xref: /freebsd/sys/amd64/amd64/vm_machdep.c (revision 076ad2f8)
1 /*-
2  * Copyright (c) 1982, 1986 The Regents of the University of California.
3  * Copyright (c) 1989, 1990 William Jolitz
4  * Copyright (c) 1994 John Dyson
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * the Systems Programming Group of the University of Utah Computer
9  * Science Department, 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  *	from: @(#)vm_machdep.c	7.3 (Berkeley) 5/13/91
40  *	Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
41  */
42 
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
45 
46 #include "opt_isa.h"
47 #include "opt_cpu.h"
48 #include "opt_compat.h"
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/bio.h>
53 #include <sys/buf.h>
54 #include <sys/kernel.h>
55 #include <sys/ktr.h>
56 #include <sys/lock.h>
57 #include <sys/malloc.h>
58 #include <sys/mbuf.h>
59 #include <sys/mutex.h>
60 #include <sys/pioctl.h>
61 #include <sys/proc.h>
62 #include <sys/smp.h>
63 #include <sys/sysctl.h>
64 #include <sys/sysent.h>
65 #include <sys/unistd.h>
66 #include <sys/vnode.h>
67 #include <sys/vmmeter.h>
68 
69 #include <machine/cpu.h>
70 #include <machine/md_var.h>
71 #include <machine/pcb.h>
72 #include <machine/smp.h>
73 #include <machine/specialreg.h>
74 #include <machine/tss.h>
75 
76 #include <vm/vm.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_kern.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_param.h>
82 
83 #include <isa/isareg.h>
84 
85 static void	cpu_reset_real(void);
86 #ifdef SMP
87 static void	cpu_reset_proxy(void);
88 static u_int	cpu_reset_proxyid;
89 static volatile u_int	cpu_reset_proxy_active;
90 #endif
91 
92 _Static_assert(OFFSETOF_CURTHREAD == offsetof(struct pcpu, pc_curthread),
93     "OFFSETOF_CURTHREAD does not correspond with offset of pc_curthread.");
94 _Static_assert(OFFSETOF_CURPCB == offsetof(struct pcpu, pc_curpcb),
95     "OFFSETOF_CURPCB does not correspond with offset of pc_curpcb.");
96 _Static_assert(OFFSETOF_MONITORBUF == offsetof(struct pcpu, pc_monitorbuf),
97     "OFFSETOF_MONINORBUF does not correspond with offset of pc_monitorbuf.");
98 
99 struct savefpu *
100 get_pcb_user_save_td(struct thread *td)
101 {
102 	vm_offset_t p;
103 
104 	p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
105 	    roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN);
106 	KASSERT((p % XSAVE_AREA_ALIGN) == 0, ("Unaligned pcb_user_save area"));
107 	return ((struct savefpu *)p);
108 }
109 
110 struct savefpu *
111 get_pcb_user_save_pcb(struct pcb *pcb)
112 {
113 	vm_offset_t p;
114 
115 	p = (vm_offset_t)(pcb + 1);
116 	return ((struct savefpu *)p);
117 }
118 
119 struct pcb *
120 get_pcb_td(struct thread *td)
121 {
122 	vm_offset_t p;
123 
124 	p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
125 	    roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN) -
126 	    sizeof(struct pcb);
127 	return ((struct pcb *)p);
128 }
129 
130 void *
131 alloc_fpusave(int flags)
132 {
133 	void *res;
134 	struct savefpu_ymm *sf;
135 
136 	res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags);
137 	if (use_xsave) {
138 		sf = (struct savefpu_ymm *)res;
139 		bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd));
140 		sf->sv_xstate.sx_hd.xstate_bv = xsave_mask;
141 	}
142 	return (res);
143 }
144 
145 /*
146  * Finish a fork operation, with process p2 nearly set up.
147  * Copy and update the pcb, set up the stack so that the child
148  * ready to run and return to user mode.
149  */
150 void
151 cpu_fork(td1, p2, td2, flags)
152 	register struct thread *td1;
153 	register struct proc *p2;
154 	struct thread *td2;
155 	int flags;
156 {
157 	register struct proc *p1;
158 	struct pcb *pcb2;
159 	struct mdproc *mdp1, *mdp2;
160 	struct proc_ldt *pldt;
161 
162 	p1 = td1->td_proc;
163 	if ((flags & RFPROC) == 0) {
164 		if ((flags & RFMEM) == 0) {
165 			/* unshare user LDT */
166 			mdp1 = &p1->p_md;
167 			mtx_lock(&dt_lock);
168 			if ((pldt = mdp1->md_ldt) != NULL &&
169 			    pldt->ldt_refcnt > 1 &&
170 			    user_ldt_alloc(p1, 1) == NULL)
171 				panic("could not copy LDT");
172 			mtx_unlock(&dt_lock);
173 		}
174 		return;
175 	}
176 
177 	/* Ensure that td1's pcb is up to date. */
178 	fpuexit(td1);
179 
180 	/* Point the pcb to the top of the stack */
181 	pcb2 = get_pcb_td(td2);
182 	td2->td_pcb = pcb2;
183 
184 	/* Copy td1's pcb */
185 	bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
186 
187 	/* Properly initialize pcb_save */
188 	pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
189 	bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2),
190 	    cpu_max_ext_state_size);
191 
192 	/* Point mdproc and then copy over td1's contents */
193 	mdp2 = &p2->p_md;
194 	bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
195 
196 	/*
197 	 * Create a new fresh stack for the new process.
198 	 * Copy the trap frame for the return to user mode as if from a
199 	 * syscall.  This copies most of the user mode register values.
200 	 */
201 	td2->td_frame = (struct trapframe *)td2->td_pcb - 1;
202 	bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
203 
204 	td2->td_frame->tf_rax = 0;		/* Child returns zero */
205 	td2->td_frame->tf_rflags &= ~PSL_C;	/* success */
206 	td2->td_frame->tf_rdx = 1;
207 
208 	/*
209 	 * If the parent process has the trap bit set (i.e. a debugger had
210 	 * single stepped the process to the system call), we need to clear
211 	 * the trap flag from the new frame unless the debugger had set PF_FORK
212 	 * on the parent.  Otherwise, the child will receive a (likely
213 	 * unexpected) SIGTRAP when it executes the first instruction after
214 	 * returning  to userland.
215 	 */
216 	if ((p1->p_pfsflags & PF_FORK) == 0)
217 		td2->td_frame->tf_rflags &= ~PSL_T;
218 
219 	/*
220 	 * Set registers for trampoline to user mode.  Leave space for the
221 	 * return address on stack.  These are the kernel mode register values.
222 	 */
223 	pcb2->pcb_r12 = (register_t)fork_return;	/* fork_trampoline argument */
224 	pcb2->pcb_rbp = 0;
225 	pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *);
226 	pcb2->pcb_rbx = (register_t)td2;		/* fork_trampoline argument */
227 	pcb2->pcb_rip = (register_t)fork_trampoline;
228 	/*-
229 	 * pcb2->pcb_dr*:	cloned above.
230 	 * pcb2->pcb_savefpu:	cloned above.
231 	 * pcb2->pcb_flags:	cloned above.
232 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
233 	 * pcb2->pcb_[fg]sbase:	cloned above
234 	 */
235 
236 	/* Setup to release spin count in fork_exit(). */
237 	td2->td_md.md_spinlock_count = 1;
238 	td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
239 	td2->td_md.md_invl_gen.gen = 0;
240 
241 	/* As an i386, do not copy io permission bitmap. */
242 	pcb2->pcb_tssp = NULL;
243 
244 	/* New segment registers. */
245 	set_pcb_flags(pcb2, PCB_FULL_IRET);
246 
247 	/* Copy the LDT, if necessary. */
248 	mdp1 = &td1->td_proc->p_md;
249 	mdp2 = &p2->p_md;
250 	mtx_lock(&dt_lock);
251 	if (mdp1->md_ldt != NULL) {
252 		if (flags & RFMEM) {
253 			mdp1->md_ldt->ldt_refcnt++;
254 			mdp2->md_ldt = mdp1->md_ldt;
255 			bcopy(&mdp1->md_ldt_sd, &mdp2->md_ldt_sd, sizeof(struct
256 			    system_segment_descriptor));
257 		} else {
258 			mdp2->md_ldt = NULL;
259 			mdp2->md_ldt = user_ldt_alloc(p2, 0);
260 			if (mdp2->md_ldt == NULL)
261 				panic("could not copy LDT");
262 			amd64_set_ldt_data(td2, 0, max_ldt_segment,
263 			    (struct user_segment_descriptor *)
264 			    mdp1->md_ldt->ldt_base);
265 		}
266 	} else
267 		mdp2->md_ldt = NULL;
268 	mtx_unlock(&dt_lock);
269 
270 	/*
271 	 * Now, cpu_switch() can schedule the new process.
272 	 * pcb_rsp is loaded pointing to the cpu_switch() stack frame
273 	 * containing the return address when exiting cpu_switch.
274 	 * This will normally be to fork_trampoline(), which will have
275 	 * %ebx loaded with the new proc's pointer.  fork_trampoline()
276 	 * will set up a stack to call fork_return(p, frame); to complete
277 	 * the return to user-mode.
278 	 */
279 }
280 
281 /*
282  * Intercept the return address from a freshly forked process that has NOT
283  * been scheduled yet.
284  *
285  * This is needed to make kernel threads stay in kernel mode.
286  */
287 void
288 cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg)
289 {
290 	/*
291 	 * Note that the trap frame follows the args, so the function
292 	 * is really called like this:  func(arg, frame);
293 	 */
294 	td->td_pcb->pcb_r12 = (long) func;	/* function */
295 	td->td_pcb->pcb_rbx = (long) arg;	/* first arg */
296 }
297 
298 void
299 cpu_exit(struct thread *td)
300 {
301 
302 	/*
303 	 * If this process has a custom LDT, release it.
304 	 */
305 	mtx_lock(&dt_lock);
306 	if (td->td_proc->p_md.md_ldt != 0)
307 		user_ldt_free(td);
308 	else
309 		mtx_unlock(&dt_lock);
310 }
311 
312 void
313 cpu_thread_exit(struct thread *td)
314 {
315 	struct pcb *pcb;
316 
317 	critical_enter();
318 	if (td == PCPU_GET(fpcurthread))
319 		fpudrop();
320 	critical_exit();
321 
322 	pcb = td->td_pcb;
323 
324 	/* Disable any hardware breakpoints. */
325 	if (pcb->pcb_flags & PCB_DBREGS) {
326 		reset_dbregs();
327 		clear_pcb_flags(pcb, PCB_DBREGS);
328 	}
329 }
330 
331 void
332 cpu_thread_clean(struct thread *td)
333 {
334 	struct pcb *pcb;
335 
336 	pcb = td->td_pcb;
337 
338 	/*
339 	 * Clean TSS/iomap
340 	 */
341 	if (pcb->pcb_tssp != NULL) {
342 		kmem_free(kernel_arena, (vm_offset_t)pcb->pcb_tssp,
343 		    ctob(IOPAGES + 1));
344 		pcb->pcb_tssp = NULL;
345 	}
346 }
347 
348 void
349 cpu_thread_swapin(struct thread *td)
350 {
351 }
352 
353 void
354 cpu_thread_swapout(struct thread *td)
355 {
356 }
357 
358 void
359 cpu_thread_alloc(struct thread *td)
360 {
361 	struct pcb *pcb;
362 	struct xstate_hdr *xhdr;
363 
364 	td->td_pcb = pcb = get_pcb_td(td);
365 	td->td_frame = (struct trapframe *)pcb - 1;
366 	pcb->pcb_save = get_pcb_user_save_pcb(pcb);
367 	if (use_xsave) {
368 		xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
369 		bzero(xhdr, sizeof(*xhdr));
370 		xhdr->xstate_bv = xsave_mask;
371 	}
372 }
373 
374 void
375 cpu_thread_free(struct thread *td)
376 {
377 
378 	cpu_thread_clean(td);
379 }
380 
381 void
382 cpu_set_syscall_retval(struct thread *td, int error)
383 {
384 
385 	switch (error) {
386 	case 0:
387 		td->td_frame->tf_rax = td->td_retval[0];
388 		td->td_frame->tf_rdx = td->td_retval[1];
389 		td->td_frame->tf_rflags &= ~PSL_C;
390 		break;
391 
392 	case ERESTART:
393 		/*
394 		 * Reconstruct pc, we know that 'syscall' is 2 bytes,
395 		 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
396 		 * We saved this in tf_err.
397 		 * %r10 (which was holding the value of %rcx) is restored
398 		 * for the next iteration.
399 		 * %r10 restore is only required for freebsd/amd64 processes,
400 		 * but shall be innocent for any ia32 ABI.
401 		 *
402 		 * Require full context restore to get the arguments
403 		 * in the registers reloaded at return to usermode.
404 		 */
405 		td->td_frame->tf_rip -= td->td_frame->tf_err;
406 		td->td_frame->tf_r10 = td->td_frame->tf_rcx;
407 		set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
408 		break;
409 
410 	case EJUSTRETURN:
411 		break;
412 
413 	default:
414 		td->td_frame->tf_rax = SV_ABI_ERRNO(td->td_proc, error);
415 		td->td_frame->tf_rflags |= PSL_C;
416 		break;
417 	}
418 }
419 
420 /*
421  * Initialize machine state, mostly pcb and trap frame for a new
422  * thread, about to return to userspace.  Put enough state in the new
423  * thread's PCB to get it to go back to the fork_return(), which
424  * finalizes the thread state and handles peculiarities of the first
425  * return to userspace for the new thread.
426  */
427 void
428 cpu_copy_thread(struct thread *td, struct thread *td0)
429 {
430 	struct pcb *pcb2;
431 
432 	/* Point the pcb to the top of the stack. */
433 	pcb2 = td->td_pcb;
434 
435 	/*
436 	 * Copy the upcall pcb.  This loads kernel regs.
437 	 * Those not loaded individually below get their default
438 	 * values here.
439 	 */
440 	bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
441 	clear_pcb_flags(pcb2, PCB_FPUINITDONE | PCB_USERFPUINITDONE |
442 	    PCB_KERNFPU);
443 	pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
444 	bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save,
445 	    cpu_max_ext_state_size);
446 	set_pcb_flags(pcb2, PCB_FULL_IRET);
447 
448 	/*
449 	 * Create a new fresh stack for the new thread.
450 	 */
451 	bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
452 
453 	/* If the current thread has the trap bit set (i.e. a debugger had
454 	 * single stepped the process to the system call), we need to clear
455 	 * the trap flag from the new frame. Otherwise, the new thread will
456 	 * receive a (likely unexpected) SIGTRAP when it executes the first
457 	 * instruction after returning to userland.
458 	 */
459 	td->td_frame->tf_rflags &= ~PSL_T;
460 
461 	/*
462 	 * Set registers for trampoline to user mode.  Leave space for the
463 	 * return address on stack.  These are the kernel mode register values.
464 	 */
465 	pcb2->pcb_r12 = (register_t)fork_return;	    /* trampoline arg */
466 	pcb2->pcb_rbp = 0;
467 	pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *);	/* trampoline arg */
468 	pcb2->pcb_rbx = (register_t)td;			    /* trampoline arg */
469 	pcb2->pcb_rip = (register_t)fork_trampoline;
470 	/*
471 	 * If we didn't copy the pcb, we'd need to do the following registers:
472 	 * pcb2->pcb_dr*:	cloned above.
473 	 * pcb2->pcb_savefpu:	cloned above.
474 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
475 	 * pcb2->pcb_[fg]sbase: cloned above
476 	 */
477 
478 	/* Setup to release spin count in fork_exit(). */
479 	td->td_md.md_spinlock_count = 1;
480 	td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
481 }
482 
483 /*
484  * Set that machine state for performing an upcall that starts
485  * the entry function with the given argument.
486  */
487 void
488 cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg,
489     stack_t *stack)
490 {
491 
492 	/*
493 	 * Do any extra cleaning that needs to be done.
494 	 * The thread may have optional components
495 	 * that are not present in a fresh thread.
496 	 * This may be a recycled thread so make it look
497 	 * as though it's newly allocated.
498 	 */
499 	cpu_thread_clean(td);
500 
501 #ifdef COMPAT_FREEBSD32
502 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
503 		/*
504 		 * Set the trap frame to point at the beginning of the entry
505 		 * function.
506 		 */
507 		td->td_frame->tf_rbp = 0;
508 		td->td_frame->tf_rsp =
509 		   (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
510 		td->td_frame->tf_rip = (uintptr_t)entry;
511 
512 		/* Pass the argument to the entry point. */
513 		suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)),
514 		    (uint32_t)(uintptr_t)arg);
515 
516 		return;
517 	}
518 #endif
519 
520 	/*
521 	 * Set the trap frame to point at the beginning of the uts
522 	 * function.
523 	 */
524 	td->td_frame->tf_rbp = 0;
525 	td->td_frame->tf_rsp =
526 	    ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f;
527 	td->td_frame->tf_rsp -= 8;
528 	td->td_frame->tf_rip = (register_t)entry;
529 	td->td_frame->tf_ds = _udatasel;
530 	td->td_frame->tf_es = _udatasel;
531 	td->td_frame->tf_fs = _ufssel;
532 	td->td_frame->tf_gs = _ugssel;
533 	td->td_frame->tf_flags = TF_HASSEGS;
534 
535 	/* Pass the argument to the entry point. */
536 	td->td_frame->tf_rdi = (register_t)arg;
537 }
538 
539 int
540 cpu_set_user_tls(struct thread *td, void *tls_base)
541 {
542 	struct pcb *pcb;
543 
544 	if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS)
545 		return (EINVAL);
546 
547 	pcb = td->td_pcb;
548 	set_pcb_flags(pcb, PCB_FULL_IRET);
549 #ifdef COMPAT_FREEBSD32
550 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
551 		pcb->pcb_gsbase = (register_t)tls_base;
552 		return (0);
553 	}
554 #endif
555 	pcb->pcb_fsbase = (register_t)tls_base;
556 	return (0);
557 }
558 
559 #ifdef SMP
560 static void
561 cpu_reset_proxy()
562 {
563 	cpuset_t tcrp;
564 
565 	cpu_reset_proxy_active = 1;
566 	while (cpu_reset_proxy_active == 1)
567 		ia32_pause(); /* Wait for other cpu to see that we've started */
568 
569 	CPU_SETOF(cpu_reset_proxyid, &tcrp);
570 	stop_cpus(tcrp);
571 	printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
572 	DELAY(1000000);
573 	cpu_reset_real();
574 }
575 #endif
576 
577 void
578 cpu_reset()
579 {
580 #ifdef SMP
581 	cpuset_t map;
582 	u_int cnt;
583 
584 	if (smp_started) {
585 		map = all_cpus;
586 		CPU_CLR(PCPU_GET(cpuid), &map);
587 		CPU_NAND(&map, &stopped_cpus);
588 		if (!CPU_EMPTY(&map)) {
589 			printf("cpu_reset: Stopping other CPUs\n");
590 			stop_cpus(map);
591 		}
592 
593 		if (PCPU_GET(cpuid) != 0) {
594 			cpu_reset_proxyid = PCPU_GET(cpuid);
595 			cpustop_restartfunc = cpu_reset_proxy;
596 			cpu_reset_proxy_active = 0;
597 			printf("cpu_reset: Restarting BSP\n");
598 
599 			/* Restart CPU #0. */
600 			CPU_SETOF(0, &started_cpus);
601 			wmb();
602 
603 			cnt = 0;
604 			while (cpu_reset_proxy_active == 0 && cnt < 10000000) {
605 				ia32_pause();
606 				cnt++;	/* Wait for BSP to announce restart */
607 			}
608 			if (cpu_reset_proxy_active == 0)
609 				printf("cpu_reset: Failed to restart BSP\n");
610 			enable_intr();
611 			cpu_reset_proxy_active = 2;
612 
613 			while (1)
614 				ia32_pause();
615 			/* NOTREACHED */
616 		}
617 
618 		DELAY(1000000);
619 	}
620 #endif
621 	cpu_reset_real();
622 	/* NOTREACHED */
623 }
624 
625 static void
626 cpu_reset_real()
627 {
628 	struct region_descriptor null_idt;
629 	int b;
630 
631 	disable_intr();
632 
633 	/*
634 	 * Attempt to do a CPU reset via the keyboard controller,
635 	 * do not turn off GateA20, as any machine that fails
636 	 * to do the reset here would then end up in no man's land.
637 	 */
638 	outb(IO_KBD + 4, 0xFE);
639 	DELAY(500000);	/* wait 0.5 sec to see if that did it */
640 
641 	/*
642 	 * Attempt to force a reset via the Reset Control register at
643 	 * I/O port 0xcf9.  Bit 2 forces a system reset when it
644 	 * transitions from 0 to 1.  Bit 1 selects the type of reset
645 	 * to attempt: 0 selects a "soft" reset, and 1 selects a
646 	 * "hard" reset.  We try a "hard" reset.  The first write sets
647 	 * bit 1 to select a "hard" reset and clears bit 2.  The
648 	 * second write forces a 0 -> 1 transition in bit 2 to trigger
649 	 * a reset.
650 	 */
651 	outb(0xcf9, 0x2);
652 	outb(0xcf9, 0x6);
653 	DELAY(500000);  /* wait 0.5 sec to see if that did it */
654 
655 	/*
656 	 * Attempt to force a reset via the Fast A20 and Init register
657 	 * at I/O port 0x92.  Bit 1 serves as an alternate A20 gate.
658 	 * Bit 0 asserts INIT# when set to 1.  We are careful to only
659 	 * preserve bit 1 while setting bit 0.  We also must clear bit
660 	 * 0 before setting it if it isn't already clear.
661 	 */
662 	b = inb(0x92);
663 	if (b != 0xff) {
664 		if ((b & 0x1) != 0)
665 			outb(0x92, b & 0xfe);
666 		outb(0x92, b | 0x1);
667 		DELAY(500000);  /* wait 0.5 sec to see if that did it */
668 	}
669 
670 	printf("No known reset method worked, attempting CPU shutdown\n");
671 	DELAY(1000000);	/* wait 1 sec for printf to complete */
672 
673 	/* Wipe the IDT. */
674 	null_idt.rd_limit = 0;
675 	null_idt.rd_base = 0;
676 	lidt(&null_idt);
677 
678 	/* "good night, sweet prince .... <THUNK!>" */
679 	breakpoint();
680 
681 	/* NOTREACHED */
682 	while(1);
683 }
684 
685 /*
686  * Software interrupt handler for queued VM system processing.
687  */
688 void
689 swi_vm(void *dummy)
690 {
691 	if (busdma_swi_pending != 0)
692 		busdma_swi();
693 }
694 
695 /*
696  * Tell whether this address is in some physical memory region.
697  * Currently used by the kernel coredump code in order to avoid
698  * dumping the ``ISA memory hole'' which could cause indefinite hangs,
699  * or other unpredictable behaviour.
700  */
701 
702 int
703 is_physical_memory(vm_paddr_t addr)
704 {
705 
706 #ifdef DEV_ISA
707 	/* The ISA ``memory hole''. */
708 	if (addr >= 0xa0000 && addr < 0x100000)
709 		return 0;
710 #endif
711 
712 	/*
713 	 * stuff other tests for known memory-mapped devices (PCI?)
714 	 * here
715 	 */
716 
717 	return 1;
718 }
719