xref: /freebsd/sys/i386/i386/trap.c (revision 3342e5967dc7193d97f99a92b81824db81efe2f1)
1 /*-
2  * SPDX-License-Identifier: BSD-4-Clause
3  *
4  * Copyright (C) 1994, David Greenman
5  * Copyright (c) 1990, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * the University of Utah, and William Jolitz.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the University of
22  *	California, Berkeley and its contributors.
23  * 4. Neither the name of the University nor the names of its contributors
24  *    may be used to endorse or promote products derived from this software
25  *    without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37  * SUCH DAMAGE.
38  */
39 
40 #include <sys/cdefs.h>
41 /*
42  * 386 Trap and System call handling
43  */
44 
45 #include "opt_clock.h"
46 #include "opt_cpu.h"
47 #include "opt_hwpmc_hooks.h"
48 #include "opt_isa.h"
49 #include "opt_kdb.h"
50 #include "opt_trap.h"
51 
52 #include <sys/param.h>
53 #include <sys/bus.h>
54 #include <sys/systm.h>
55 #include <sys/proc.h>
56 #include <sys/ptrace.h>
57 #include <sys/kdb.h>
58 #include <sys/kernel.h>
59 #include <sys/ktr.h>
60 #include <sys/lock.h>
61 #include <sys/mutex.h>
62 #include <sys/resourcevar.h>
63 #include <sys/signalvar.h>
64 #include <sys/syscall.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysent.h>
67 #include <sys/uio.h>
68 #include <sys/vmmeter.h>
69 #ifdef HWPMC_HOOKS
70 #include <sys/pmckern.h>
71 PMC_SOFT_DEFINE( , , page_fault, all);
72 PMC_SOFT_DEFINE( , , page_fault, read);
73 PMC_SOFT_DEFINE( , , page_fault, write);
74 #endif
75 #include <security/audit/audit.h>
76 
77 #include <vm/vm.h>
78 #include <vm/vm_param.h>
79 #include <vm/pmap.h>
80 #include <vm/vm_kern.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_page.h>
83 #include <vm/vm_extern.h>
84 
85 #include <machine/cpu.h>
86 #include <machine/intr_machdep.h>
87 #include <x86/mca.h>
88 #include <machine/md_var.h>
89 #include <machine/pcb.h>
90 #ifdef SMP
91 #include <machine/smp.h>
92 #endif
93 #include <machine/stack.h>
94 #include <machine/trap.h>
95 #include <machine/tss.h>
96 #include <machine/vm86.h>
97 
98 #ifdef POWERFAIL_NMI
99 #include <sys/syslog.h>
100 #include <machine/clock.h>
101 #endif
102 
103 #ifdef KDTRACE_HOOKS
104 #include <sys/dtrace_bsd.h>
105 #endif
106 
107 void trap(struct trapframe *frame);
108 void syscall(struct trapframe *frame);
109 
110 static int trap_pfault(struct trapframe *, bool, vm_offset_t, int *, int *);
111 static void trap_fatal(struct trapframe *, vm_offset_t);
112 #ifdef KDTRACE_HOOKS
113 static bool trap_user_dtrace(struct trapframe *,
114     int (**hook)(struct trapframe *));
115 #endif
116 void dblfault_handler(void);
117 
118 extern inthand_t IDTVEC(bpt), IDTVEC(dbg), IDTVEC(int0x80_syscall);
119 extern uint64_t pg_nx;
120 
121 struct trap_data {
122 	bool		ei;
123 	const char	*msg;
124 };
125 
126 static const struct trap_data trap_data[] = {
127 	[T_PRIVINFLT] =	{ .ei = true,	.msg = "privileged instruction fault" },
128 	[T_BPTFLT] =	{ .ei = false,	.msg = "breakpoint instruction fault" },
129 	[T_ARITHTRAP] =	{ .ei = true,	.msg = "arithmetic trap" },
130 	[T_PROTFLT] =	{ .ei = true,	.msg = "general protection fault" },
131 	[T_TRCTRAP] =	{ .ei = false,	.msg = "debug exception" },
132 	[T_PAGEFLT] =	{ .ei = true,	.msg = "page fault" },
133 	[T_ALIGNFLT] = 	{ .ei = true,	.msg = "alignment fault" },
134 	[T_DIVIDE] =	{ .ei = true,	.msg = "integer divide fault" },
135 	[T_NMI] =	{ .ei = false,	.msg = "non-maskable interrupt trap" },
136 	[T_OFLOW] =	{ .ei = true,	.msg = "overflow trap" },
137 	[T_BOUND] =	{ .ei = true,	.msg = "FPU bounds check fault" },
138 	[T_DNA] =	{ .ei = true,	.msg = "FPU device not available" },
139 	[T_DOUBLEFLT] =	{ .ei = false,	.msg = "double fault" },
140 	[T_FPOPFLT] =	{ .ei = true,	.msg = "FPU operand fetch fault" },
141 	[T_TSSFLT] =	{ .ei = true,	.msg = "invalid TSS fault" },
142 	[T_SEGNPFLT] =	{ .ei = true,	.msg = "segment not present fault" },
143 	[T_STKFLT] =	{ .ei = true,	.msg = "stack fault" },
144 	[T_MCHK] =	{ .ei = true,	.msg = "machine check trap" },
145 	[T_XMMFLT] =	{ .ei = true,	.msg = "SIMD floating-point exception" },
146 	[T_DTRACE_RET] ={ .ei = true,	.msg = "DTrace pid return trap" },
147 };
148 
149 static bool
trap_enable_intr(int trapno)150 trap_enable_intr(int trapno)
151 {
152 
153 	MPASS(trapno > 0);
154 	if (trapno < nitems(trap_data) && trap_data[trapno].msg != NULL)
155 		return (trap_data[trapno].ei);
156 	return (false);
157 }
158 
159 static const char *
trap_msg(int trapno)160 trap_msg(int trapno)
161 {
162 	const char *res;
163 	static const char unkn[] = "UNKNOWN";
164 
165 	res = NULL;
166 	if (trapno < nitems(trap_data))
167 		res = trap_data[trapno].msg;
168 	if (res == NULL)
169 		res = unkn;
170 	return (res);
171 }
172 
173 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
174 int has_f00f_bug = 0;		/* Initialized so that it can be patched. */
175 #endif
176 
177 static int uprintf_signal;
178 SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RW,
179     &uprintf_signal, 0,
180     "Print debugging information on trap signal to ctty");
181 
182 
183 #ifdef INVARIANTS
184 static __inline register_t
read_esp(void)185 read_esp(void)
186 {
187 	register_t res;
188 
189 	__asm __volatile("movl\t%%esp,%0" : "=r" (res));
190 	return (res);
191 }
192 
193 void
trap_check_kstack(void)194 trap_check_kstack(void)
195 {
196 	struct thread *td;
197 	vm_offset_t stk;
198 
199 	td = curthread;
200 	stk = read_esp();
201 	if (stk >= PMAP_TRM_MIN_ADDRESS)
202 		panic("td %p stack %#x in trampoline", td, stk);
203 	if (!kstack_contains(td, stk, 0))
204 		panic("td %p stack %#x not in kstack VA %#x %d",
205 		    td, stk, td->td_kstack, td->td_kstack_pages);
206 }
207 #endif
208 
209 /*
210  * Exception, fault, and trap interface to the FreeBSD kernel.
211  * This common code is called from assembly language IDT gate entry
212  * routines that prepare a suitable stack frame, and restore this
213  * frame after the exception has been processed.
214  */
215 
216 void
trap(struct trapframe * frame)217 trap(struct trapframe *frame)
218 {
219 	ksiginfo_t ksi;
220 	struct thread *td;
221 	struct proc *p;
222 	int pf, signo, ucode;
223 	u_int type;
224 	register_t addr, dr6;
225 	vm_offset_t eva;
226 #ifdef POWERFAIL_NMI
227 	static int lastalert = 0;
228 #endif
229 
230 	td = curthread;
231 	p = td->td_proc;
232 	dr6 = 0;
233 
234 	VM_CNT_INC(v_trap);
235 	type = frame->tf_trapno;
236 
237 	KASSERT((read_eflags() & PSL_I) == 0,
238 	    ("trap: interrupts enabled, type %d frame %p", type, frame));
239 
240 #ifdef KDB
241 	if (kdb_active) {
242 		kdb_reenter();
243 		return;
244 	}
245 #endif
246 	trap_check_kstack();
247 
248 	if (type == T_NMI) {
249 		nmi_handle_intr(frame);
250 		return;
251 	}
252 
253 	if (type == T_RESERVED) {
254 		trap_fatal(frame, 0);
255 		return;
256 	}
257 
258 	if (type == T_MCHK) {
259 		mca_intr();
260 		return;
261 	}
262 
263 #ifdef KDTRACE_HOOKS
264 	/*
265 	 * A trap can occur while DTrace executes a probe. Before
266 	 * executing the probe, DTrace blocks re-scheduling and sets
267 	 * a flag in its per-cpu flags to indicate that it doesn't
268 	 * want to fault. On returning from the probe, the no-fault
269 	 * flag is cleared and finally re-scheduling is enabled.
270 	 */
271 	if ((type == T_PROTFLT || type == T_PAGEFLT) &&
272 	    dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type))
273 		return;
274 #endif
275 
276 	/*
277 	 * We must not allow context switches until %cr2 is read.
278 	 * Also, for some Cyrix CPUs, %cr2 is clobbered by interrupts.
279 	 * All faults use interrupt gates, so %cr2 can be safely read
280 	 * now, before optional enable of the interrupts below.
281 	 */
282 	if (type == T_PAGEFLT)
283 		eva = rcr2();
284 
285 	/*
286 	 * Buggy application or kernel code has disabled interrupts
287 	 * and then trapped.  Enabling interrupts now is wrong, but it
288 	 * is better than running with interrupts disabled until they
289 	 * are accidentally enabled later.
290 	 */
291 	if ((frame->tf_eflags & PSL_I) == 0 && TRAPF_USERMODE(frame) &&
292 	    (curpcb->pcb_flags & PCB_VM86CALL) == 0)
293 		uprintf("pid %ld (%s): usermode trap %d (%s) with "
294 		    "interrupts disabled\n",
295 		    (long)curproc->p_pid, curthread->td_name, type,
296 		    trap_data[type].msg);
297 
298 	/*
299 	 * Conditionally reenable interrupts.  If we hold a spin lock,
300 	 * then we must not reenable interrupts.  This might be a
301 	 * spurious page fault.
302 	 */
303 	if (trap_enable_intr(type) && td->td_md.md_spinlock_count == 0 &&
304 	    frame->tf_eip != (int)cpu_switch_load_gs)
305 		enable_intr();
306 
307         if (TRAPF_USERMODE(frame) && (curpcb->pcb_flags & PCB_VM86CALL) == 0) {
308 		/* user trap */
309 
310 		td->td_pticks = 0;
311 		td->td_frame = frame;
312 		addr = frame->tf_eip;
313 		if (td->td_cowgen != atomic_load_int(&p->p_cowgen))
314 			thread_cow_update(td);
315 
316 		switch (type) {
317 		case T_PRIVINFLT:	/* privileged instruction fault */
318 			signo = SIGILL;
319 			ucode = ILL_PRVOPC;
320 			break;
321 
322 		case T_BPTFLT:		/* bpt instruction fault */
323 #ifdef KDTRACE_HOOKS
324 			if (trap_user_dtrace(frame, &dtrace_pid_probe_ptr))
325 				return;
326 #else
327 			enable_intr();
328 #endif
329 			signo = SIGTRAP;
330 			ucode = TRAP_BRKPT;
331 			break;
332 
333 		case T_TRCTRAP:		/* debug exception */
334 			enable_intr();
335 user_trctrap_out:
336 			signo = SIGTRAP;
337 			ucode = TRAP_TRACE;
338 			dr6 = rdr6();
339 			if ((dr6 & DBREG_DR6_BS) != 0) {
340 				PROC_LOCK(td->td_proc);
341 				if ((td->td_dbgflags & TDB_STEP) != 0) {
342 					td->td_frame->tf_eflags &= ~PSL_T;
343 					td->td_dbgflags &= ~TDB_STEP;
344 				}
345 				PROC_UNLOCK(td->td_proc);
346 			}
347 			break;
348 
349 		case T_ARITHTRAP:	/* arithmetic trap */
350 			ucode = npxtrap_x87();
351 			if (ucode == -1)
352 				return;
353 			signo = SIGFPE;
354 			break;
355 
356 		/*
357 		 * The following two traps can happen in vm86 mode,
358 		 * and, if so, we want to handle them specially.
359 		 */
360 		case T_PROTFLT:		/* general protection fault */
361 		case T_STKFLT:		/* stack fault */
362 			if (frame->tf_eflags & PSL_VM) {
363 				signo = vm86_emulate((struct vm86frame *)frame);
364 				ucode = 0;	/* XXXKIB: better code ? */
365 				if (signo == SIGTRAP) {
366 					load_dr6(rdr6() | 0x4000);
367 					goto user_trctrap_out;
368 				}
369 				if (signo == 0)
370 					goto user;
371 				break;
372 			}
373 			signo = SIGBUS;
374 			ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
375 			break;
376 		case T_SEGNPFLT:	/* segment not present fault */
377 			signo = SIGBUS;
378 			ucode = BUS_ADRERR;
379 			break;
380 		case T_TSSFLT:		/* invalid TSS fault */
381 			signo = SIGBUS;
382 			ucode = BUS_OBJERR;
383 			break;
384 		case T_ALIGNFLT:
385 			signo = SIGBUS;
386 			ucode = BUS_ADRALN;
387 			break;
388 		case T_DOUBLEFLT:	/* double fault */
389 		default:
390 			signo = SIGBUS;
391 			ucode = BUS_OBJERR;
392 			break;
393 
394 		case T_PAGEFLT:		/* page fault */
395 			addr = eva;
396 			pf = trap_pfault(frame, true, eva, &signo, &ucode);
397 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
398 			if (pf == -2) {
399 				/*
400 				 * The f00f hack workaround has triggered, so
401 				 * treat the fault as an illegal instruction
402 				 * (T_PRIVINFLT) instead of a page fault.
403 				 */
404 				type = frame->tf_trapno = T_PRIVINFLT;
405 				break;
406 			}
407 #endif
408 			if (pf == -1)
409 				return;
410 			if (pf == 0)
411 				goto user;
412 			break;
413 
414 		case T_DIVIDE:		/* integer divide fault */
415 			ucode = FPE_INTDIV;
416 			signo = SIGFPE;
417 			break;
418 
419 		case T_NMI:
420 #ifdef POWERFAIL_NMI
421 #ifndef TIMER_FREQ
422 #  define TIMER_FREQ 1193182
423 #endif
424 			if (time_second - lastalert > 10) {
425 				log(LOG_WARNING, "NMI: power fail\n");
426 				sysbeep(880, SBT_1S);
427 				lastalert = time_second;
428 			}
429 			return;
430 #else /* !POWERFAIL_NMI */
431 			nmi_handle_intr(frame);
432 			return;
433 #endif /* POWERFAIL_NMI */
434 
435 		case T_OFLOW:		/* integer overflow fault */
436 			ucode = FPE_INTOVF;
437 			signo = SIGFPE;
438 			break;
439 
440 		case T_BOUND:		/* bounds check fault */
441 			ucode = FPE_FLTSUB;
442 			signo = SIGFPE;
443 			break;
444 
445 		case T_DNA:
446 			KASSERT(PCB_USER_FPU(td->td_pcb),
447 			    ("kernel FPU ctx has leaked"));
448 			/* transparent fault (due to context switch "late") */
449 			if (npxdna())
450 				return;
451 			uprintf("pid %d killed due to lack of floating point\n",
452 				p->p_pid);
453 			signo = SIGKILL;
454 			ucode = 0;
455 			break;
456 
457 		case T_FPOPFLT:		/* FPU operand fetch fault */
458 			ucode = ILL_COPROC;
459 			signo = SIGILL;
460 			break;
461 
462 		case T_XMMFLT:		/* SIMD floating-point exception */
463 			ucode = npxtrap_sse();
464 			if (ucode == -1)
465 				return;
466 			signo = SIGFPE;
467 			break;
468 #ifdef KDTRACE_HOOKS
469 		case T_DTRACE_RET:
470 			(void)trap_user_dtrace(frame, &dtrace_return_probe_ptr);
471 			return;
472 #endif
473 		}
474 	} else {
475 		/* kernel trap */
476 
477 		KASSERT(cold || td->td_ucred != NULL,
478 		    ("kernel trap doesn't have ucred"));
479 		switch (type) {
480 		case T_PAGEFLT:			/* page fault */
481 			(void)trap_pfault(frame, false, eva, NULL, NULL);
482 			return;
483 
484 		case T_DNA:
485 			if (PCB_USER_FPU(td->td_pcb))
486 				panic("Unregistered use of FPU in kernel");
487 			if (npxdna())
488 				return;
489 			break;
490 
491 		case T_ARITHTRAP:	/* arithmetic trap */
492 		case T_XMMFLT:		/* SIMD floating-point exception */
493 		case T_FPOPFLT:		/* FPU operand fetch fault */
494 			/*
495 			 * XXXKIB for now disable any FPU traps in kernel
496 			 * handler registration seems to be overkill
497 			 */
498 			trap_fatal(frame, 0);
499 			return;
500 
501 			/*
502 			 * The following two traps can happen in
503 			 * vm86 mode, and, if so, we want to handle
504 			 * them specially.
505 			 */
506 		case T_PROTFLT:		/* general protection fault */
507 		case T_STKFLT:		/* stack fault */
508 			if (frame->tf_eflags & PSL_VM) {
509 				signo = vm86_emulate((struct vm86frame *)frame);
510 				if (signo == SIGTRAP) {
511 					type = T_TRCTRAP;
512 					load_dr6(rdr6() | 0x4000);
513 					goto kernel_trctrap;
514 				}
515 				if (signo != 0)
516 					/*
517 					 * returns to original process
518 					 */
519 					vm86_trap((struct vm86frame *)frame);
520 				return;
521 			}
522 			/* FALL THROUGH */
523 		case T_SEGNPFLT:	/* segment not present fault */
524 			if (curpcb->pcb_flags & PCB_VM86CALL)
525 				break;
526 
527 			/*
528 			 * Invalid %fs's and %gs's can be created using
529 			 * procfs or PT_SETREGS or by invalidating the
530 			 * underlying LDT entry.  This causes a fault
531 			 * in kernel mode when the kernel attempts to
532 			 * switch contexts.  Lose the bad context
533 			 * (XXX) so that we can continue, and generate
534 			 * a signal.
535 			 */
536 			if (frame->tf_eip == (int)cpu_switch_load_gs) {
537 				curpcb->pcb_gs = 0;
538 #if 0
539 				PROC_LOCK(p);
540 				kern_psignal(p, SIGBUS);
541 				PROC_UNLOCK(p);
542 #endif
543 				return;
544 			}
545 
546 			if (td->td_intr_nesting_level != 0)
547 				break;
548 
549 			/*
550 			 * Invalid segment selectors and out of bounds
551 			 * %eip's and %esp's can be set up in user mode.
552 			 * This causes a fault in kernel mode when the
553 			 * kernel tries to return to user mode.  We want
554 			 * to get this fault so that we can fix the
555 			 * problem here and not have to check all the
556 			 * selectors and pointers when the user changes
557 			 * them.
558 			 *
559 			 * N.B. Comparing to long mode, 32-bit mode
560 			 * does not push %esp on the trap frame,
561 			 * because iretl faulted while in ring 0.  As
562 			 * the consequence, there is no need to fixup
563 			 * the stack pointer for doreti_iret_fault,
564 			 * the fixup and the complimentary trap() call
565 			 * are executed on the main thread stack, not
566 			 * on the trampoline stack.
567 			 */
568 			if (frame->tf_eip == (int)doreti_iret + setidt_disp) {
569 				frame->tf_eip = (int)doreti_iret_fault +
570 				    setidt_disp;
571 				return;
572 			}
573 			if (type == T_STKFLT)
574 				break;
575 
576 			if (frame->tf_eip == (int)doreti_popl_ds +
577 			    setidt_disp) {
578 				frame->tf_eip = (int)doreti_popl_ds_fault +
579 				    setidt_disp;
580 				return;
581 			}
582 			if (frame->tf_eip == (int)doreti_popl_es +
583 			    setidt_disp) {
584 				frame->tf_eip = (int)doreti_popl_es_fault +
585 				    setidt_disp;
586 				return;
587 			}
588 			if (frame->tf_eip == (int)doreti_popl_fs +
589 			    setidt_disp) {
590 				frame->tf_eip = (int)doreti_popl_fs_fault +
591 				    setidt_disp;
592 				return;
593 			}
594 			if (curpcb->pcb_onfault != NULL) {
595 				frame->tf_eip = (int)curpcb->pcb_onfault;
596 				return;
597 			}
598 			break;
599 
600 		case T_TSSFLT:
601 			/*
602 			 * PSL_NT can be set in user mode and isn't cleared
603 			 * automatically when the kernel is entered.  This
604 			 * causes a TSS fault when the kernel attempts to
605 			 * `iret' because the TSS link is uninitialized.  We
606 			 * want to get this fault so that we can fix the
607 			 * problem here and not every time the kernel is
608 			 * entered.
609 			 */
610 			if (frame->tf_eflags & PSL_NT) {
611 				frame->tf_eflags &= ~PSL_NT;
612 				return;
613 			}
614 			break;
615 
616 		case T_TRCTRAP:	 /* debug exception */
617 kernel_trctrap:
618 			/* Clear any pending debug events. */
619 			dr6 = rdr6();
620 			load_dr6(0);
621 
622 			/*
623 			 * Ignore debug register exceptions due to
624 			 * accesses in the user's address space, which
625 			 * can happen under several conditions such as
626 			 * if a user sets a watchpoint on a buffer and
627 			 * then passes that buffer to a system call.
628 			 * We still want to get TRCTRAPS for addresses
629 			 * in kernel space because that is useful when
630 			 * debugging the kernel.
631 			 */
632 			if (user_dbreg_trap(dr6) &&
633 			   !(curpcb->pcb_flags & PCB_VM86CALL))
634 				return;
635 
636 			/*
637 			 * Malicious user code can configure a debug
638 			 * register watchpoint to trap on data access
639 			 * to the top of stack and then execute 'pop
640 			 * %ss; int 3'.  Due to exception deferral for
641 			 * 'pop %ss', the CPU will not interrupt 'int
642 			 * 3' to raise the DB# exception for the debug
643 			 * register but will postpone the DB# until
644 			 * execution of the first instruction of the
645 			 * BP# handler (in kernel mode).  Normally the
646 			 * previous check would ignore DB# exceptions
647 			 * for watchpoints on user addresses raised in
648 			 * kernel mode.  However, some CPU errata
649 			 * include cases where DB# exceptions do not
650 			 * properly set bits in %dr6, e.g. Haswell
651 			 * HSD23 and Skylake-X SKZ24.
652 			 *
653 			 * A deferred DB# can also be raised on the
654 			 * first instructions of system call entry
655 			 * points or single-step traps via similar use
656 			 * of 'pop %ss' or 'mov xxx, %ss'.
657 			 */
658 			if (frame->tf_eip ==
659 			    (uintptr_t)IDTVEC(int0x80_syscall) + setidt_disp ||
660 			    frame->tf_eip == (uintptr_t)IDTVEC(bpt) +
661 			    setidt_disp ||
662 			    frame->tf_eip == (uintptr_t)IDTVEC(dbg) +
663 			    setidt_disp)
664 				return;
665 			/*
666 			 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
667 			 */
668 		case T_BPTFLT:
669 			/*
670 			 * If KDB is enabled, let it handle the debugger trap.
671 			 * Otherwise, debugger traps "can't happen".
672 			 */
673 #ifdef KDB
674 			if (kdb_trap(type, dr6, frame))
675 				return;
676 #endif
677 			break;
678 
679 		case T_NMI:
680 #ifdef POWERFAIL_NMI
681 			if (time_second - lastalert > 10) {
682 				log(LOG_WARNING, "NMI: power fail\n");
683 				sysbeep(880, SBT_1S);
684 				lastalert = time_second;
685 			}
686 			return;
687 #else /* !POWERFAIL_NMI */
688 			nmi_handle_intr(frame);
689 			return;
690 #endif /* POWERFAIL_NMI */
691 		}
692 
693 		trap_fatal(frame, eva);
694 		return;
695 	}
696 
697 	ksiginfo_init_trap(&ksi);
698 	ksi.ksi_signo = signo;
699 	ksi.ksi_code = ucode;
700 	ksi.ksi_addr = (void *)addr;
701 	ksi.ksi_trapno = type;
702 	if (uprintf_signal) {
703 		uprintf("pid %d comm %s: signal %d err %#x code %d type %d "
704 		    "addr %#x ss %#04x esp %#08x cs %#04x eip %#08x eax %#08x"
705 		    "<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
706 		    p->p_pid, p->p_comm, signo, frame->tf_err, ucode, type,
707 		    addr, frame->tf_ss, frame->tf_esp, frame->tf_cs,
708 		    frame->tf_eip, frame->tf_eax,
709 		    fubyte((void *)(frame->tf_eip + 0)),
710 		    fubyte((void *)(frame->tf_eip + 1)),
711 		    fubyte((void *)(frame->tf_eip + 2)),
712 		    fubyte((void *)(frame->tf_eip + 3)),
713 		    fubyte((void *)(frame->tf_eip + 4)),
714 		    fubyte((void *)(frame->tf_eip + 5)),
715 		    fubyte((void *)(frame->tf_eip + 6)),
716 		    fubyte((void *)(frame->tf_eip + 7)));
717 	}
718 	KASSERT((read_eflags() & PSL_I) != 0, ("interrupts disabled"));
719 	trapsignal(td, &ksi);
720 
721 user:
722 	userret(td, frame);
723 	KASSERT(PCB_USER_FPU(td->td_pcb),
724 	    ("Return from trap with kernel FPU ctx leaked"));
725 }
726 
727 /*
728  * Handle all details of a page fault.
729  * Returns:
730  * -2 if the fault was caused by triggered workaround for Intel Pentium
731  *    0xf00f bug.
732  * -1 if this fault was fatal, typically from kernel mode
733  *    (cannot happen, but we need to return something).
734  * 0  if this fault was handled by updating either the user or kernel
735  *    page table, execution can continue.
736  * 1  if this fault was from usermode and it was not handled, a synchronous
737  *    signal should be delivered to the thread.  *signo returns the signal
738  *    number, *ucode gives si_code.
739  */
740 static int
trap_pfault(struct trapframe * frame,bool usermode,vm_offset_t eva,int * signo,int * ucode)741 trap_pfault(struct trapframe *frame, bool usermode, vm_offset_t eva,
742     int *signo, int *ucode)
743 {
744 	struct thread *td;
745 	struct proc *p;
746 	vm_map_t map;
747 	int rv;
748 	vm_prot_t ftype;
749 
750 	MPASS(!usermode || (signo != NULL && ucode != NULL));
751 
752 	td = curthread;
753 	p = td->td_proc;
754 
755 	if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
756 		/*
757 		 * Due to both processor errata and lazy TLB invalidation when
758 		 * access restrictions are removed from virtual pages, memory
759 		 * accesses that are allowed by the physical mapping layer may
760 		 * nonetheless cause one spurious page fault per virtual page.
761 		 * When the thread is executing a "no faulting" section that
762 		 * is bracketed by vm_fault_{disable,enable}_pagefaults(),
763 		 * every page fault is treated as a spurious page fault,
764 		 * unless it accesses the same virtual address as the most
765 		 * recent page fault within the same "no faulting" section.
766 		 */
767 		if (td->td_md.md_spurflt_addr != eva ||
768 		    (td->td_pflags & TDP_RESETSPUR) != 0) {
769 			/*
770 			 * Do nothing to the TLB.  A stale TLB entry is
771 			 * flushed automatically by a page fault.
772 			 */
773 			td->td_md.md_spurflt_addr = eva;
774 			td->td_pflags &= ~TDP_RESETSPUR;
775 			return (0);
776 		}
777 	} else {
778 		/*
779 		 * If we get a page fault while in a critical section, then
780 		 * it is most likely a fatal kernel page fault.  The kernel
781 		 * is already going to panic trying to get a sleep lock to
782 		 * do the VM lookup, so just consider it a fatal trap so the
783 		 * kernel can print out a useful trap message and even get
784 		 * to the debugger.
785 		 *
786 		 * If we get a page fault while holding a non-sleepable
787 		 * lock, then it is most likely a fatal kernel page fault.
788 		 * If WITNESS is enabled, then it's going to whine about
789 		 * bogus LORs with various VM locks, so just skip to the
790 		 * fatal trap handling directly.
791 		 */
792 		if (td->td_critnest != 0 ||
793 		    WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
794 		    "Kernel page fault") != 0) {
795 			trap_fatal(frame, eva);
796 			return (-1);
797 		}
798 	}
799 	if (eva >= PMAP_TRM_MIN_ADDRESS) {
800 		/*
801 		 * Don't allow user-mode faults in kernel address space.
802 		 * An exception:  if the faulting address is the invalid
803 		 * instruction entry in the IDT, then the Intel Pentium
804 		 * F00F bug workaround was triggered, and we need to
805 		 * treat it is as an illegal instruction, and not a page
806 		 * fault.
807 		 */
808 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
809 		if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
810 			*ucode = ILL_PRVOPC;
811 			*signo = SIGILL;
812 			return (-2);
813 		}
814 #endif
815 		if (usermode) {
816 			*signo = SIGSEGV;
817 			*ucode = SEGV_MAPERR;
818 			return (1);
819 		}
820 		trap_fatal(frame, eva);
821 		return (-1);
822 	} else {
823 		map = usermode ? &p->p_vmspace->vm_map : kernel_map;
824 
825 		/*
826 		 * Kernel cannot access a user-space address directly
827 		 * because user pages are not mapped.  Also, page
828 		 * faults must not be caused during the interrupts.
829 		 */
830 		if (!usermode && td->td_intr_nesting_level != 0) {
831 			trap_fatal(frame, eva);
832 			return (-1);
833 		}
834 	}
835 
836 	/*
837 	 * If the trap was caused by errant bits in the PTE then panic.
838 	 */
839 	if (frame->tf_err & PGEX_RSV) {
840 		trap_fatal(frame, eva);
841 		return (-1);
842 	}
843 
844 	/*
845 	 * PGEX_I is defined only if the execute disable bit capability is
846 	 * supported and enabled.
847 	 */
848 	if (frame->tf_err & PGEX_W)
849 		ftype = VM_PROT_WRITE;
850 	else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
851 		ftype = VM_PROT_EXECUTE;
852 	else
853 		ftype = VM_PROT_READ;
854 
855 	/* Fault in the page. */
856 	rv = vm_fault_trap(map, eva, ftype, VM_FAULT_NORMAL, signo, ucode);
857 	if (rv == KERN_SUCCESS) {
858 #ifdef HWPMC_HOOKS
859 		if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
860 			PMC_SOFT_CALL_TF( , , page_fault, all, frame);
861 			if (ftype == VM_PROT_READ)
862 				PMC_SOFT_CALL_TF( , , page_fault, read,
863 				    frame);
864 			else
865 				PMC_SOFT_CALL_TF( , , page_fault, write,
866 				    frame);
867 		}
868 #endif
869 		return (0);
870 	}
871 	if (usermode)
872 		return (1);
873 	if (td->td_intr_nesting_level == 0 &&
874 	    curpcb->pcb_onfault != NULL) {
875 		frame->tf_eip = (int)curpcb->pcb_onfault;
876 		return (0);
877 	}
878 	trap_fatal(frame, eva);
879 	return (-1);
880 }
881 
882 static void
trap_fatal(struct trapframe * frame,vm_offset_t eva)883 trap_fatal(struct trapframe *frame, vm_offset_t eva)
884 {
885 	int code, ss, esp;
886 	u_int type;
887 	struct soft_segment_descriptor softseg;
888 #ifdef KDB
889 	bool handled;
890 #endif
891 
892 	code = frame->tf_err;
893 	type = frame->tf_trapno;
894 	sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
895 
896 	printf("\n\nFatal trap %d: %s while in %s mode\n", type, trap_msg(type),
897 	    frame->tf_eflags & PSL_VM ? "vm86" :
898 	    ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
899 #ifdef SMP
900 	/* two separate prints in case of a trap on an unmapped page */
901 	printf("cpuid = %d; ", PCPU_GET(cpuid));
902 	printf("apic id = %02x\n", PCPU_GET(apic_id));
903 #endif
904 	if (type == T_PAGEFLT) {
905 		printf("fault virtual address	= 0x%x\n", eva);
906 		printf("fault code		= %s %s%s, %s\n",
907 			code & PGEX_U ? "user" : "supervisor",
908 			code & PGEX_W ? "write" : "read",
909 			pg_nx != 0 ?
910 			(code & PGEX_I ? " instruction" : " data") :
911 			"",
912 			code & PGEX_RSV ? "reserved bits in PTE" :
913 			code & PGEX_P ? "protection violation" : "page not present");
914 	} else {
915 		printf("error code		= %#x\n", code);
916 	}
917 	printf("instruction pointer	= 0x%x:0x%x\n",
918 	       frame->tf_cs & 0xffff, frame->tf_eip);
919         if (TF_HAS_STACKREGS(frame)) {
920 		ss = frame->tf_ss & 0xffff;
921 		esp = frame->tf_esp;
922 	} else {
923 		ss = GSEL(GDATA_SEL, SEL_KPL);
924 		esp = (int)&frame->tf_esp;
925 	}
926 	printf("stack pointer	        = 0x%x:0x%x\n", ss, esp);
927 	printf("frame pointer	        = 0x%x:0x%x\n", ss, frame->tf_ebp);
928 	printf("code segment		= base 0x%x, limit 0x%x, type 0x%x\n",
929 	       softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
930 	printf("			= DPL %d, pres %d, def32 %d, gran %d\n",
931 	       softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
932 	       softseg.ssd_gran);
933 	printf("processor eflags	= ");
934 	if (frame->tf_eflags & PSL_T)
935 		printf("trace trap, ");
936 	if (frame->tf_eflags & PSL_I)
937 		printf("interrupt enabled, ");
938 	if (frame->tf_eflags & PSL_NT)
939 		printf("nested task, ");
940 	if (frame->tf_eflags & PSL_RF)
941 		printf("resume, ");
942 	if (frame->tf_eflags & PSL_VM)
943 		printf("vm86, ");
944 	printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
945 	printf("current process		= %d (%s)\n",
946 	    curproc->p_pid, curthread->td_name);
947 
948 #ifdef KDB
949 	if (debugger_on_trap) {
950 		kdb_why = KDB_WHY_TRAP;
951 		frame->tf_err = eva;	/* smuggle fault address to ddb */
952 		handled = kdb_trap(type, 0, frame);
953 		frame->tf_err = code;	/* restore error code */
954 		kdb_why = KDB_WHY_UNSET;
955 		if (handled)
956 			return;
957 	}
958 #endif
959 	printf("trap number		= %d\n", type);
960 	if (trap_msg(type) != NULL)
961 		panic("%s", trap_msg(type));
962 	else
963 		panic("unknown/reserved trap");
964 }
965 
966 #ifdef KDTRACE_HOOKS
967 /*
968  * Invoke a userspace DTrace hook.  The hook pointer is cleared when no
969  * userspace probes are enabled, so we must synchronize with DTrace to ensure
970  * that a trapping thread is able to call the hook before it is cleared.
971  */
972 static bool
trap_user_dtrace(struct trapframe * frame,int (** hookp)(struct trapframe *))973 trap_user_dtrace(struct trapframe *frame, int (**hookp)(struct trapframe *))
974 {
975 	int (*hook)(struct trapframe *);
976 
977 	hook = atomic_load_ptr(hookp);
978 	enable_intr();
979 	if (hook != NULL)
980 		return ((hook)(frame) == 0);
981 	return (false);
982 }
983 #endif
984 
985 /*
986  * Double fault handler. Called when a fault occurs while writing
987  * a frame for a trap/exception onto the stack. This usually occurs
988  * when the stack overflows (such is the case with infinite recursion,
989  * for example).
990  *
991  * XXX Note that the current PTD gets replaced by IdlePTD when the
992  * task switch occurs. This means that the stack that was active at
993  * the time of the double fault is not available at <kstack> unless
994  * the machine was idle when the double fault occurred. The downside
995  * of this is that "trace <ebp>" in ddb won't work.
996  */
997 void
dblfault_handler(void)998 dblfault_handler(void)
999 {
1000 	struct i386tss *t;
1001 
1002 #ifdef KDTRACE_HOOKS
1003 	if (dtrace_doubletrap_func != NULL)
1004 		(*dtrace_doubletrap_func)();
1005 #endif
1006 	printf("\nFatal double fault:\n");
1007 	t = PCPU_GET(common_tssp);
1008 	printf(
1009 	    "eip = %#08x esp = %#08x ebp = %#08x eax = %#08x\n"
1010 	    "edx = %#08x ecx = %#08x edi = %#08x esi = %#08x\n"
1011 	    "ebx = %#08x\n"
1012 	    "psl = %#08x cs  = %#08x ss  = %#08x ds  = %#08x\n"
1013 	    "es  = %#08x fs  = %#08x gs  = %#08x cr3 = %#08x\n",
1014 	    t->tss_eip, t->tss_esp, t->tss_ebp, t->tss_eax,
1015 	    t->tss_edx, t->tss_ecx, t->tss_edi, t->tss_esi,
1016 	    t->tss_ebx,
1017 	    t->tss_eflags, t->tss_cs, t->tss_ss, t->tss_ds,
1018 	    t->tss_es, t->tss_fs, t->tss_gs, t->tss_cr3);
1019 #ifdef SMP
1020 	printf("cpuid = %d; apic id = %02x\n", PCPU_GET(cpuid),
1021 	    PCPU_GET(apic_id));
1022 #endif
1023 	panic("double fault");
1024 }
1025 
1026 int
cpu_fetch_syscall_args(struct thread * td)1027 cpu_fetch_syscall_args(struct thread *td)
1028 {
1029 	struct proc *p;
1030 	struct trapframe *frame;
1031 	struct syscall_args *sa;
1032 	caddr_t params;
1033 	long tmp;
1034 	int error;
1035 #ifdef COMPAT_43
1036 	u_int32_t eip;
1037 	int cs;
1038 #endif
1039 
1040 	p = td->td_proc;
1041 	frame = td->td_frame;
1042 	sa = &td->td_sa;
1043 
1044 #ifdef COMPAT_43
1045 	if (__predict_false(frame->tf_cs == 7 && frame->tf_eip == 2)) {
1046 		/*
1047 		 * In lcall $7,$0 after int $0x80.  Convert the user
1048 		 * frame to what it would be for a direct int 0x80 instead
1049 		 * of lcall $7,$0, by popping the lcall return address.
1050 		 */
1051 		error = fueword32((void *)frame->tf_esp, &eip);
1052 		if (error == -1)
1053 			return (EFAULT);
1054 		cs = fuword16((void *)(frame->tf_esp + sizeof(u_int32_t)));
1055 		if (cs == -1)
1056 			return (EFAULT);
1057 
1058 		/*
1059 		 * Unwind in-kernel frame after all stack frame pieces
1060 		 * were successfully read.
1061 		 */
1062 		frame->tf_eip = eip;
1063 		frame->tf_cs = cs;
1064 		frame->tf_esp += 2 * sizeof(u_int32_t);
1065 		frame->tf_err = 7;	/* size of lcall $7,$0 */
1066 	}
1067 #endif
1068 
1069 	sa->code = frame->tf_eax;
1070 	sa->original_code = sa->code;
1071 	params = (caddr_t)frame->tf_esp + sizeof(uint32_t);
1072 
1073 	/*
1074 	 * Need to check if this is a 32 bit or 64 bit syscall.
1075 	 */
1076 	if (sa->code == SYS_syscall) {
1077 		/*
1078 		 * Code is first argument, followed by actual args.
1079 		 */
1080 		error = fueword(params, &tmp);
1081 		if (error == -1)
1082 			return (EFAULT);
1083 		sa->code = tmp;
1084 		params += sizeof(uint32_t);
1085 	} else if (sa->code == SYS___syscall) {
1086 		/*
1087 		 * Like syscall, but code is a quad, so as to maintain
1088 		 * quad alignment for the rest of the arguments.
1089 		 */
1090 		error = fueword(params, &tmp);
1091 		if (error == -1)
1092 			return (EFAULT);
1093 		sa->code = tmp;
1094 		params += sizeof(quad_t);
1095 	}
1096 
1097  	if (sa->code >= p->p_sysent->sv_size)
1098 		sa->callp = &nosys_sysent;
1099   	else
1100  		sa->callp = &p->p_sysent->sv_table[sa->code];
1101 
1102 	if (params != NULL && sa->callp->sy_narg != 0)
1103 		error = copyin(params, (caddr_t)sa->args,
1104 		    (u_int)(sa->callp->sy_narg * sizeof(uint32_t)));
1105 	else
1106 		error = 0;
1107 
1108 	if (error == 0) {
1109 		td->td_retval[0] = 0;
1110 		td->td_retval[1] = frame->tf_edx;
1111 	}
1112 
1113 	return (error);
1114 }
1115 
1116 #include "../../kern/subr_syscall.c"
1117 
1118 /*
1119  * syscall - system call request C handler.  A system call is
1120  * essentially treated as a trap by reusing the frame layout.
1121  */
1122 void
syscall(struct trapframe * frame)1123 syscall(struct trapframe *frame)
1124 {
1125 	struct thread *td;
1126 	register_t orig_tf_eflags;
1127 	ksiginfo_t ksi;
1128 
1129 #ifdef DIAGNOSTIC
1130 	if (!(TRAPF_USERMODE(frame) &&
1131 	    (curpcb->pcb_flags & PCB_VM86CALL) == 0)) {
1132 		panic("syscall");
1133 		/* NOT REACHED */
1134 	}
1135 #endif
1136 	trap_check_kstack();
1137 	orig_tf_eflags = frame->tf_eflags;
1138 
1139 	td = curthread;
1140 	td->td_frame = frame;
1141 
1142 	syscallenter(td);
1143 
1144 	/*
1145 	 * Traced syscall.
1146 	 */
1147 	if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1148 		frame->tf_eflags &= ~PSL_T;
1149 		ksiginfo_init_trap(&ksi);
1150 		ksi.ksi_signo = SIGTRAP;
1151 		ksi.ksi_code = TRAP_TRACE;
1152 		ksi.ksi_addr = (void *)frame->tf_eip;
1153 		trapsignal(td, &ksi);
1154 	}
1155 
1156 	KASSERT(PCB_USER_FPU(td->td_pcb),
1157 	    ("System call %s returning with kernel FPU ctx leaked",
1158 	     syscallname(td->td_proc, td->td_sa.code)));
1159 	KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td),
1160 	    ("System call %s returning with mangled pcb_save",
1161 	     syscallname(td->td_proc, td->td_sa.code)));
1162 
1163 	syscallret(td);
1164 }
1165