    addr < p->p_usrstack &&
545 			    grow(addr))
546 				res = 0;
547 		} else {
548 			res = pagefault(addr,
549 			    (errcode & PF_ERR_PROT)? F_PROT: F_INVAL, rw, 1);
550 		}
551 		(void) new_mstate(cur_thread, mstate);
552 
553 		/*
554 		 * Restore lofault. If we resolved the fault, exit.
555 		 * If we didn't and lofault wasn't set, die.
556 		 */
557 		cur_thread->t_lofault = lofault;
558 		if (res == 0)
559 			goto cleanup;
560 
561 #if defined(OPTERON_ERRATUM_93) && defined(_LP64)
562 		if (lofault == 0 && opteron_erratum_93) {
563 			/*
564 			 * Workaround for Opteron Erratum 93. On return from
565 			 * a System Managment Interrupt at a HLT instruction
566 			 * the %rip might be truncated to a 32 bit value.
567 			 * BIOS is supposed to fix this, but some don't.
568 			 * If this occurs we simply restore the high order bits.
569 			 * The HLT instruction is 1 byte of 0xf4.
570 			 */
571 			uintptr_t	rip = rp->r_pc;
572 
573 			if ((rip & 0xfffffffful) == rip) {
574 				rip |= 0xfffffffful << 32;
575 				if (hat_getpfnum(kas.a_hat, (caddr_t)rip) !=
576 				    PFN_INVALID &&
577 				    (*(uchar_t *)rip == 0xf4 ||
578 				    *(uchar_t *)(rip - 1) == 0xf4)) {
579 					rp->r_pc = rip;
580 					goto cleanup;
581 				}
582 			}
583 		}
584 #endif /* OPTERON_ERRATUM_93 && _LP64 */
585 
586 #ifdef OPTERON_ERRATUM_91
587 		if (lofault == 0 && opteron_erratum_91) {
588 			/*
589 			 * Workaround for Opteron Erratum 91. Prefetches may
590 			 * generate a page fault (they're not supposed to do
591 			 * that!). If this occurs we simply return back to the
592 			 * instruction.
593 			 */
594 			caddr_t		pc = (caddr_t)rp->r_pc;
595 
596 			/*
597 			 * If the faulting PC is not mapped, this is a
598 			 * legitimate kernel page fault that must result in a
599 			 * panic. If the faulting PC is mapped, it could contain
600 			 * a prefetch instruction. Check for that here.
601 			 */
602 			if (hat_getpfnum(kas.a_hat, pc) != PFN_INVALID) {
603 				if (cmp_to_prefetch((uchar_t *)pc)) {
604 #ifdef DEBUG
605 					cmn_err(CE_WARN, "Opteron erratum 91 "
606 					    "occurred: kernel prefetch"
607 					    " at %p generated a page fault!",
608 					    (void *)rp->r_pc);
609 #endif /* DEBUG */
610 					goto cleanup;
611 				}
612 			}
613 			(void) die(type, rp, addr, cpuid);
614 		}
615 #endif /* OPTERON_ERRATUM_91 */
616 
617 		if (lofault == 0)
618 			(void) die(type, rp, addr, cpuid);
619 
620 		/*
621 		 * Cannot resolve fault.  Return to lofault.
622 		 */
623 		if (lodebug) {
624 			showregs(type, rp, addr);
625 			traceregs(rp);
626 		}
627 		if (FC_CODE(res) == FC_OBJERR)
628 			res = FC_ERRNO(res);
629 		else
630 			res = EFAULT;
631 		rp->r_r0 = res;
632 		rp->r_pc = cur_thread->t_lofault;
633 		goto cleanup;
634 
635 	case T_PGFLT + USER:	/* user page fault */
636 		if (faultdebug) {
637 			char *fault_str;
638 
639 			switch (rw) {
640 			case S_READ:
641 				fault_str = "read";
642 				break;
643 			case S_WRITE:
644 				fault_str = "write";
645 				break;
646 			case S_EXEC:
647 				fault_str = "exec";
648 				break;
649 			default:
650 				fault_str = "";
651 				break;
652 			}
653 			printf("user %s fault:  addr=0x%lx errcode=0x%x\n",
654 			    fault_str, (uintptr_t)addr, errcode);
655 		}
656 
657 #if defined(OPTERON_ERRATUM_100) && defined(_LP64)
658 		/*
659 		 * Workaround for AMD erratum 100
660 		 *
661 		 * A 32-bit process may receive a page fault on a non
662 		 * 32-bit address by mistake. The range of the faulting
663 		 * address will be
664 		 *
665 		 *	0xffffffff80000000 .. 0xffffffffffffffff or
666 		 *	0x0000000100000000 .. 0x000000017fffffff
667 		 *
668 		 * The fault is always due to an instruction fetch, however
669 		 * the value of r_pc should be correct (in 32 bit range),
670 		 * so we ignore the page fault on the bogus address.
671 		 */
672 		if (p->p_model == DATAMODEL_ILP32 &&
673 		    (0xffffffff80000000 <= (uintptr_t)addr ||
674 		    (0x100000000 <= (uintptr_t)addr &&
675 		    (uintptr_t)addr <= 0x17fffffff))) {
676 			if (!opteron_erratum_100)
677 				panic("unexpected erratum #100");
678 			if (rp->r_pc <= 0xffffffff)
679 				goto out;
680 		}
681 #endif /* OPTERON_ERRATUM_100 && _LP64 */
682 
683 		ASSERT(!(curthread->t_flag & T_WATCHPT));
684 		watchpage = (pr_watch_active(p) && pr_is_watchpage(addr, rw));
685 #ifdef __i386
686 		/*
687 		 * In 32-bit mode, the lcall (system call) instruction fetches
688 		 * one word from the stack, at the stack pointer, because of the
689 		 * way the call gate is constructed.  This is a bogus
690 		 * read and should not be counted as a read watchpoint.
691 		 * We work around the problem here by testing to see if
692 		 * this situation applies and, if so, simply jumping to
693 		 * the code in locore.s that fields the system call trap.
694 		 * The registers on the stack are already set up properly
695 		 * due to the match between the call gate sequence and the
696 		 * trap gate sequence.  We just have to adjust the pc.
697 		 */
698 		if (watchpage && addr == (caddr_t)rp->r_sp &&
699 		    rw == S_READ && instr_is_syscall((caddr_t)rp->r_pc)) {
700 			extern void watch_syscall(void);
701 
702 			rp->r_pc += LCALLSIZE;
703 			watch_syscall();	/* never returns */
704 			/* NOTREACHED */
705 		}
706 #endif /* __i386 */
707 		vaddr = addr;
708 		if (!watchpage || (sz = instr_size(rp, &vaddr, rw)) <= 0)
709 			fault_type = (errcode & PF_ERR_PROT)? F_PROT: F_INVAL;
710 		else if ((watchcode = pr_is_watchpoint(&vaddr, &ta,
711 		    sz, NULL, rw)) != 0) {
712 			if (ta) {
713 				do_watch_step(vaddr, sz, rw,
714 					watchcode, rp->r_pc);
715 				fault_type = F_INVAL;
716 			} else {
717 				bzero(&siginfo, sizeof (siginfo));
718 				siginfo.si_signo = SIGTRAP;
719 				siginfo.si_code = watchcode;
720 				siginfo.si_addr = vaddr;
721 				siginfo.si_trapafter = 0;
722 				siginfo.si_pc = (caddr_t)rp->r_pc;
723 				fault = FLTWATCH;
724 				break;
725 			}
726 		} else {
727 			/* XXX pr_watch_emul() never succeeds (for now) */
728 			if (rw != S_EXEC && pr_watch_emul(rp, vaddr, rw))
729 				goto out;
730 			do_watch_step(vaddr, sz, rw, 0, 0);
731 			fault_type = F_INVAL;
732 		}
733 
734 		res = pagefault(addr, fault_type, rw, 0);
735 
736 		/*
737 		 * If pagefault() succeeded, ok.
738 		 * Otherwise attempt to grow the stack.
739 		 */
740 		if (res == 0 ||
741 		    (res == FC_NOMAP &&
742 		    addr < p->p_usrstack &&
743 		    grow(addr))) {
744 			lwp->lwp_lastfault = FLTPAGE;
745 			lwp->lwp_lastfaddr = addr;
746 			if (prismember(&p->p_fltmask, FLTPAGE)) {
747 				bzero(&siginfo, sizeof (siginfo));
748 				siginfo.si_addr = addr;
749 				(void) stop_on_fault(FLTPAGE, &siginfo);
750 			}
751 			goto out;
752 		} else if (res == FC_PROT && addr < p->p_usrstack &&
753 		    (mmu.pt_nx != 0 && (errcode & PF_ERR_EXEC))) {
754 			report_stack_exec(p, addr);
755 		}
756 
757 #ifdef OPTERON_ERRATUM_91
758 		/*
759 		 * Workaround for Opteron Erratum 91. Prefetches may generate a
760 		 * page fault (they're not supposed to do that!). If this
761 		 * occurs we simply return back to the instruction.
762 		 *
763 		 * We rely on copyin to properly fault in the page with r_pc.
764 		 */
765 		if (opteron_erratum_91 &&
766 		    addr != (caddr_t)rp->r_pc &&
767 		    instr_is_prefetch((caddr_t)rp->r_pc)) {
768 #ifdef DEBUG
769 			cmn_err(CE_WARN, "Opteron erratum 91 occurred: "
770 			    "prefetch at %p in pid %d generated a trap!",
771 			    (void *)rp->r_pc, p->p_pid);
772 #endif /* DEBUG */
773 			goto out;
774 		}
775 #endif /* OPTERON_ERRATUM_91 */
776 
777 		if (tudebug)
778 			showregs(type, rp, addr);
779 		/*
780 		 * In the case where both pagefault and grow fail,
781 		 * set the code to the value provided by pagefault.
782 		 * We map all errors returned from pagefault() to SIGSEGV.
783 		 */
784 		bzero(&siginfo, sizeof (siginfo));
785 		siginfo.si_addr = addr;
786 		switch (FC_CODE(res)) {
787 		case FC_HWERR:
788 		case FC_NOSUPPORT:
789 			siginfo.si_signo = SIGBUS;
790 			siginfo.si_code = BUS_ADRERR;
791 			fault = FLTACCESS;
792 			break;
793 		case FC_ALIGN:
794 			siginfo.si_signo = SIGBUS;
795 			siginfo.si_code = BUS_ADRALN;
796 			fault = FLTACCESS;
797 			break;
798 		case FC_OBJERR:
799 			if ((siginfo.si_errno = FC_ERRNO(res)) != EINTR) {
800 				siginfo.si_signo = SIGBUS;
801 				siginfo.si_code = BUS_OBJERR;
802 				fault = FLTACCESS;
803 			}
804 			break;
805 		default:	/* FC_NOMAP or FC_PROT */
806 			siginfo.si_signo = SIGSEGV;
807 			siginfo.si_code =
808 			    (res == FC_NOMAP)? SEGV_MAPERR : SEGV_ACCERR;
809 			fault = FLTBOUNDS;
810 			break;
811 		}
812 		break;
813 
814 	case T_ILLINST + USER:	/* invalid opcode fault */
815 		/*
816 		 * If the syscall instruction is disabled due to LDT usage, a
817 		 * user program that attempts to execute it will trigger a #ud
818 		 * trap. Check for that case here. If this occurs on a CPU which
819 		 * doesn't even support syscall, the result of all of this will
820 		 * be to emulate that particular instruction.
821 		 */
822 		if (p->p_ldt != NULL &&
823 		    instr_is_fast_syscall((caddr_t)rp->r_pc, X86_ASYSC)) {
824 			if (rewrite_syscall((caddr_t)rp->r_pc) == 0)
825 				goto out;
826 #ifdef DEBUG
827 			else
828 				cmn_err(CE_WARN, "failed to rewrite syscall "
829 				    "instruction in process %d",
830 				    curthread->t_procp->p_pid);
831 #endif /* DEBUG */
832 		}
833 
834 #ifdef __amd64
835 		/*
836 		 * Emulate the LAHF and SAHF instructions if needed.
837 		 * See the instr_is_lsahf function for details.
838 		 */
839 		if (p->p_model == DATAMODEL_LP64 &&
840 		    instr_is_lsahf((caddr_t)rp->r_pc, &instr)) {
841 			emulate_lsahf(rp, instr);
842 			goto out;
843 		}
844 #endif
845 
846 		/*FALLTHROUGH*/
847 
848 		if (tudebug)
849 			showregs(type, rp, (caddr_t)0);
850 		siginfo.si_signo = SIGILL;
851 		siginfo.si_code  = ILL_ILLOPC;
852 		siginfo.si_addr  = (caddr_t)rp->r_pc;
853 		fault = FLTILL;
854 		break;
855 
856 	case T_ZERODIV + USER:		/* integer divide by zero */
857 		if (tudebug && tudebugfpe)
858 			showregs(type, rp, (caddr_t)0);
859 		siginfo.si_signo = SIGFPE;
860 		siginfo.si_code  = FPE_INTDIV;
861 		siginfo.si_addr  = (caddr_t)rp->r_pc;
862 		fault = FLTIZDIV;
863 		break;
864 
865 	case T_OVFLW + USER:	/* integer overflow */
866 		if (tudebug && tudebugfpe)
867 			showregs(type, rp, (caddr_t)0);
868 		siginfo.si_signo = SIGFPE;
869 		siginfo.si_code  = FPE_INTOVF;
870 		siginfo.si_addr  = (caddr_t)rp->r_pc;
871 		fault = FLTIOVF;
872 		break;
873 
874 	case T_NOEXTFLT + USER:	/* math coprocessor not available */
875 		if (tudebug && tudebugfpe)
876 			showregs(type, rp, addr);
877 		if (fpnoextflt(rp)) {
878 			siginfo.si_signo = SIGFPE;
879 			siginfo.si_code  = ILL_ILLOPC;
880 			siginfo.si_addr  = (caddr_t)rp->r_pc;
881 			fault = FLTFPE;
882 		}
883 		break;
884 
885 	case T_EXTOVRFLT:	/* extension overrun fault */
886 		/* check if we took a kernel trap on behalf of user */
887 		{
888 			extern  void ndptrap_frstor(void);
889 			if (rp->r_pc != (uintptr_t)ndptrap_frstor)
890 				(void) die(type, rp, addr, cpuid);
891 			type |= USER;
892 		}
893 		/*FALLTHROUGH*/
894 	case T_EXTOVRFLT + USER:	/* extension overrun fault */
895 		if (tudebug && tudebugfpe)
896 			showregs(type, rp, addr);
897 		if (fpextovrflt(rp)) {
898 			siginfo.si_signo = SIGSEGV;
899 			siginfo.si_code  = SEGV_MAPERR;
900 			siginfo.si_addr  = (caddr_t)rp->r_pc;
901 			fault = FLTBOUNDS;
902 		}
903 		break;
904 
905 	case T_EXTERRFLT:	/* x87 floating point exception pending */
906 		/* check if we took a kernel trap on behalf of user */
907 		{
908 			extern  void ndptrap_frstor(void);
909 			if (rp->r_pc != (uintptr_t)ndptrap_frstor)
910 				(void) die(type, rp, addr, cpuid);
911 			type |= USER;
912 		}
913 		/*FALLTHROUGH*/
914 
915 	case T_EXTERRFLT + USER: /* x87 floating point exception pending */
916 		if (tudebug && tudebugfpe)
917 			showregs(type, rp, addr);
918 		if (sicode = fpexterrflt(rp)) {
919 			siginfo.si_signo = SIGFPE;
920 			siginfo.si_code  = sicode;
921 			siginfo.si_addr  = (caddr_t)rp->r_pc;
922 			fault = FLTFPE;
923 		}
924 		break;
925 
926 	case T_SIMDFPE + USER:		/* SSE and SSE2 exceptions */
927 		if (tudebug && tudebugsse)
928 			showregs(type, rp, addr);
929 		if ((x86_feature & (X86_SSE|X86_SSE2)) == 0) {
930 			/*
931 			 * There are rumours that some user instructions
932 			 * on older CPUs can cause this trap to occur; in
933 			 * which case send a SIGILL instead of a SIGFPE.
934 			 */
935 			siginfo.si_signo = SIGILL;
936 			siginfo.si_code  = ILL_ILLTRP;
937 			siginfo.si_addr  = (caddr_t)rp->r_pc;
938 			siginfo.si_trapno = type & ~USER;
939 			fault = FLTILL;
940 		} else if ((sicode = fpsimderrflt(rp)) != 0) {
941 			siginfo.si_signo = SIGFPE;
942 			siginfo.si_code = sicode;
943 			siginfo.si_addr = (caddr_t)rp->r_pc;
944 			fault = FLTFPE;
945 		}
946 		break;
947 
948 	case T_BPTFLT:	/* breakpoint trap */
949 		/*
950 		 * Kernel breakpoint traps should only happen when kmdb is
951 		 * active, and even then, it'll have interposed on the IDT, so
952 		 * control won't get here.  If it does, we've hit a breakpoint
953 		 * without the debugger, which is very strange, and very
954 		 * fatal.
955 		 */
956 		if (tudebug && tudebugbpt)
957 			showregs(type, rp, (caddr_t)0);
958 
959 		(void) die(type, rp, addr, cpuid);
960 		break;
961 
962 	case T_SGLSTP: /* single step/hw breakpoint exception */
963 		if (tudebug && tudebugbpt)
964 			showregs(type, rp, (caddr_t)0);
965 
966 		/* Now evaluate how we got here */
967 		if (lwp != NULL && (lwp->lwp_pcb.pcb_drstat & DR_SINGLESTEP)) {
968 			/*
969 			 * i386 single-steps even through lcalls which
970 			 * change the privilege level. So we take a trap at
971 			 * the first instruction in privileged mode.
972 			 *
973 			 * Set a flag to indicate that upon completion of
974 			 * the system call, deal with the single-step trap.
975 			 *
976 			 * The same thing happens for sysenter, too.
977 			 */
978 #if defined(__amd64)
979 			if (rp->r_pc == (uintptr_t)sys_sysenter) {
980 				/*
981 				 * Adjust the pc so that we don't execute the
982 				 * swapgs instruction at the head of the
983 				 * handler and completely confuse things.
984 				 */
985 				rp->r_pc = (uintptr_t)
986 				    _sys_sysenter_post_swapgs;
987 #elif defined(__i386)
988 			if (rp->r_pc == (uintptr_t)sys_call ||
989 			    rp->r_pc == (uintptr_t)sys_sysenter) {
990 #endif
991 				rp->r_ps &= ~PS_T; /* turn off trace */
992 				lwp->lwp_pcb.pcb_flags |= DEBUG_PENDING;
993 				cur_thread->t_post_sys = 1;
994 				aston(curthread);
995 				goto cleanup;
996 			}
997 		}
998 		/* XXX - needs review on debugger interface? */
999 		if (boothowto & RB_DEBUG)
1000 			debug_enter((char *)NULL);
1001 		else
1002 			(void) die(type, rp, addr, cpuid);
1003 		break;
1004 
1005 	case T_NMIFLT:	/* NMI interrupt */
1006 		printf("Unexpected NMI in system mode\n");
1007 		goto cleanup;
1008 
1009 	case T_NMIFLT + USER:	/* NMI interrupt */
1010 		printf("Unexpected NMI in user mode\n");
1011 		break;
1012 
1013 	case T_GPFLT:	/* general protection violation */
1014 #if defined(__amd64)
1015 		/*
1016 		 * On amd64, we can get a #gp from referencing addresses
1017 		 * in the virtual address hole e.g. from a copyin.
1018 		 */
1019 
1020 		/*
1021 		 * If we're under on_trap() protection (see <sys/ontrap.h>),
1022 		 * set ot_trap and longjmp back to the on_trap() call site.
1023 		 */
1024 		if ((cur_thread->t_ontrap != NULL) &&
1025 		    (cur_thread->t_ontrap->ot_prot & OT_DATA_ACCESS)) {
1026 			curthread->t_ontrap->ot_trap |= OT_DATA_ACCESS;
1027 			longjmp(&curthread->t_ontrap->ot_jmpbuf);
1028 		}
1029 
1030 		/*
1031 		 * If we're under lofault protection (copyin etc.),
1032 		 * longjmp back to lofault with an EFAULT.
1033 		 */
1034 		if (cur_thread->t_lofault) {
1035 			/*
1036 			 * Fault is not resolvable, so just return to lofault
1037 			 */
1038 			if (lodebug) {
1039 				showregs(type, rp, addr);
1040 				traceregs(rp);
1041 			}
1042 			rp->r_r0 = EFAULT;
1043 			rp->r_pc = cur_thread->t_lofault;
1044 			goto cleanup;
1045 		}
1046 		/*FALLTHROUGH*/
1047 #endif
1048 	case T_STKFLT:	/* stack fault */
1049 	case T_TSSFLT:	/* invalid TSS fault */
1050 	case T_SEGFLT:	/* segment not present fault */
1051 		if (tudebug)
1052 			showregs(type, rp, (caddr_t)0);
1053 		if (kern_gpfault(rp))
1054 			(void) die(type, rp, addr, cpuid);
1055 		goto cleanup;
1056 		/*FALLTHROUGH*/
1057 
1058 /*
1059  * ONLY 32-bit PROCESSES can USE a PRIVATE LDT! 64-bit apps should have
1060  * no legacy need for them, so we put a stop to it here.
1061  *
1062  * So: not-present fault is ONLY valid for 32-bit processes with a private LDT
1063  * trying to do a system call. Emulate it.
1064  *
1065  * #gp fault is ONLY valid for 32-bit processes also, which DO NOT have private
1066  * LDT, and are trying to do a system call. Emulate it.
1067  */
1068 	case T_SEGFLT + USER:	/* segment not present fault */
1069 	case T_GPFLT + USER:	/* general protection violation */
1070 #ifdef _SYSCALL32_IMPL
1071 		if (p->p_model != DATAMODEL_NATIVE) {
1072 #endif /* _SYSCALL32_IMPL */
1073 		if (instr_is_syscall((caddr_t)rp->r_pc)) {
1074 			if (type == T_SEGFLT + USER)
1075 				ASSERT(p->p_ldt != NULL);
1076 
1077 			if ((p->p_ldt == NULL && type == T_GPFLT + USER) ||
1078 			    type == T_SEGFLT + USER) {
1079 
1080 			/*
1081 			 * The user attempted a system call via the obsolete
1082 			 * call gate mechanism. Because the process doesn't have
1083 			 * an LDT (i.e. the ldtr contains 0), a #gp results.
1084 			 * Emulate the syscall here, just as we do above for a
1085 			 * #np trap.
1086 			 */
1087 
1088 			/*
1089 			 * Since this is a not-present trap, rp->r_pc points to
1090 			 * the trapping lcall instruction. We need to bump it
1091 			 * to the next insn so the app can continue on.
1092 			 */
1093 			rp->r_pc += LCALLSIZE;
1094 			lwp->lwp_regs = rp;
1095 
1096 			/*
1097 			 * Normally the microstate of the LWP is forced back to
1098 			 * LMS_USER by the syscall handlers. Emulate that
1099 			 * behavior here.
1100 			 */
1101 			mstate = LMS_USER;
1102 
1103 			dosyscall();
1104 			goto out;
1105 			}
1106 		}
1107 #ifdef _SYSCALL32_IMPL
1108 		}
1109 #endif /* _SYSCALL32_IMPL */
1110 		/*
1111 		 * If the current process is using a private LDT and the
1112 		 * trapping instruction is sysenter, the sysenter instruction
1113 		 * has been disabled on the CPU because it destroys segment
1114 		 * registers. If this is the case, rewrite the instruction to
1115 		 * be a safe system call and retry it. If this occurs on a CPU
1116 		 * which doesn't even support sysenter, the result of all of
1117 		 * this will be to emulate that particular instruction.
1118 		 */
1119 		if (p->p_ldt != NULL &&
1120 		    instr_is_fast_syscall((caddr_t)rp->r_pc, X86_SEP)) {
1121 			if (rewrite_syscall((caddr_t)rp->r_pc) == 0)
1122 				goto out;
1123 #ifdef DEBUG
1124 			else
1125 				cmn_err(CE_WARN, "failed to rewrite sysenter "
1126 				    "instruction in process %d",
1127 				    curthread->t_procp->p_pid);
1128 #endif /* DEBUG */
1129 		}
1130 		/*FALLTHROUGH*/
1131 
1132 	case T_BOUNDFLT + USER:	/* bound fault */
1133 	case T_STKFLT + USER:	/* stack fault */
1134 	case T_TSSFLT + USER:	/* invalid TSS fault */
1135 		if (tudebug)
1136 			showregs(type, rp, (caddr_t)0);
1137 		siginfo.si_signo = SIGSEGV;
1138 		siginfo.si_code  = SEGV_MAPERR;
1139 		siginfo.si_addr  = (caddr_t)rp->r_pc;
1140 		fault = FLTBOUNDS;
1141 		break;
1142 
1143 	case T_ALIGNMENT + USER:	/* user alignment error (486) */
1144 		if (tudebug)
1145 			showregs(type, rp, (caddr_t)0);
1146 		bzero(&siginfo, sizeof (siginfo));
1147 		siginfo.si_signo = SIGBUS;
1148 		siginfo.si_code = BUS_ADRALN;
1149 		siginfo.si_addr = (caddr_t)rp->r_pc;
1150 		fault = FLTACCESS;
1151 		break;
1152 
1153 	case T_SGLSTP + USER: /* single step/hw breakpoint exception */
1154 		if (tudebug && tudebugbpt)
1155 			showregs(type, rp, (caddr_t)0);
1156 
1157 		/* Was it single-stepping? */
1158 		if (lwp->lwp_pcb.pcb_drstat & DR_SINGLESTEP) {
1159 			pcb_t *pcb = &lwp->lwp_pcb;
1160 
1161 			rp->r_ps &= ~PS_T;
1162 			/*
1163 			 * If both NORMAL_STEP and WATCH_STEP are in effect,
1164 			 * give precedence to NORMAL_STEP.  If neither is set,
1165 			 * user must have set the PS_T bit in %efl; treat this
1166 			 * as NORMAL_STEP.
1167 			 */
1168 			if ((pcb->pcb_flags & NORMAL_STEP) ||
1169 			    !(pcb->pcb_flags & WATCH_STEP)) {
1170 				siginfo.si_signo = SIGTRAP;
1171 				siginfo.si_code = TRAP_TRACE;
1172 				siginfo.si_addr = (caddr_t)rp->r_pc;
1173 				fault = FLTTRACE;
1174 				if (pcb->pcb_flags & WATCH_STEP)
1175 					(void) undo_watch_step(NULL);
1176 			} else {
1177 				fault = undo_watch_step(&siginfo);
1178 			}
1179 			pcb->pcb_flags &= ~(NORMAL_STEP|WATCH_STEP);
1180 		} else {
1181 			cmn_err(CE_WARN,
1182 			    "Unexpected INT 1 in user mode, dr6=%lx",
1183 			    lwp->lwp_pcb.pcb_drstat);
1184 		}
1185 		break;
1186 
1187 	case T_BPTFLT + USER:	/* breakpoint trap */
1188 		if (tudebug && tudebugbpt)
1189 			showregs(type, rp, (caddr_t)0);
1190 		/*
1191 		 * int 3 (the breakpoint instruction) leaves the pc referring
1192 		 * to the address one byte after the breakpointed address.
1193 		 * If the P_PR_BPTADJ flag has been set via /proc, We adjust
1194 		 * it back so it refers to the breakpointed address.
1195 		 */
1196 		if (p->p_proc_flag & P_PR_BPTADJ)
1197 			rp->r_pc--;
1198 		siginfo.si_signo = SIGTRAP;
1199 		siginfo.si_code  = TRAP_BRKPT;
1200 		siginfo.si_addr  = (caddr_t)rp->r_pc;
1201 		fault = FLTBPT;
1202 		break;
1203 
1204 	case T_AST:
1205 		/*
1206 		 * This occurs only after the cs register has been made to
1207 		 * look like a kernel selector, either through debugging or
1208 		 * possibly by functions like setcontext().  The thread is
1209 		 * about to cause a general protection fault at common_iret()
1210 		 * in locore.  We let that happen immediately instead of
1211 		 * doing the T_AST processing.
1212 		 */
1213 		goto cleanup;
1214 
1215 	case T_AST + USER:		/* profiling or resched pseudo trap */
1216 		if (lwp->lwp_pcb.pcb_flags & CPC_OVERFLOW) {
1217 			lwp->lwp_pcb.pcb_flags &= ~CPC_OVERFLOW;
1218 			if (kcpc_overflow_ast()) {
1219 				/*
1220 				 * Signal performance counter overflow
1221 				 */
1222 				if (tudebug)
1223 					showregs(type, rp, (caddr_t)0);
1224 				bzero(&siginfo, sizeof (siginfo));
1225 				siginfo.si_signo = SIGEMT;
1226 				siginfo.si_code = EMT_CPCOVF;
1227 				siginfo.si_addr = (caddr_t)rp->r_pc;
1228 				fault = FLTCPCOVF;
1229 			}
1230 		}
1231 		break;
1232 	}
1233 
1234 	/*
1235 	 * We can't get here from a system trap
1236 	 */
1237 	ASSERT(type & USER);
1238 
1239 	if (fault) {
1240 		/*
1241 		 * Remember the fault and fault adddress
1242 		 * for real-time (SIGPROF) profiling.
1243 		 */
1244 		lwp->lwp_lastfault = fault;
1245 		lwp->lwp_lastfaddr = siginfo.si_addr;
1246 
1247 		DTRACE_PROC2(fault, int, fault, ksiginfo_t *, &siginfo);
1248 
1249 		/*
1250 		 * If a debugger has declared this fault to be an
1251 		 * event of interest, stop the lwp.  Otherwise just
1252 		 * deliver the associated signal.
1253 		 */
1254 		if (siginfo.si_signo != SIGKILL &&
1255 		    prismember(&p->p_fltmask, fault) &&
1256 		    stop_on_fault(fault, &siginfo) == 0)
1257 			siginfo.si_signo = 0;
1258 	}
1259 
1260 	if (siginfo.si_signo)
1261 		trapsig(&siginfo, (fault == FLTCPCOVF)? 0 : 1);
1262 
1263 	if (lwp->lwp_oweupc)
1264 		profil_tick(rp->r_pc);
1265 
1266 	if (cur_thread->t_astflag | cur_thread->t_sig_check) {
1267 		/*
1268 		 * Turn off the AST flag before checking all the conditions that
1269 		 * may have caused an AST.  This flag is on whenever a signal or
1270 		 * unusual condition should be handled after the next trap or
1271 		 * syscall.
1272 		 */
1273 		astoff(cur_thread);
1274 		/*
1275 		 * If a single-step trap occurred on a syscall (see above)
1276 		 * recognize it now.  Do this before checking for signals
1277 		 * because deferred_singlestep_trap() may generate a SIGTRAP to
1278 		 * the LWP or may otherwise mark the LWP to call issig(FORREAL).
1279 		 */
1280 		if (lwp->lwp_pcb.pcb_flags & DEBUG_PENDING)
1281 			deferred_singlestep_trap((caddr_t)rp->r_pc);
1282 
1283 		cur_thread->t_sig_check = 0;
1284 
1285 		mutex_enter(&p->p_lock);
1286 		if (curthread->t_proc_flag & TP_CHANGEBIND) {
1287 			timer_lwpbind();
1288 			curthread->t_proc_flag &= ~TP_CHANGEBIND;
1289 		}
1290 		mutex_exit(&p->p_lock);
1291 
1292 		/*
1293 		 * for kaio requests that are on the per-process poll queue,
1294 		 * aiop->aio_pollq, they're AIO_POLL bit is set, the kernel
1295 		 * should copyout their result_t to user memory. by copying
1296 		 * out the result_t, the user can poll on memory waiting
1297 		 * for the kaio request to complete.
1298 		 */
1299 		if (p->p_aio)
1300 			aio_cleanup(0);
1301 		/*
1302 		 * If this LWP was asked to hold, call holdlwp(), which will
1303 		 * stop.  holdlwps() sets this up and calls pokelwps() which
1304 		 * sets the AST flag.
1305 		 *
1306 		 * Also check TP_EXITLWP, since this is used by fresh new LWPs
1307 		 * through lwp_rtt().  That flag is set if the lwp_create(2)
1308 		 * syscall failed after creating the LWP.
1309 		 */
1310 		if (ISHOLD(p))
1311 			holdlwp();
1312 
1313 		/*
1314 		 * All code that sets signals and makes ISSIG evaluate true must
1315 		 * set t_astflag afterwards.
1316 		 */
1317 		if (ISSIG_PENDING(cur_thread, lwp, p)) {
1318 			if (issig(FORREAL))
1319 				psig();
1320 			cur_thread->t_sig_check = 1;
1321 		}
1322 
1323 		if (cur_thread->t_rprof != NULL) {
1324 			realsigprof(0, 0);
1325 			cur_thread->t_sig_check = 1;
1326 		}
1327 
1328 		/*
1329 		 * /proc can't enable/disable the trace bit itself
1330 		 * because that could race with the call gate used by
1331 		 * system calls via "lcall". If that happened, an
1332 		 * invalid EFLAGS would result. prstep()/prnostep()
1333 		 * therefore schedule an AST for the purpose.
1334 		 */
1335 		if (lwp->lwp_pcb.pcb_flags & REQUEST_STEP) {
1336 			lwp->lwp_pcb.pcb_flags &= ~REQUEST_STEP;
1337 			rp->r_ps |= PS_T;
1338 		}
1339 		if (lwp->lwp_pcb.pcb_flags & REQUEST_NOSTEP) {
1340 			lwp->lwp_pcb.pcb_flags &= ~REQUEST_NOSTEP;
1341 			rp->r_ps &= ~PS_T;
1342 		}
1343 	}
1344 
1345 out:	/* We can't get here from a system trap */
1346 	ASSERT(type & USER);
1347 
1348 	if (ISHOLD(p))
1349 		holdlwp();
1350 
1351 	/*
1352 	 * Set state to LWP_USER here so preempt won't give us a kernel
1353 	 * priority if it occurs after this point.  Call CL_TRAPRET() to
1354 	 * restore the user-level priority.
1355 	 *
1356 	 * It is important that no locks (other than spinlocks) be entered
1357 	 * after this point before returning to user mode (unless lwp_state
1358 	 * is set back to LWP_SYS).
1359 	 */
1360 	lwp->lwp_state = LWP_USER;
1361 
1362 	if (cur_thread->t_trapret) {
1363 		cur_thread->t_trapret = 0;
1364 		thread_lock(cur_thread);
1365 		CL_TRAPRET(cur_thread);
1366 		thread_unlock(cur_thread);
1367 	}
1368 	if (CPU->cpu_runrun)
1369 		preempt();
1370 	(void) new_mstate(cur_thread, mstate);
1371 
1372 	/* Kernel probe */
1373 	TNF_PROBE_1(thread_state, "thread", /* CSTYLED */,
1374 	    tnf_microstate, state, LMS_USER);
1375 
1376 	return;
1377 
1378 cleanup:	/* system traps end up here */
1379 	ASSERT(!(type & USER));
1380 }
1381 
1382 /*
1383  * Patch non-zero to disable preemption of threads in the kernel.
1384  */
1385 int IGNORE_KERNEL_PREEMPTION = 0;	/* XXX - delete this someday */
1386 
1387 struct kpreempt_cnts {		/* kernel preemption statistics */
1388 	int	kpc_idle;	/* executing idle thread */
1389 	int	kpc_intr;	/* executing interrupt thread */
1390 	int	kpc_clock;	/* executing clock thread */
1391 	int	kpc_blocked;	/* thread has blocked preemption (t_preempt) */
1392 	int	kpc_notonproc;	/* thread is surrendering processor */
1393 	int	kpc_inswtch;	/* thread has ratified scheduling decision */
1394 	int	kpc_prilevel;	/* processor interrupt level is too high */
1395 	int	kpc_apreempt;	/* asynchronous preemption */
1396 	int	kpc_spreempt;	/* synchronous preemption */
1397 } kpreempt_cnts;
1398 
1399 /*
1400  * kernel preemption: forced rescheduling, preempt the running kernel thread.
1401  *	the argument is old PIL for an interrupt,
1402  *	or the distingished value KPREEMPT_SYNC.
1403  */
1404 void
1405 kpreempt(int asyncspl)
1406 {
1407 	kthread_t *cur_thread = curthread;
1408 
1409 	if (IGNORE_KERNEL_PREEMPTION) {
1410 		aston(CPU->cpu_dispthread);
1411 		return;
1412 	}
1413 
1414 	/*
1415 	 * Check that conditions are right for kernel preemption
1416 	 */
1417 	do {
1418 		if (cur_thread->t_preempt) {
1419 			/*
1420 			 * either a privileged thread (idle, panic, interrupt)
1421 			 *	or will check when t_preempt is lowered
1422 			 */
1423 			if (cur_thread->t_pri < 0)
1424 				kpreempt_cnts.kpc_idle++;
1425 			else if (cur_thread->t_flag & T_INTR_THREAD) {
1426 				kpreempt_cnts.kpc_intr++;
1427 				if (cur_thread->t_pil == CLOCK_LEVEL)
1428 					kpreempt_cnts.kpc_clock++;
1429 			} else
1430 				kpreempt_cnts.kpc_blocked++;
1431 			aston(CPU->cpu_dispthread);
1432 			return;
1433 		}
1434 		if (cur_thread->t_state != TS_ONPROC ||
1435 		    cur_thread->t_disp_queue != CPU->cpu_disp) {
1436 			/* this thread will be calling swtch() shortly */
1437 			kpreempt_cnts.kpc_notonproc++;
1438 			if (CPU->cpu_thread != CPU->cpu_dispthread) {
1439 				/* already in swtch(), force another */
1440 				kpreempt_cnts.kpc_inswtch++;
1441 				siron();
1442 			}
1443 			return;
1444 		}
1445 		if (getpil() >= DISP_LEVEL) {
1446 			/*
1447 			 * We can't preempt this thread if it is at
1448 			 * a PIL >= DISP_LEVEL since it may be holding
1449 			 * a spin lock (like sched_lock).
1450 			 */
1451 			siron();	/* check back later */
1452 			kpreempt_cnts.kpc_prilevel++;
1453 			return;
1454 		}
1455 
1456 		if (asyncspl != KPREEMPT_SYNC)
1457 			kpreempt_cnts.kpc_apreempt++;
1458 		else
1459 			kpreempt_cnts.kpc_spreempt++;
1460 
1461 		cur_thread->t_preempt++;
1462 		preempt();
1463 		cur_thread->t_preempt--;
1464 	} while (CPU->cpu_kprunrun);
1465 }
1466 
1467 /*
1468  * Print out debugging info.
1469  */
1470 static void
1471 showregs(uint_t type, struct regs *rp, caddr_t addr)
1472 {
1473 	int s;
1474 
1475 	s = spl7();
1476 	type &= ~USER;
1477 	if (u.u_comm[0])
1478 		printf("%s: ", u.u_comm);
1479 	if (type < TRAP_TYPES)
1480 		printf("#%s %s\n", trap_type_mnemonic[type], trap_type[type]);
1481 	else
1482 		switch (type) {
1483 		case T_SYSCALL:
1484 			printf("Syscall Trap:\n");
1485 			break;
1486 		case T_AST:
1487 			printf("AST\n");
1488 			break;
1489 		default:
1490 			printf("Bad Trap = %d\n", type);
1491 			break;
1492 		}
1493 	if (type == T_PGFLT) {
1494 		printf("Bad %s fault at addr=0x%lx\n",
1495 		    USERMODE(rp->r_cs) ? "user": "kernel", (uintptr_t)addr);
1496 	} else if (addr) {
1497 		printf("addr=0x%lx\n", (uintptr_t)addr);
1498 	}
1499 
1500 	printf("pid=%d, pc=0x%lx, sp=0x%lx, eflags=0x%lx\n",
1501 	    (ttoproc(curthread) && ttoproc(curthread)->p_pidp) ?
1502 	    ttoproc(curthread)->p_pid : 0, rp->r_pc, rp->r_sp, rp->r_ps);
1503 
1504 #if defined(__lint)
1505 	/*
1506 	 * this clause can be deleted when lint bug 4870403 is fixed
1507 	 * (lint thinks that bit 32 is illegal in a %b format string)
1508 	 */
1509 	printf("cr0: %x cr4: %b\n",
1510 	    (uint_t)getcr0(), (uint_t)getcr4(), FMT_CR4);
1511 #else
1512 	printf("cr0: %b cr4: %b\n",
1513 	    (uint_t)getcr0(), FMT_CR0, (uint_t)getcr4(), FMT_CR4);
1514 #endif
1515 
1516 #if defined(__amd64)
1517 	printf("cr2: %lx cr3: %lx cr8: %lx\n", getcr2(), getcr3(), getcr8());
1518 #elif defined(__i386)
1519 	printf("cr2: %lx cr3: %lx\n", getcr2(), getcr3());
1520 #endif
1521 
1522 	dumpregs(rp);
1523 	splx(s);
1524 }
1525 
1526 static void
1527 dumpregs(struct regs *rp)
1528 {
1529 #if defined(__amd64)
1530 	const char fmt[] = "\t%3s: %16lx %3s: %16lx %3s: %16lx\n";
1531 
1532 	printf(fmt, "rdi", rp->r_rdi, "rsi", rp->r_rsi, "rdx", rp->r_rdx);
1533 	printf(fmt, "rcx", rp->r_rcx, " r8", rp->r_r8, " r9", rp->r_r9);
1534 	printf(fmt, "rax", rp->r_rax, "rbx", rp->r_rbx, "rbp", rp->r_rbp);
1535 	printf(fmt, "r10", rp->r_r10, "r11", rp->r_r11, "r12", rp->r_r12);
1536 	printf(fmt, "r13", rp->r_r13, "r14", rp->r_r14, "r15", rp->r_r15);
1537 
1538 	printf(fmt, "fsb", rp->r_fsbase, "gsb", rp->r_gsbase, " ds", rp->r_ds);
1539 	printf(fmt, " es", rp->r_es, " fs", rp->r_fs, " gs", rp->r_gs);
1540 
1541 	printf(fmt, "trp", rp->r_trapno, "err", rp->r_err, "rip", rp->r_rip);
1542 	printf(fmt, " cs", rp->r_cs, "rfl", rp->r_rfl, "rsp", rp->r_rsp);
1543 
1544 	printf("\t%3s: %16lx\n", " ss", rp->r_ss);
1545 
1546 #elif defined(__i386)
1547 	const char fmt[] = "\t%3s: %8lx %3s: %8lx %3s: %8lx %3s: %8lx\n";
1548 
1549 	printf(fmt, " gs", rp->r_gs, " fs", rp->r_fs,
1550 	    " es", rp->r_es, " ds", rp->r_ds);
1551 	printf(fmt, "edi", rp->r_edi, "esi", rp->r_esi,
1552 	    "ebp", rp->r_ebp, "esp", rp->r_esp);
1553 	printf(fmt, "ebx", rp->r_ebx, "edx", rp->r_edx,
1554 	    "ecx", rp->r_ecx, "eax", rp->r_eax);
1555 	printf(fmt, "trp", rp->r_trapno, "err", rp->r_err,
1556 	    "eip", rp->r_eip, " cs", rp->r_cs);
1557 	printf("\t%3s: %8lx %3s: %8lx %3s: %8lx\n",
1558 	    "efl", rp->r_efl, "usp", rp->r_uesp, " ss", rp->r_ss);
1559 
1560 #endif	/* __i386 */
1561 }
1562 
1563 /*
1564  * Handle #gp faults in kernel mode.
1565  *
1566  * One legitimate way this can happen is if we attempt to update segment
1567  * registers to naughty values on the way out of the kernel.
1568  *
1569  * This can happen in a couple of ways: someone - either accidentally or
1570  * on purpose - creates (setcontext(2), lwp_create(2)) or modifies
1571  * (signal(2)) a ucontext that contains silly segment register values.
1572  * Or someone - either accidentally or on purpose - modifies the prgregset_t
1573  * of a subject process via /proc to contain silly segment register values.
1574  *
1575  * (The unfortunate part is that we can end up discovering the bad segment
1576  * register value in the middle of an 'iret' after we've popped most of the
1577  * stack.  So it becomes quite difficult to associate an accurate ucontext
1578  * with the lwp, because the act of taking the #gp trap overwrites most of
1579  * what we were going to send the lwp.)
1580  *
1581  * OTOH if it turns out that's -not- the problem, and we're -not- an lwp
1582  * trying to return to user mode and we get a #gp fault, then we need
1583  * to die() -- which will happen if we return non-zero from this routine.
1584  */
1585 static int
1586 kern_gpfault(struct regs *rp)
1587 {
1588 	kthread_t *t = curthread;
1589 	proc_t *p = ttoproc(t);
1590 	klwp_t *lwp = ttolwp(t);
1591 	struct regs tmpregs, *trp = NULL;
1592 	caddr_t pc = (caddr_t)rp->r_pc;
1593 	int v;
1594 
1595 	extern void _sys_rtt(), sr_sup();
1596 
1597 #if defined(__amd64)
1598 	extern void _update_sregs(), _update_sregs_done();
1599 	static const uint8_t iretq_insn[2] = { 0x48, 0xcf };
1600 
1601 #elif defined(__i386)
1602 	static const uint8_t iret_insn[1] = { 0xcf };
1603 
1604 	/*
1605 	 * Note carefully the appallingly awful dependency between
1606 	 * the instruction sequence used in __SEGREGS_POP and these
1607 	 * instructions encoded here.
1608 	 *
1609 	 * XX64	Add some commentary to locore.s/privregs.h to document this.
1610 	 */
1611 	static const uint8_t movw_0_esp_gs[4] = { 0x8e, 0x6c, 0x24, 0x0 };
1612 	static const uint8_t movw_4_esp_fs[4] = { 0x8e, 0x64, 0x24, 0x4 };
1613 	static const uint8_t movw_8_esp_es[4] = { 0x8e, 0x44, 0x24, 0x8 };
1614 	static const uint8_t movw_c_esp_ds[4] = { 0x8e, 0x5c, 0x24, 0xc };
1615 #endif
1616 	/*
1617 	 * if we're not an lwp, or the pc range is outside _sys_rtt, then
1618 	 * we should immediately be die()ing horribly
1619 	 */
1620 	if (lwp == NULL ||
1621 	    (uintptr_t)pc < (uintptr_t)_sys_rtt ||
1622 	    (uintptr_t)pc > (uintptr_t)sr_sup)
1623 		return (1);
1624 
1625 	/*
1626 	 * So at least we're in the right part of the kernel.
1627 	 *
1628 	 * Disassemble the instruction at the faulting pc.
1629 	 * Once we know what it is, we carefully reconstruct the stack
1630 	 * based on the order in which the stack is deconstructed in
1631 	 * _sys_rtt. Ew.
1632 	 */
1633 
1634 #if defined(__amd64)
1635 
1636 	if (bcmp(pc, iretq_insn, sizeof (iretq_insn)) == 0) {
1637 		/*
1638 		 * We took the #gp while trying to perform the iretq.
1639 		 * This means that either %cs or %ss are bad.
1640 		 * All we know for sure is that most of the general
1641 		 * registers have been restored, including the
1642 		 * segment registers, and all we have left on the
1643 		 * topmost part of the lwp's stack are the
1644 		 * registers that the iretq was unable to consume.
1645 		 *
1646 		 * All the rest of the state was crushed by the #gp
1647 		 * which pushed -its- registers atop our old save area
1648 		 * (because we had to decrement the stack pointer, sigh) so
1649 		 * all that we can try and do is to reconstruct the
1650 		 * crushed frame from the #gp trap frame itself.
1651 		 */
1652 		trp = &tmpregs;
1653 		trp->r_ss = lwptoregs(lwp)->r_ss;
1654 		trp->r_sp = lwptoregs(lwp)->r_sp;
1655 		trp->r_ps = lwptoregs(lwp)->r_ps;
1656 		trp->r_cs = lwptoregs(lwp)->r_cs;
1657 		trp->r_pc = lwptoregs(lwp)->r_pc;
1658 		bcopy(rp, trp, offsetof(struct regs, r_pc));
1659 
1660 		/*
1661 		 * Validate simple math
1662 		 */
1663 		ASSERT(trp->r_pc == lwptoregs(lwp)->r_pc);
1664 		ASSERT(trp->r_err == rp->r_err);
1665 
1666 	} else if ((lwp->lwp_pcb.pcb_flags & RUPDATE_PENDING) != 0 &&
1667 	    pc >= (caddr_t)_update_sregs &&
1668 	    pc < (caddr_t)_update_sregs_done) {
1669 		/*
1670 		 * This is the common case -- we're trying to load
1671 		 * a bad segment register value in the only section
1672 		 * of kernel code that ever loads segment registers.
1673 		 *
1674 		 * We don't need to do anything at this point because
1675 		 * the pcb contains all the pending segment register
1676 		 * state, and the regs are still intact because we
1677 		 * didn't adjust the stack pointer yet.  Given the fidelity
1678 		 * of all this, we could conceivably send a signal
1679 		 * to the lwp, rather than core-ing.
1680 		 */
1681 		trp = lwptoregs(lwp);
1682 		ASSERT((caddr_t)trp == (caddr_t)rp->r_sp);
1683 	}
1684 
1685 #elif defined(__i386)
1686 
1687 	if (bcmp(pc, iret_insn, sizeof (iret_insn)) == 0) {
1688 		/*
1689 		 * We took the #gp while trying to perform the iret.
1690 		 * This means that either %cs or %ss are bad.
1691 		 * All we know for sure is that most of the general
1692 		 * registers have been restored, including the
1693 		 * segment registers, and all we have left on the
1694 		 * topmost part of the lwp's stack are the registers that
1695 		 * the iret was unable to consume.
1696 		 *
1697 		 * All the rest of the state was crushed by the #gp
1698 		 * which pushed -its- registers atop our old save area
1699 		 * (because we had to decrement the stack pointer, sigh) so
1700 		 * all that we can try and do is to reconstruct the
1701 		 * crushed frame from the #gp trap frame itself.
1702 		 */
1703 		trp = &tmpregs;
1704 		trp->r_ss = lwptoregs(lwp)->r_ss;
1705 		trp->r_sp = lwptoregs(lwp)->r_sp;
1706 		trp->r_ps = lwptoregs(lwp)->r_ps;
1707 		trp->r_cs = lwptoregs(lwp)->r_cs;
1708 		trp->r_pc = lwptoregs(lwp)->r_pc;
1709 		bcopy(rp, trp, offsetof(struct regs, r_pc));
1710 
1711 		ASSERT(trp->r_pc == lwptoregs(lwp)->r_pc);
1712 		ASSERT(trp->r_err == rp->r_err);
1713 
1714 	} else {
1715 		/*
1716 		 * Segment registers are reloaded in _sys_rtt
1717 		 * via the following sequence:
1718 		 *
1719 		 *	movw	0(%esp), %gs
1720 		 *	movw	4(%esp), %fs
1721 		 *	movw	8(%esp), %es
1722 		 *	movw	12(%esp), %ds
1723 		 *	addl	$16, %esp
1724 		 *
1725 		 * Thus if any of them fault, we know the user
1726 		 * registers are left unharmed on the stack.
1727 		 */
1728 		if (bcmp(pc, movw_0_esp_gs, sizeof (movw_0_esp_gs)) == 0 ||
1729 		    bcmp(pc, movw_4_esp_fs, sizeof (movw_4_esp_fs)) == 0 ||
1730 		    bcmp(pc, movw_8_esp_es, sizeof (movw_8_esp_es)) == 0 ||
1731 		    bcmp(pc, movw_c_esp_ds, sizeof (movw_c_esp_ds)) == 0)
1732 			trp = lwptoregs(lwp);
1733 	}
1734 #endif	/* __amd64 */
1735 
1736 	if (trp == NULL)
1737 		return (1);
1738 
1739 	/*
1740 	 * If we get to here, we're reasonably confident that we've
1741 	 * correctly decoded what happened on the way out of the kernel.
1742 	 * Rewrite the lwp's registers so that we can create a core dump
1743 	 * the (at least vaguely) represents the mcontext we were
1744 	 * being asked to restore when things went so terribly wrong.
1745 	 */
1746 
1747 	/*
1748 	 * Make sure that we have a meaningful %trapno and %err.
1749 	 */
1750 	trp->r_trapno = rp->r_trapno;
1751 	trp->r_err = rp->r_err;
1752 
1753 	if ((caddr_t)trp != (caddr_t)lwptoregs(lwp))
1754 		bcopy(trp, lwptoregs(lwp), sizeof (*trp));
1755 
1756 	mutex_enter(&p->p_lock);
1757 	lwp->lwp_cursig = SIGSEGV;
1758 	mutex_exit(&p->p_lock);
1759 
1760 	/*
1761 	 * Terminate all LWPs but don't discard them.  If another lwp beat us to
1762 	 * the punch by calling exit(), evaporate now.
1763 	 */
1764 	proc_is_exiting(p);
1765 	if (exitlwps(1) != 0) {
1766 		mutex_enter(&p->p_lock);
1767 		lwp_exit();
1768 	}
1769 
1770 #ifdef C2_AUDIT
1771 	if (audit_active)		/* audit core dump */
1772 		audit_core_start(SIGSEGV);
1773 #endif
1774 	v = core(SIGSEGV, B_FALSE);
1775 #ifdef C2_AUDIT
1776 	if (audit_active)		/* audit core dump */
1777 		audit_core_finish(v ? CLD_KILLED : CLD_DUMPED);
1778 #endif
1779 	exit(v ? CLD_KILLED : CLD_DUMPED, SIGSEGV);
1780 	return (0);
1781 }
1782 
1783 /*
1784  * dump_tss() - Display the TSS structure
1785  */
1786 
1787 #if defined(__amd64)
1788 
1789 static void
1790 dump_tss(void)
1791 {
1792 	const char tss_fmt[] = "tss.%s:\t0x%p\n";  /* Format string */
1793 	struct tss *tss = CPU->cpu_tss;
1794 
1795 	printf(tss_fmt, "tss_rsp0", (void *)tss->tss_rsp0);
1796 	printf(tss_fmt, "tss_rsp1", (void *)tss->tss_rsp1);
1797 	printf(tss_fmt, "tss_rsp2", (void *)tss->tss_rsp2);
1798 
1799 	printf(tss_fmt, "tss_ist1", (void *)tss->tss_ist1);
1800 	printf(tss_fmt, "tss_ist2", (void *)tss->tss_ist2);
1801 	printf(tss_fmt, "tss_ist3", (void *)tss->tss_ist3);
1802 	printf(tss_fmt, "tss_ist4", (void *)tss->tss_ist4);
1803 	printf(tss_fmt, "tss_ist5", (void *)tss->tss_ist5);
1804 	printf(tss_fmt, "tss_ist6", (void *)tss->tss_ist6);
1805 	printf(tss_fmt, "tss_ist7", (void *)tss->tss_ist7);
1806 }
1807 
1808 #elif defined(__i386)
1809 
1810 static void
1811 dump_tss(void)
1812 {
1813 	const char tss_fmt[] = "tss.%s:\t0x%p\n";  /* Format string */
1814 	struct tss *tss = CPU->cpu_tss;
1815 
1816 	printf(tss_fmt, "tss_link", (void *)(uintptr_t)tss->tss_link);
1817 	printf(tss_fmt, "tss_esp0", (void *)(uintptr_t)tss->tss_esp0);
1818 	printf(tss_fmt, "tss_ss0", (void *)(uintptr_t)tss->tss_ss0);
1819 	printf(tss_fmt, "tss_esp1", (void *)(uintptr_t)tss->tss_esp1);
1820 	printf(tss_fmt, "tss_ss1", (void *)(uintptr_t)tss->tss_ss1);
1821 	printf(tss_fmt, "tss_esp2", (void *)(uintptr_t)tss->tss_esp2);
1822 	printf(tss_fmt, "tss_ss2", (void *)(uintptr_t)tss->tss_ss2);
1823 	printf(tss_fmt, "tss_cr3", (void *)(uintptr_t)tss->tss_cr3);
1824 	printf(tss_fmt, "tss_eip", (void *)(uintptr_t)tss->tss_eip);
1825 	printf(tss_fmt, "tss_eflags", (void *)(uintptr_t)tss->tss_eflags);
1826 	printf(tss_fmt, "tss_eax", (void *)(uintptr_t)tss->tss_eax);
1827 	printf(tss_fmt, "tss_ebx", (void *)(uintptr_t)tss->tss_ebx);
1828 	printf(tss_fmt, "tss_ecx", (void *)(uintptr_t)tss->tss_ecx);
1829 	printf(tss_fmt, "tss_edx", (void *)(uintptr_t)tss->tss_edx);
1830 	printf(tss_fmt, "tss_esp", (void *)(uintptr_t)tss->tss_esp);
1831 }
1832 
1833 #endif	/* __amd64 */
1834 
1835 #if defined(TRAPTRACE)
1836 
1837 int ttrace_nrec = 0;		/* number of records to dump out */
1838 int ttrace_dump_nregs = 5;	/* dump out this many records with regs too */
1839 
1840 /*
1841  * Dump out the last ttrace_nrec traptrace records on each CPU
1842  */
1843 static void
1844 dump_ttrace(void)
1845 {
1846 	trap_trace_ctl_t *ttc;
1847 	trap_trace_rec_t *rec;
1848 	uintptr_t current;
1849 	int i, j, k;
1850 	int n = NCPU;
1851 #if defined(__amd64)
1852 	const char banner[] =
1853 		"\ncpu          address    timestamp "
1854 		"type  vc  handler   pc\n";
1855 	const char fmt1[] = "%3d %016lx %12llx ";
1856 #elif defined(__i386)
1857 	const char banner[] =
1858 		"\ncpu address     timestamp type  vc  handler   pc\n";
1859 	const char fmt1[] = "%3d %08lx %12llx ";
1860 #endif
1861 	const char fmt2[] = "%4s %3x ";
1862 	const char fmt3[] = "%8s ";
1863 
1864 	if (ttrace_nrec == 0)
1865 		return;
1866 
1867 	printf(banner);
1868 
1869 	for (i = 0; i < n; i++) {
1870 		ttc = &trap_trace_ctl[i];
1871 		if (ttc->ttc_first == NULL)
1872 			continue;
1873 
1874 		current = ttc->ttc_next - sizeof (trap_trace_rec_t);
1875 		for (j = 0; j < ttrace_nrec; j++) {
1876 			struct sysent	*sys;
1877 			struct autovec	*vec;
1878 			extern struct av_head autovect[];
1879 			int type;
1880 			ulong_t	off;
1881 			char *sym, *stype;
1882 
1883 			if (current < ttc->ttc_first)
1884 				current =
1885 				    ttc->ttc_limit - sizeof (trap_trace_rec_t);
1886 
1887 			if (current == NULL)
1888 				continue;
1889 
1890 			rec = (trap_trace_rec_t *)current;
1891 
1892 			if (rec->ttr_stamp == 0)
1893 				break;
1894 
1895 			printf(fmt1, i, (uintptr_t)rec, rec->ttr_stamp);
1896 
1897 			switch (rec->ttr_marker) {
1898 			case TT_SYSCALL:
1899 			case TT_SYSENTER:
1900 			case TT_SYSC:
1901 			case TT_SYSC64:
1902 #if defined(__amd64)
1903 				sys = &sysent32[rec->ttr_sysnum];
1904 				switch (rec->ttr_marker) {
1905 				case TT_SYSC64:
1906 					sys = &sysent[rec->ttr_sysnum];
1907 					/*FALLTHROUGH*/
1908 #elif defined(__i386)
1909 				sys = &sysent[rec->ttr_sysnum];
1910 				switch (rec->ttr_marker) {
1911 				case TT_SYSC64:
1912 #endif
1913 				case TT_SYSC:
1914 					stype = "sysc";	/* syscall */
1915 					break;
1916 				case TT_SYSCALL:
1917 					stype = "lcal";	/* lcall */
1918 					break;
1919 				case TT_SYSENTER:
1920 					stype = "syse";	/* sysenter */
1921 					break;
1922 				default:
1923 					break;
1924 				}
1925 				printf(fmt2, "sysc", rec->ttr_sysnum);
1926 				if (sys != NULL) {
1927 					sym = kobj_getsymname(
1928 					    (uintptr_t)sys->sy_callc,
1929 					    &off);
1930 					if (sym != NULL)
1931 						printf("%s ", sym);
1932 					else
1933 						printf("%p ", sys->sy_callc);
1934 				} else {
1935 					printf("unknown ");
1936 				}
1937 				break;
1938 
1939 			case TT_INTERRUPT:
1940 				printf(fmt2, "intr", rec->ttr_vector);
1941 				vec = (&autovect[rec->ttr_vector])->avh_link;
1942 				if (vec != NULL) {
1943 					sym = kobj_getsymname(
1944 					    (uintptr_t)vec->av_vector, &off);
1945 					if (sym != NULL)
1946 						printf("%s ", sym);
1947 					else
1948 						printf("%p ", vec->av_vector);
1949 				} else {
1950 					printf("unknown ");
1951 				}
1952 				break;
1953 
1954 			case TT_TRAP:
1955 				type = rec->ttr_regs.r_trapno;
1956 				printf(fmt2, "trap", type);
1957 				printf("#%s ", type < TRAP_TYPES ?
1958 				    trap_type_mnemonic[type] : "trap");
1959 				break;
1960 
1961 			default:
1962 				break;
1963 			}
1964 
1965 			sym = kobj_getsymname(rec->ttr_regs.r_pc, &off);
1966 			if (sym != NULL)
1967 				printf("%s+%lx\n", sym, off);
1968 			else
1969 				printf("%lx\n", rec->ttr_regs.r_pc);
1970 
1971 			if (ttrace_dump_nregs-- > 0) {
1972 				int s;
1973 
1974 				if (rec->ttr_marker == TT_INTERRUPT)
1975 					printf(
1976 					    "\t\tipl %x spl %x pri %x\n",
1977 					    rec->ttr_ipl,
1978 					    rec->ttr_spl,
1979 					    rec->ttr_pri);
1980 
1981 				dumpregs(&rec->ttr_regs);
1982 
1983 				printf("\t%3s: %p\n\n", " ct",
1984 				    (void *)rec->ttr_curthread);
1985 
1986 				/*
1987 				 * print out the pc stack that we recorded
1988 				 * at trap time (if any)
1989 				 */
1990 				for (s = 0; s < rec->ttr_sdepth; s++) {
1991 					uintptr_t fullpc;
1992 
1993 					if (s >= TTR_STACK_DEPTH) {
1994 						printf("ttr_sdepth corrupt\n");
1995 						break;
1996 					}
1997 
1998 					fullpc = (uintptr_t)rec->ttr_stack[s];
1999 
2000 					sym = kobj_getsymname(fullpc, &off);
2001 					if (sym != NULL)
2002 						printf("-> %s+0x%lx()\n",
2003 						    sym, off);
2004 					else
2005 						printf("-> 0x%lx()\n", fullpc);
2006 				}
2007 				printf("\n");
2008 			}
2009 			current -= sizeof (trap_trace_rec_t);
2010 		}
2011 	}
2012 }
2013 
2014 #endif	/* TRAPTRACE */
2015 
2016 void
2017 panic_showtrap(struct trap_info *tip)
2018 {
2019 	showregs(tip->trap_type, tip->trap_regs, tip->trap_addr);
2020 
2021 #if defined(TRAPTRACE)
2022 	dump_ttrace();
2023 #endif	/* TRAPTRACE */
2024 
2025 	if (tip->trap_type == T_DBLFLT)
2026 		dump_tss();
2027 }
2028 
2029 void
2030 panic_savetrap(panic_data_t *pdp, struct trap_info *tip)
2031 {
2032 	panic_saveregs(pdp, tip->trap_regs);
2033 }
2034