xref: /netbsd/sys/arch/amd64/amd64/trap.c (revision 2904b500)
1 /*	$NetBSD: trap.c,v 1.128 2020/09/05 07:26:37 maxv Exp $	*/
2 
3 /*
4  * Copyright (c) 1998, 2000, 2017 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Charles M. Hannum, and by Maxime Villard.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Copyright (c) 1990 The Regents of the University of California.
34  * All rights reserved.
35  *
36  * This code is derived from software contributed to Berkeley by
37  * the University of Utah, and William Jolitz.
38  *
39  * Redistribution and use in source and binary forms, with or without
40  * modification, are permitted provided that the following conditions
41  * are met:
42  * 1. Redistributions of source code must retain the above copyright
43  *    notice, this list of conditions and the following disclaimer.
44  * 2. Redistributions in binary form must reproduce the above copyright
45  *    notice, this list of conditions and the following disclaimer in the
46  *    documentation and/or other materials provided with the distribution.
47  * 3. Neither the name of the University nor the names of its contributors
48  *    may be used to endorse or promote products derived from this software
49  *    without specific prior written permission.
50  *
51  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
52  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
53  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
54  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
55  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
56  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
57  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
58  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
59  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
60  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
61  * SUCH DAMAGE.
62  *
63  *	@(#)trap.c	7.4 (Berkeley) 5/13/91
64  */
65 
66 #include <sys/cdefs.h>
67 __KERNEL_RCSID(0, "$NetBSD: trap.c,v 1.128 2020/09/05 07:26:37 maxv Exp $");
68 
69 #include "opt_ddb.h"
70 #include "opt_kgdb.h"
71 #include "opt_xen.h"
72 #include "opt_dtrace.h"
73 
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/proc.h>
77 #include <sys/acct.h>
78 #include <sys/kauth.h>
79 #include <sys/kernel.h>
80 #include <sys/kmem.h>
81 #include <sys/ras.h>
82 #include <sys/signal.h>
83 #include <sys/syscall.h>
84 #include <sys/cpu.h>
85 #include <sys/ucontext.h>
86 #include <sys/module_hook.h>
87 #include <sys/compat_stub.h>
88 
89 #include <uvm/uvm_extern.h>
90 
91 #include <machine/cpufunc.h>
92 #include <x86/fpu.h>
93 #include <x86/dbregs.h>
94 #include <machine/psl.h>
95 #include <machine/reg.h>
96 #include <machine/trap.h>
97 #include <machine/userret.h>
98 #include <machine/db_machdep.h>
99 
100 #include <x86/nmi.h>
101 
102 #ifndef XENPV
103 #include "isa.h"
104 #endif
105 
106 #include <sys/kgdb.h>
107 
108 #ifdef KDTRACE_HOOKS
109 #include <sys/dtrace_bsd.h>
110 /*
111  * This is a hook which is initialized by the dtrace module to handle traps
112  * which might occur during DTrace probe execution.
113  */
114 dtrace_trap_func_t dtrace_trap_func = NULL;
115 dtrace_doubletrap_func_t dtrace_doubletrap_func = NULL;
116 #endif
117 
118 /*
119  * Module hook for amd64_oosyscall
120  */
121 struct amd64_oosyscall_hook_t amd64_oosyscall_hook;
122 
123 void nmitrap(struct trapframe *);
124 void doubletrap(struct trapframe *);
125 void trap(struct trapframe *);
126 
127 const char * const trap_type[] = {
128 	"privileged instruction fault",		/*  0 T_PRIVINFLT */
129 	"breakpoint trap",			/*  1 T_BPTFLT */
130 	"arithmetic trap",			/*  2 T_ARITHTRAP */
131 	"asynchronous system trap",		/*  3 T_ASTFLT */
132 	"protection fault",			/*  4 T_PROTFLT */
133 	"trace trap",				/*  5 T_TRCTRAP */
134 	"page fault",				/*  6 T_PAGEFLT */
135 	"alignment fault",			/*  7 T_ALIGNFLT */
136 	"integer divide fault",			/*  8 T_DIVIDE */
137 	"non-maskable interrupt",		/*  9 T_NMI */
138 	"overflow trap",			/* 10 T_OFLOW */
139 	"bounds check fault",			/* 11 T_BOUND */
140 	"FPU not available fault",		/* 12 T_DNA */
141 	"double fault",				/* 13 T_DOUBLEFLT */
142 	"FPU operand fetch fault",		/* 14 T_FPOPFLT */
143 	"invalid TSS fault",			/* 15 T_TSSFLT */
144 	"segment not present fault",		/* 16 T_SEGNPFLT */
145 	"stack fault",				/* 17 T_STKFLT */
146 	"machine check fault",			/* 18 T_MCA */
147 	"SSE FP exception",			/* 19 T_XMM */
148 	"reserved trap",			/* 20 T_RESERVED */
149 };
150 int trap_types = __arraycount(trap_type);
151 
152 #ifdef TRAP_SIGDEBUG
153 static void sigdebug(const struct trapframe *, const ksiginfo_t *, int);
154 #define SIGDEBUG(a, b, c) sigdebug(a, b, c)
155 #else
156 #define SIGDEBUG(a, b, c)
157 #endif
158 
159 static void
onfault_restore(struct trapframe * frame,void * onfault,int error)160 onfault_restore(struct trapframe *frame, void *onfault, int error)
161 {
162 	frame->tf_rip = (uintptr_t)onfault;
163 	frame->tf_rax = error;
164 }
165 
166 static void *
onfault_handler(const struct pcb * pcb,const struct trapframe * tf)167 onfault_handler(const struct pcb *pcb, const struct trapframe *tf)
168 {
169 	struct onfault_table {
170 		uintptr_t start;
171 		uintptr_t end;
172 		void *handler;
173 	};
174 	extern const struct onfault_table onfault_table[];
175 	const struct onfault_table *p;
176 	uintptr_t pc;
177 
178 	if (pcb->pcb_onfault != NULL) {
179 		return pcb->pcb_onfault;
180 	}
181 
182 	pc = tf->tf_rip;
183 	for (p = onfault_table; p->start; p++) {
184 		if (p->start <= pc && pc < p->end) {
185 			return p->handler;
186 		}
187 	}
188 	return NULL;
189 }
190 
191 static void
trap_print(const struct trapframe * frame,const lwp_t * l)192 trap_print(const struct trapframe *frame, const lwp_t *l)
193 {
194 	const int type = frame->tf_trapno;
195 
196 	if (frame->tf_trapno < trap_types) {
197 		printf("fatal %s", trap_type[type]);
198 	} else {
199 		printf("unknown trap %d", type);
200 	}
201 	printf(" in %s mode\n", (type & T_USER) ? "user" : "supervisor");
202 
203 	printf("trap type %d code %#lx rip %#lx cs %#lx rflags %#lx cr2 %#lx "
204 	    "ilevel %#x rsp %#lx\n",
205 	    type, frame->tf_err, (u_long)frame->tf_rip, frame->tf_cs,
206 	    frame->tf_rflags, rcr2(), curcpu()->ci_ilevel, frame->tf_rsp);
207 
208 	printf("curlwp %p pid %d.%d lowest kstack %p\n",
209 	    l, l->l_proc->p_pid, l->l_lid, KSTACK_LOWEST_ADDR(l));
210 }
211 
212 void
nmitrap(struct trapframe * frame)213 nmitrap(struct trapframe *frame)
214 {
215 	const int type = T_NMI;
216 
217 	if (nmi_dispatch(frame))
218 		return;
219 	/* NMI can be hooked up to a pushbutton for debugging */
220 	if (kgdb_trap(type, frame))
221 		return;
222 	if (kdb_trap(type, 0, frame))
223 		return;
224 	/* machine/parity/power fail/"kitchen sink" faults */
225 
226 	x86_nmi();
227 }
228 
229 void
doubletrap(struct trapframe * frame)230 doubletrap(struct trapframe *frame)
231 {
232 	const int type = T_DOUBLEFLT;
233 	struct lwp *l = curlwp;
234 
235 	trap_print(frame, l);
236 
237 	if (kdb_trap(type, 0, frame))
238 		return;
239 	if (kgdb_trap(type, frame))
240 		return;
241 
242 	panic("double fault");
243 }
244 
245 /*
246  * trap(frame): exception, fault, and trap interface to BSD kernel.
247  *
248  * This common code is called from assembly language IDT gate entry routines
249  * that prepare a suitable stack frame, and restore this frame after the
250  * exception has been processed. Note that the effect is as if the arguments
251  * were passed call by reference.
252  *
253  * Note that the fpu traps (07 T_DNA, 10 T_ARITHTRAP and 13 T_XMM)
254  * jump directly into the code in x86/fpu.c so they get processed
255  * without interrupts being enabled.
256  */
257 void
trap(struct trapframe * frame)258 trap(struct trapframe *frame)
259 {
260 	struct lwp *l = curlwp;
261 	struct proc *p;
262 	struct pcb *pcb;
263 	extern char kcopy_fault[];
264 	ksiginfo_t ksi;
265 	void *onfault;
266 	int type, error;
267 	uint64_t cr2;
268 	bool pfail;
269 
270 	if (__predict_true(l != NULL)) {
271 		pcb = lwp_getpcb(l);
272 		p = l->l_proc;
273 	} else {
274 		/*
275 		 * This can happen eg on break points in early on boot.
276 		 */
277 		pcb = NULL;
278 		p = NULL;
279 	}
280 	type = frame->tf_trapno;
281 
282 	if (!KERNELMODE(frame->tf_cs)) {
283 		type |= T_USER;
284 		l->l_md.md_regs = frame;
285 		LWP_CACHE_CREDS(l, p);
286 	}
287 
288 #ifdef KDTRACE_HOOKS
289 	/*
290 	 * A trap can occur while DTrace executes a probe. Before
291 	 * executing the probe, DTrace blocks re-scheduling and sets
292 	 * a flag in its per-cpu flags to indicate that it doesn't
293 	 * want to fault. On returning from the probe, the no-fault
294 	 * flag is cleared and finally re-scheduling is enabled.
295 	 *
296 	 * If the DTrace kernel module has registered a trap handler,
297 	 * call it and if it returns non-zero, assume that it has
298 	 * handled the trap and modified the trap frame so that this
299 	 * function can return normally.
300 	 */
301 	if ((type == T_PROTFLT || type == T_PAGEFLT) &&
302 	    dtrace_trap_func != NULL) {
303 		if ((*dtrace_trap_func)(frame, type)) {
304 			return;
305 		}
306 	}
307 #endif
308 
309 	switch (type) {
310 
311 	default:
312 	we_re_toast:
313 		trap_print(frame, l);
314 
315 		if (kdb_trap(type, 0, frame))
316 			return;
317 		if (kgdb_trap(type, frame))
318 			return;
319 		/*
320 		 * If this is a breakpoint, don't panic if we're not connected.
321 		 */
322 		if (type == T_BPTFLT && kgdb_disconnected()) {
323 			printf("kgdb: ignored %s\n", trap_type[type]);
324 			return;
325 		}
326 		panic("trap");
327 		/*NOTREACHED*/
328 
329 	case T_PROTFLT:
330 	case T_SEGNPFLT:
331 	case T_ALIGNFLT:
332 	case T_STKFLT:
333 	case T_TSSFLT:
334 		if (p == NULL)
335 			goto we_re_toast;
336 
337 		/* Check for copyin/copyout fault. */
338 		onfault = onfault_handler(pcb, frame);
339 		if (onfault != NULL) {
340 			onfault_restore(frame, onfault, EFAULT);
341 			return;
342 		}
343 
344 		goto we_re_toast;
345 
346 	case T_PROTFLT|T_USER:		/* protection fault */
347 	{	int hook_ret;
348 
349 		MODULE_HOOK_CALL(amd64_oosyscall_hook, (p, frame),
350 			ENOSYS, hook_ret);
351 		if (hook_ret == 0) {
352 			/* Do the syscall */
353 			p->p_md.md_syscall(frame);
354 			goto out;
355 		}
356 	}
357 		/* FALLTHROUGH */
358 	case T_TSSFLT|T_USER:
359 	case T_SEGNPFLT|T_USER:
360 	case T_STKFLT|T_USER:
361 	case T_ALIGNFLT|T_USER:
362 		KSI_INIT_TRAP(&ksi);
363 		ksi.ksi_trap = type & ~T_USER;
364 		ksi.ksi_addr = (void *)frame->tf_rip;
365 		switch (type) {
366 		case T_SEGNPFLT|T_USER:
367 		case T_STKFLT|T_USER:
368 			ksi.ksi_signo = SIGBUS;
369 			ksi.ksi_code = BUS_ADRERR;
370 			break;
371 		case T_TSSFLT|T_USER:
372 			ksi.ksi_signo = SIGBUS;
373 			ksi.ksi_code = BUS_OBJERR;
374 			break;
375 		case T_ALIGNFLT|T_USER:
376 			ksi.ksi_signo = SIGBUS;
377 			ksi.ksi_code = BUS_ADRALN;
378 			break;
379 		case T_PROTFLT|T_USER:
380 			ksi.ksi_signo = SIGSEGV;
381 			ksi.ksi_code = SEGV_ACCERR;
382 			break;
383 		default:
384 			KASSERT(0);
385 			break;
386 		}
387 		goto trapsignal;
388 
389 	case T_PRIVINFLT|T_USER:	/* privileged instruction fault */
390 	case T_FPOPFLT|T_USER:		/* coprocessor operand fault */
391 		KSI_INIT_TRAP(&ksi);
392 		ksi.ksi_signo = SIGILL;
393 		ksi.ksi_trap = type & ~T_USER;
394 		ksi.ksi_addr = (void *) frame->tf_rip;
395 		switch (type) {
396 		case T_PRIVINFLT|T_USER:
397 			ksi.ksi_code = ILL_PRVOPC;
398 			break;
399 		case T_FPOPFLT|T_USER:
400 			ksi.ksi_code = ILL_COPROC;
401 			break;
402 		default:
403 			KASSERT(0);
404 			break;
405 		}
406 		goto trapsignal;
407 
408 	case T_ASTFLT|T_USER:
409 		/* Allow process switch. */
410 		//curcpu()->ci_data.cpu_nast++;
411 		if (l->l_pflag & LP_OWEUPC) {
412 			l->l_pflag &= ~LP_OWEUPC;
413 			ADDUPROF(l);
414 		}
415 		goto out;
416 
417 	case T_BOUND|T_USER:
418 	case T_OFLOW|T_USER:
419 	case T_DIVIDE|T_USER:
420 		KSI_INIT_TRAP(&ksi);
421 		ksi.ksi_signo = SIGFPE;
422 		ksi.ksi_trap = type & ~T_USER;
423 		ksi.ksi_addr = (void *)frame->tf_rip;
424 		switch (type) {
425 		case T_BOUND|T_USER:
426 			ksi.ksi_code = FPE_FLTSUB;
427 			break;
428 		case T_OFLOW|T_USER:
429 			ksi.ksi_code = FPE_INTOVF;
430 			break;
431 		case T_DIVIDE|T_USER:
432 			ksi.ksi_code = FPE_INTDIV;
433 			break;
434 		default:
435 			KASSERT(0);
436 			break;
437 		}
438 		goto trapsignal;
439 
440 	case T_PAGEFLT:
441 		/* Allow page faults in kernel mode. */
442 		if (__predict_false(l == NULL))
443 			goto we_re_toast;
444 
445 		onfault = pcb->pcb_onfault;
446 
447 		if (cpu_intr_p() || (l->l_pflag & LP_INTR) != 0) {
448 			goto we_re_toast;
449 		}
450 
451 		cr2 = rcr2();
452 
453 		if (frame->tf_err & PGEX_I) {
454 			/* SMEP might have brought us here */
455 			if (cr2 < VM_MAXUSER_ADDRESS) {
456 				printf("prevented execution of %p (SMEP)\n",
457 				    (void *)cr2);
458 				goto we_re_toast;
459 			}
460 		}
461 
462 		if ((frame->tf_err & PGEX_P) &&
463 		    cr2 < VM_MAXUSER_ADDRESS) {
464 			/* SMAP might have brought us here */
465 			if (onfault_handler(pcb, frame) == NULL) {
466 				printf("prevented access to %p (SMAP)\n",
467 				    (void *)cr2);
468 				goto we_re_toast;
469 			}
470 		}
471 
472 		goto pagefltcommon;
473 
474 	case T_PAGEFLT|T_USER: {
475 		register vaddr_t va;
476 		register struct vmspace *vm;
477 		register struct vm_map *map;
478 		vm_prot_t ftype;
479 		extern struct vm_map *kernel_map;
480 
481 		cr2 = rcr2();
482 		if (p->p_emul->e_usertrap != NULL &&
483 		    (*p->p_emul->e_usertrap)(l, cr2, frame) != 0)
484 			return;
485 pagefltcommon:
486 		vm = p->p_vmspace;
487 		if (__predict_false(vm == NULL)) {
488 			goto we_re_toast;
489 		}
490 		pcb->pcb_cr2 = cr2;
491 		va = trunc_page((vaddr_t)cr2);
492 		/*
493 		 * It is only a kernel address space fault iff:
494 		 *	1. (type & T_USER) == 0  and
495 		 *	2. pcb_onfault not set or
496 		 *	3. pcb_onfault set but supervisor space fault
497 		 * The last can occur during an exec() copyin where the
498 		 * argument space is lazy-allocated.
499 		 */
500 		if (type == T_PAGEFLT && va >= VM_MIN_KERNEL_ADDRESS)
501 			map = kernel_map;
502 		else
503 			map = &vm->vm_map;
504 		if (frame->tf_err & PGEX_W)
505 			ftype = VM_PROT_WRITE;
506 		else if (frame->tf_err & PGEX_I)
507 			ftype = VM_PROT_EXECUTE;
508 		else
509 			ftype = VM_PROT_READ;
510 
511 #ifdef DIAGNOSTIC
512 		if (map == kernel_map && va == 0) {
513 			printf("trap: bad kernel access at %lx\n", va);
514 			goto we_re_toast;
515 		}
516 #endif
517 		/* Fault the original page in. */
518 		onfault = pcb->pcb_onfault;
519 		pcb->pcb_onfault = NULL;
520 		error = uvm_fault(map, va, ftype);
521 		pcb->pcb_onfault = onfault;
522 		if (error == 0) {
523 			if (map != kernel_map && (void *)va >= vm->vm_maxsaddr)
524 				uvm_grow(p, va);
525 
526 			pfail = false;
527 			while (type == T_PAGEFLT) {
528 				/*
529 				 * we need to switch pmap now if we're in
530 				 * the middle of copyin/out.
531 				 *
532 				 * but we don't need to do so for kcopy as
533 				 * it never touch userspace.
534  				 */
535 				kpreempt_disable();
536 				if (curcpu()->ci_want_pmapload) {
537 					onfault = onfault_handler(pcb, frame);
538 					if (onfault != kcopy_fault) {
539 						pmap_load();
540 					}
541 				}
542 				/*
543 				 * We need to keep the pmap loaded and
544 				 * so avoid being preempted until back
545 				 * into the copy functions.  Disable
546 				 * interrupts at the hardware level before
547 				 * re-enabling preemption.  Interrupts
548 				 * will be re-enabled by 'iret' when
549 				 * returning back out of the trap stub.
550 				 * They'll only be re-enabled when the
551 				 * program counter is once again in
552 				 * the copy functions, and so visible
553 				 * to cpu_kpreempt_exit().
554 				 */
555 #ifndef XENPV
556 				x86_disable_intr();
557 #endif
558 				l->l_nopreempt--;
559 				if (l->l_nopreempt > 0 || !l->l_dopreempt ||
560 				    pfail) {
561 					return;
562 				}
563 #ifndef XENPV
564 				x86_enable_intr();
565 #endif
566 				/*
567 				 * If preemption fails for some reason,
568 				 * don't retry it.  The conditions won't
569 				 * change under our nose.
570 				 */
571 				pfail = kpreempt(0);
572 			}
573 			goto out;
574 		}
575 
576 		if (type == T_PAGEFLT) {
577 			onfault = onfault_handler(pcb, frame);
578 			if (onfault != NULL) {
579 				onfault_restore(frame, onfault, error);
580 				return;
581 			}
582 
583 			printf("uvm_fault(%p, 0x%lx, %d) -> %x\n",
584 			    map, va, ftype, error);
585 			goto we_re_toast;
586 		}
587 
588 		KSI_INIT_TRAP(&ksi);
589 		ksi.ksi_trap = type & ~T_USER;
590 		ksi.ksi_addr = (void *)cr2;
591 		switch (error) {
592 		case EINVAL:
593 			ksi.ksi_signo = SIGBUS;
594 			ksi.ksi_code = BUS_ADRERR;
595 			break;
596 		case EACCES:
597 			ksi.ksi_signo = SIGSEGV;
598 			ksi.ksi_code = SEGV_ACCERR;
599 			error = EFAULT;
600 			break;
601 		case ENOMEM:
602 			ksi.ksi_signo = SIGKILL;
603 			printf("UVM: pid %d.%d (%s), uid %d killed: "
604 			    "out of swap\n", p->p_pid, l->l_lid, p->p_comm,
605 			    l->l_cred ?  kauth_cred_geteuid(l->l_cred) : -1);
606 			break;
607 		default:
608 			ksi.ksi_signo = SIGSEGV;
609 			ksi.ksi_code = SEGV_MAPERR;
610 			break;
611 		}
612 
613 		SIGDEBUG(frame, &ksi, error);
614  		(*p->p_emul->e_trapsignal)(l, &ksi);
615 		break;
616 	}
617 
618 	case T_TRCTRAP:
619 		/*
620 		 * Ignore debug register trace traps due to
621 		 * accesses in the user's address space, which
622 		 * can happen under several conditions such as
623 		 * if a user sets a watchpoint on a buffer and
624 		 * then passes that buffer to a system call.
625 		 * We still want to get TRCTRAPS for addresses
626 		 * in kernel space because that is useful when
627 		 * debugging the kernel.
628 		 */
629 		if (x86_dbregs_user_trap())
630 			break;
631 
632 		goto we_re_toast;
633 
634 	case T_BPTFLT|T_USER:		/* bpt instruction fault */
635 	case T_TRCTRAP|T_USER:		/* trace trap */
636 		/*
637 		 * Don't go single-stepping into a RAS.
638 		 */
639 		if (p->p_raslist == NULL ||
640 		    (ras_lookup(p, (void *)frame->tf_rip) == (void *)-1)) {
641 			KSI_INIT_TRAP(&ksi);
642 			ksi.ksi_signo = SIGTRAP;
643 			ksi.ksi_trap = type & ~T_USER;
644 			if (x86_dbregs_user_trap()) {
645 				x86_dbregs_store_dr6(l);
646 				ksi.ksi_code = TRAP_DBREG;
647 			} else if (type == (T_BPTFLT|T_USER))
648 				ksi.ksi_code = TRAP_BRKPT;
649 			else
650 				ksi.ksi_code = TRAP_TRACE;
651 			(*p->p_emul->e_trapsignal)(l, &ksi);
652 		}
653 		break;
654 	}
655 
656 	if ((type & T_USER) == 0)
657 		return;
658 out:
659 	userret(l);
660 	return;
661 trapsignal:
662 	SIGDEBUG(frame, &ksi, 0);
663 	(*p->p_emul->e_trapsignal)(l, &ksi);
664 	userret(l);
665 }
666 
667 /*
668  * startlwp: start of a new LWP.
669  */
670 void
startlwp(void * arg)671 startlwp(void *arg)
672 {
673 	ucontext_t *uc = arg;
674 	lwp_t *l = curlwp;
675 	int error __diagused;
676 
677 	error = cpu_setmcontext(l, &uc->uc_mcontext, uc->uc_flags);
678 	KASSERT(error == 0);
679 
680 	kmem_free(uc, sizeof(ucontext_t));
681 	userret(l);
682 }
683 
684 #ifdef TRAP_SIGDEBUG
685 static void
frame_dump(const struct trapframe * tf,struct pcb * pcb)686 frame_dump(const struct trapframe *tf, struct pcb *pcb)
687 {
688 
689 	printf("trapframe %p\n", tf);
690 	printf("rip %#018lx  rsp %#018lx  rfl %#018lx\n",
691 	    tf->tf_rip, tf->tf_rsp, tf->tf_rflags);
692 	printf("rdi %#018lx  rsi %#018lx  rdx %#018lx\n",
693 	    tf->tf_rdi, tf->tf_rsi, tf->tf_rdx);
694 	printf("rcx %#018lx  r8  %#018lx  r9  %#018lx\n",
695 	    tf->tf_rcx, tf->tf_r8, tf->tf_r9);
696 	printf("r10 %#018lx  r11 %#018lx  r12 %#018lx\n",
697 	    tf->tf_r10, tf->tf_r11, tf->tf_r12);
698 	printf("r13 %#018lx  r14 %#018lx  r15 %#018lx\n",
699 	    tf->tf_r13, tf->tf_r14, tf->tf_r15);
700 	printf("rbp %#018lx  rbx %#018lx  rax %#018lx\n",
701 	    tf->tf_rbp, tf->tf_rbx, tf->tf_rax);
702 	printf("cs %#04lx  ds %#04lx  es %#04lx  "
703 	    "fs %#04lx  gs %#04lx  ss %#04lx\n",
704 	    tf->tf_cs & 0xffff, tf->tf_ds & 0xffff, tf->tf_es & 0xffff,
705 	    tf->tf_fs & 0xffff, tf->tf_gs & 0xffff, tf->tf_ss & 0xffff);
706 	printf("fsbase %#018lx gsbase %#018lx\n", pcb->pcb_fs, pcb->pcb_gs);
707 	printf("\n");
708 	hexdump(printf, "Stack dump", tf, 256);
709 }
710 
711 static void
sigdebug(const struct trapframe * tf,const ksiginfo_t * ksi,int e)712 sigdebug(const struct trapframe *tf, const ksiginfo_t *ksi, int e)
713 {
714 	struct lwp *l = curlwp;
715 	struct proc *p = l->l_proc;
716 
717 	printf("pid %d.%d (%s): signal %d code=%d (trap %#lx) "
718 	    "@rip %#lx addr %#lx error=%d\n",
719 	    p->p_pid, l->l_lid, p->p_comm, ksi->ksi_signo, ksi->ksi_code,
720 	    tf->tf_trapno, tf->tf_rip, rcr2(), e);
721 	frame_dump(tf, lwp_getpcb(l));
722 }
723 #endif
724