vm_machdep.c revision 331643
1/*-
2 * SPDX-License-Identifier: BSD-4-Clause
3 *
4 * Copyright (c) 1982, 1986 The Regents of the University of California.
5 * Copyright (c) 1989, 1990 William Jolitz
6 * Copyright (c) 1994 John Dyson
7 * All rights reserved.
8 *
9 * This code is derived from software contributed to Berkeley by
10 * the Systems Programming Group of the University of Utah Computer
11 * Science Department, and William Jolitz.
12 *
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
16 * 1. Redistributions of source code must retain the above copyright
17 *    notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 *    notice, this list of conditions and the following disclaimer in the
20 *    documentation and/or other materials provided with the distribution.
21 * 3. All advertising materials mentioning features or use of this software
22 *    must display the following acknowledgement:
23 *	This product includes software developed by the University of
24 *	California, Berkeley and its contributors.
25 * 4. Neither the name of the University nor the names of its contributors
26 *    may be used to endorse or promote products derived from this software
27 *    without specific prior written permission.
28 *
29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 * SUCH DAMAGE.
40 *
41 *	from: @(#)vm_machdep.c	7.3 (Berkeley) 5/13/91
42 *	Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
43 */
44
45#include <sys/cdefs.h>
46__FBSDID("$FreeBSD: stable/11/sys/amd64/amd64/vm_machdep.c 331643 2018-03-27 18:52:27Z dim $");
47
48#include "opt_isa.h"
49#include "opt_cpu.h"
50#include "opt_compat.h"
51
52#include <sys/param.h>
53#include <sys/systm.h>
54#include <sys/bio.h>
55#include <sys/buf.h>
56#include <sys/kernel.h>
57#include <sys/ktr.h>
58#include <sys/lock.h>
59#include <sys/malloc.h>
60#include <sys/mbuf.h>
61#include <sys/mutex.h>
62#include <sys/pioctl.h>
63#include <sys/proc.h>
64#include <sys/smp.h>
65#include <sys/sysctl.h>
66#include <sys/sysent.h>
67#include <sys/unistd.h>
68#include <sys/vnode.h>
69#include <sys/vmmeter.h>
70
71#include <machine/cpu.h>
72#include <machine/md_var.h>
73#include <machine/pcb.h>
74#include <machine/smp.h>
75#include <machine/specialreg.h>
76#include <machine/tss.h>
77
78#include <vm/vm.h>
79#include <vm/vm_extern.h>
80#include <vm/vm_kern.h>
81#include <vm/vm_page.h>
82#include <vm/vm_map.h>
83#include <vm/vm_param.h>
84
85#include <isa/isareg.h>
86
87static void	cpu_reset_real(void);
88#ifdef SMP
89static void	cpu_reset_proxy(void);
90static u_int	cpu_reset_proxyid;
91static volatile u_int	cpu_reset_proxy_active;
92#endif
93
94_Static_assert(OFFSETOF_CURTHREAD == offsetof(struct pcpu, pc_curthread),
95    "OFFSETOF_CURTHREAD does not correspond with offset of pc_curthread.");
96_Static_assert(OFFSETOF_CURPCB == offsetof(struct pcpu, pc_curpcb),
97    "OFFSETOF_CURPCB does not correspond with offset of pc_curpcb.");
98_Static_assert(OFFSETOF_MONITORBUF == offsetof(struct pcpu, pc_monitorbuf),
99    "OFFSETOF_MONINORBUF does not correspond with offset of pc_monitorbuf.");
100
101struct savefpu *
102get_pcb_user_save_td(struct thread *td)
103{
104	vm_offset_t p;
105
106	p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
107	    roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN);
108	KASSERT((p % XSAVE_AREA_ALIGN) == 0, ("Unaligned pcb_user_save area"));
109	return ((struct savefpu *)p);
110}
111
112struct savefpu *
113get_pcb_user_save_pcb(struct pcb *pcb)
114{
115	vm_offset_t p;
116
117	p = (vm_offset_t)(pcb + 1);
118	return ((struct savefpu *)p);
119}
120
121struct pcb *
122get_pcb_td(struct thread *td)
123{
124	vm_offset_t p;
125
126	p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
127	    roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN) -
128	    sizeof(struct pcb);
129	return ((struct pcb *)p);
130}
131
132void *
133alloc_fpusave(int flags)
134{
135	void *res;
136	struct savefpu_ymm *sf;
137
138	res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags);
139	if (use_xsave) {
140		sf = (struct savefpu_ymm *)res;
141		bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd));
142		sf->sv_xstate.sx_hd.xstate_bv = xsave_mask;
143	}
144	return (res);
145}
146
147/*
148 * Finish a fork operation, with process p2 nearly set up.
149 * Copy and update the pcb, set up the stack so that the child
150 * ready to run and return to user mode.
151 */
152void
153cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
154{
155	struct proc *p1;
156	struct pcb *pcb2;
157	struct mdproc *mdp1, *mdp2;
158	struct proc_ldt *pldt;
159
160	p1 = td1->td_proc;
161	if ((flags & RFPROC) == 0) {
162		if ((flags & RFMEM) == 0) {
163			/* unshare user LDT */
164			mdp1 = &p1->p_md;
165			mtx_lock(&dt_lock);
166			if ((pldt = mdp1->md_ldt) != NULL &&
167			    pldt->ldt_refcnt > 1 &&
168			    user_ldt_alloc(p1, 1) == NULL)
169				panic("could not copy LDT");
170			mtx_unlock(&dt_lock);
171		}
172		return;
173	}
174
175	/* Ensure that td1's pcb is up to date. */
176	fpuexit(td1);
177	update_pcb_bases(td1->td_pcb);
178
179	/* Point the pcb to the top of the stack */
180	pcb2 = get_pcb_td(td2);
181	td2->td_pcb = pcb2;
182
183	/* Copy td1's pcb */
184	bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
185
186	/* Properly initialize pcb_save */
187	pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
188	bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2),
189	    cpu_max_ext_state_size);
190
191	/* Point mdproc and then copy over td1's contents */
192	mdp2 = &p2->p_md;
193	bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
194
195	/*
196	 * Create a new fresh stack for the new process.
197	 * Copy the trap frame for the return to user mode as if from a
198	 * syscall.  This copies most of the user mode register values.
199	 */
200	td2->td_frame = (struct trapframe *)td2->td_pcb - 1;
201	bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
202
203	td2->td_frame->tf_rax = 0;		/* Child returns zero */
204	td2->td_frame->tf_rflags &= ~PSL_C;	/* success */
205	td2->td_frame->tf_rdx = 1;
206
207	/*
208	 * If the parent process has the trap bit set (i.e. a debugger had
209	 * single stepped the process to the system call), we need to clear
210	 * the trap flag from the new frame unless the debugger had set PF_FORK
211	 * on the parent.  Otherwise, the child will receive a (likely
212	 * unexpected) SIGTRAP when it executes the first instruction after
213	 * returning  to userland.
214	 */
215	if ((p1->p_pfsflags & PF_FORK) == 0)
216		td2->td_frame->tf_rflags &= ~PSL_T;
217
218	/*
219	 * Set registers for trampoline to user mode.  Leave space for the
220	 * return address on stack.  These are the kernel mode register values.
221	 */
222	pcb2->pcb_r12 = (register_t)fork_return;	/* fork_trampoline argument */
223	pcb2->pcb_rbp = 0;
224	pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *);
225	pcb2->pcb_rbx = (register_t)td2;		/* fork_trampoline argument */
226	pcb2->pcb_rip = (register_t)fork_trampoline;
227	/*-
228	 * pcb2->pcb_dr*:	cloned above.
229	 * pcb2->pcb_savefpu:	cloned above.
230	 * pcb2->pcb_flags:	cloned above.
231	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
232	 * pcb2->pcb_[fg]sbase:	cloned above
233	 */
234
235	/* Setup to release spin count in fork_exit(). */
236	td2->td_md.md_spinlock_count = 1;
237	td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
238	td2->td_md.md_invl_gen.gen = 0;
239
240	/* As an i386, do not copy io permission bitmap. */
241	pcb2->pcb_tssp = NULL;
242
243	/* New segment registers. */
244	set_pcb_flags_raw(pcb2, PCB_FULL_IRET);
245
246	/* Copy the LDT, if necessary. */
247	mdp1 = &td1->td_proc->p_md;
248	mdp2 = &p2->p_md;
249	mtx_lock(&dt_lock);
250	if (mdp1->md_ldt != NULL) {
251		if (flags & RFMEM) {
252			mdp1->md_ldt->ldt_refcnt++;
253			mdp2->md_ldt = mdp1->md_ldt;
254			bcopy(&mdp1->md_ldt_sd, &mdp2->md_ldt_sd, sizeof(struct
255			    system_segment_descriptor));
256		} else {
257			mdp2->md_ldt = NULL;
258			mdp2->md_ldt = user_ldt_alloc(p2, 0);
259			if (mdp2->md_ldt == NULL)
260				panic("could not copy LDT");
261			amd64_set_ldt_data(td2, 0, max_ldt_segment,
262			    (struct user_segment_descriptor *)
263			    mdp1->md_ldt->ldt_base);
264		}
265	} else
266		mdp2->md_ldt = NULL;
267	mtx_unlock(&dt_lock);
268
269	/*
270	 * Now, cpu_switch() can schedule the new process.
271	 * pcb_rsp is loaded pointing to the cpu_switch() stack frame
272	 * containing the return address when exiting cpu_switch.
273	 * This will normally be to fork_trampoline(), which will have
274	 * %ebx loaded with the new proc's pointer.  fork_trampoline()
275	 * will set up a stack to call fork_return(p, frame); to complete
276	 * the return to user-mode.
277	 */
278}
279
280/*
281 * Intercept the return address from a freshly forked process that has NOT
282 * been scheduled yet.
283 *
284 * This is needed to make kernel threads stay in kernel mode.
285 */
286void
287cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg)
288{
289	/*
290	 * Note that the trap frame follows the args, so the function
291	 * is really called like this:  func(arg, frame);
292	 */
293	td->td_pcb->pcb_r12 = (long) func;	/* function */
294	td->td_pcb->pcb_rbx = (long) arg;	/* first arg */
295}
296
297void
298cpu_exit(struct thread *td)
299{
300
301	/*
302	 * If this process has a custom LDT, release it.
303	 */
304	mtx_lock(&dt_lock);
305	if (td->td_proc->p_md.md_ldt != 0)
306		user_ldt_free(td);
307	else
308		mtx_unlock(&dt_lock);
309}
310
311void
312cpu_thread_exit(struct thread *td)
313{
314	struct pcb *pcb;
315
316	critical_enter();
317	if (td == PCPU_GET(fpcurthread))
318		fpudrop();
319	critical_exit();
320
321	pcb = td->td_pcb;
322
323	/* Disable any hardware breakpoints. */
324	if (pcb->pcb_flags & PCB_DBREGS) {
325		reset_dbregs();
326		clear_pcb_flags(pcb, PCB_DBREGS);
327	}
328}
329
330void
331cpu_thread_clean(struct thread *td)
332{
333	struct pcb *pcb;
334
335	pcb = td->td_pcb;
336
337	/*
338	 * Clean TSS/iomap
339	 */
340	if (pcb->pcb_tssp != NULL) {
341		pmap_pti_remove_kva((vm_offset_t)pcb->pcb_tssp,
342		    (vm_offset_t)pcb->pcb_tssp + ctob(IOPAGES + 1));
343		kmem_free(kernel_arena, (vm_offset_t)pcb->pcb_tssp,
344		    ctob(IOPAGES + 1));
345		pcb->pcb_tssp = NULL;
346	}
347}
348
349void
350cpu_thread_swapin(struct thread *td)
351{
352}
353
354void
355cpu_thread_swapout(struct thread *td)
356{
357}
358
359void
360cpu_thread_alloc(struct thread *td)
361{
362	struct pcb *pcb;
363	struct xstate_hdr *xhdr;
364
365	td->td_pcb = pcb = get_pcb_td(td);
366	td->td_frame = (struct trapframe *)pcb - 1;
367	pcb->pcb_save = get_pcb_user_save_pcb(pcb);
368	if (use_xsave) {
369		xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
370		bzero(xhdr, sizeof(*xhdr));
371		xhdr->xstate_bv = xsave_mask;
372	}
373}
374
375void
376cpu_thread_free(struct thread *td)
377{
378
379	cpu_thread_clean(td);
380}
381
382void
383cpu_set_syscall_retval(struct thread *td, int error)
384{
385
386	switch (error) {
387	case 0:
388		td->td_frame->tf_rax = td->td_retval[0];
389		td->td_frame->tf_rdx = td->td_retval[1];
390		td->td_frame->tf_rflags &= ~PSL_C;
391		break;
392
393	case ERESTART:
394		/*
395		 * Reconstruct pc, we know that 'syscall' is 2 bytes,
396		 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
397		 * We saved this in tf_err.
398		 * %r10 (which was holding the value of %rcx) is restored
399		 * for the next iteration.
400		 * %r10 restore is only required for freebsd/amd64 processes,
401		 * but shall be innocent for any ia32 ABI.
402		 *
403		 * Require full context restore to get the arguments
404		 * in the registers reloaded at return to usermode.
405		 */
406		td->td_frame->tf_rip -= td->td_frame->tf_err;
407		td->td_frame->tf_r10 = td->td_frame->tf_rcx;
408		set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
409		break;
410
411	case EJUSTRETURN:
412		break;
413
414	default:
415		td->td_frame->tf_rax = SV_ABI_ERRNO(td->td_proc, error);
416		td->td_frame->tf_rflags |= PSL_C;
417		break;
418	}
419}
420
421/*
422 * Initialize machine state, mostly pcb and trap frame for a new
423 * thread, about to return to userspace.  Put enough state in the new
424 * thread's PCB to get it to go back to the fork_return(), which
425 * finalizes the thread state and handles peculiarities of the first
426 * return to userspace for the new thread.
427 */
428void
429cpu_copy_thread(struct thread *td, struct thread *td0)
430{
431	struct pcb *pcb2;
432
433	/* Point the pcb to the top of the stack. */
434	pcb2 = td->td_pcb;
435
436	/*
437	 * Copy the upcall pcb.  This loads kernel regs.
438	 * Those not loaded individually below get their default
439	 * values here.
440	 */
441	update_pcb_bases(td0->td_pcb);
442	bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
443	clear_pcb_flags(pcb2, PCB_FPUINITDONE | PCB_USERFPUINITDONE |
444	    PCB_KERNFPU);
445	pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
446	bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save,
447	    cpu_max_ext_state_size);
448	set_pcb_flags_raw(pcb2, PCB_FULL_IRET);
449
450	/*
451	 * Create a new fresh stack for the new thread.
452	 */
453	bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
454
455	/* If the current thread has the trap bit set (i.e. a debugger had
456	 * single stepped the process to the system call), we need to clear
457	 * the trap flag from the new frame. Otherwise, the new thread will
458	 * receive a (likely unexpected) SIGTRAP when it executes the first
459	 * instruction after returning to userland.
460	 */
461	td->td_frame->tf_rflags &= ~PSL_T;
462
463	/*
464	 * Set registers for trampoline to user mode.  Leave space for the
465	 * return address on stack.  These are the kernel mode register values.
466	 */
467	pcb2->pcb_r12 = (register_t)fork_return;	    /* trampoline arg */
468	pcb2->pcb_rbp = 0;
469	pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *);	/* trampoline arg */
470	pcb2->pcb_rbx = (register_t)td;			    /* trampoline arg */
471	pcb2->pcb_rip = (register_t)fork_trampoline;
472	/*
473	 * If we didn't copy the pcb, we'd need to do the following registers:
474	 * pcb2->pcb_dr*:	cloned above.
475	 * pcb2->pcb_savefpu:	cloned above.
476	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
477	 * pcb2->pcb_[fg]sbase: cloned above
478	 */
479
480	/* Setup to release spin count in fork_exit(). */
481	td->td_md.md_spinlock_count = 1;
482	td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
483}
484
485/*
486 * Set that machine state for performing an upcall that starts
487 * the entry function with the given argument.
488 */
489void
490cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg,
491    stack_t *stack)
492{
493
494	/*
495	 * Do any extra cleaning that needs to be done.
496	 * The thread may have optional components
497	 * that are not present in a fresh thread.
498	 * This may be a recycled thread so make it look
499	 * as though it's newly allocated.
500	 */
501	cpu_thread_clean(td);
502
503#ifdef COMPAT_FREEBSD32
504	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
505		/*
506		 * Set the trap frame to point at the beginning of the entry
507		 * function.
508		 */
509		td->td_frame->tf_rbp = 0;
510		td->td_frame->tf_rsp =
511		   (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
512		td->td_frame->tf_rip = (uintptr_t)entry;
513
514		/* Return address sentinel value to stop stack unwinding. */
515		suword32((void *)td->td_frame->tf_rsp, 0);
516
517		/* Pass the argument to the entry point. */
518		suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)),
519		    (uint32_t)(uintptr_t)arg);
520
521		return;
522	}
523#endif
524
525	/*
526	 * Set the trap frame to point at the beginning of the uts
527	 * function.
528	 */
529	td->td_frame->tf_rbp = 0;
530	td->td_frame->tf_rsp =
531	    ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f;
532	td->td_frame->tf_rsp -= 8;
533	td->td_frame->tf_rip = (register_t)entry;
534	td->td_frame->tf_ds = _udatasel;
535	td->td_frame->tf_es = _udatasel;
536	td->td_frame->tf_fs = _ufssel;
537	td->td_frame->tf_gs = _ugssel;
538	td->td_frame->tf_flags = TF_HASSEGS;
539
540	/* Return address sentinel value to stop stack unwinding. */
541	suword((void *)td->td_frame->tf_rsp, 0);
542
543	/* Pass the argument to the entry point. */
544	td->td_frame->tf_rdi = (register_t)arg;
545}
546
547int
548cpu_set_user_tls(struct thread *td, void *tls_base)
549{
550	struct pcb *pcb;
551
552	if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS)
553		return (EINVAL);
554
555	pcb = td->td_pcb;
556	set_pcb_flags(pcb, PCB_FULL_IRET);
557#ifdef COMPAT_FREEBSD32
558	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
559		pcb->pcb_gsbase = (register_t)tls_base;
560		return (0);
561	}
562#endif
563	pcb->pcb_fsbase = (register_t)tls_base;
564	return (0);
565}
566
567#ifdef SMP
568static void
569cpu_reset_proxy()
570{
571	cpuset_t tcrp;
572
573	cpu_reset_proxy_active = 1;
574	while (cpu_reset_proxy_active == 1)
575		ia32_pause(); /* Wait for other cpu to see that we've started */
576
577	CPU_SETOF(cpu_reset_proxyid, &tcrp);
578	stop_cpus(tcrp);
579	printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
580	DELAY(1000000);
581	cpu_reset_real();
582}
583#endif
584
585void
586cpu_reset()
587{
588#ifdef SMP
589	cpuset_t map;
590	u_int cnt;
591
592	if (smp_started) {
593		map = all_cpus;
594		CPU_CLR(PCPU_GET(cpuid), &map);
595		CPU_NAND(&map, &stopped_cpus);
596		if (!CPU_EMPTY(&map)) {
597			printf("cpu_reset: Stopping other CPUs\n");
598			stop_cpus(map);
599		}
600
601		if (PCPU_GET(cpuid) != 0) {
602			cpu_reset_proxyid = PCPU_GET(cpuid);
603			cpustop_restartfunc = cpu_reset_proxy;
604			cpu_reset_proxy_active = 0;
605			printf("cpu_reset: Restarting BSP\n");
606
607			/* Restart CPU #0. */
608			CPU_SETOF(0, &started_cpus);
609			wmb();
610
611			cnt = 0;
612			while (cpu_reset_proxy_active == 0 && cnt < 10000000) {
613				ia32_pause();
614				cnt++;	/* Wait for BSP to announce restart */
615			}
616			if (cpu_reset_proxy_active == 0)
617				printf("cpu_reset: Failed to restart BSP\n");
618			enable_intr();
619			cpu_reset_proxy_active = 2;
620
621			while (1)
622				ia32_pause();
623			/* NOTREACHED */
624		}
625
626		DELAY(1000000);
627	}
628#endif
629	cpu_reset_real();
630	/* NOTREACHED */
631}
632
633static void
634cpu_reset_real()
635{
636	struct region_descriptor null_idt;
637	int b;
638
639	disable_intr();
640
641	/*
642	 * Attempt to do a CPU reset via the keyboard controller,
643	 * do not turn off GateA20, as any machine that fails
644	 * to do the reset here would then end up in no man's land.
645	 */
646	outb(IO_KBD + 4, 0xFE);
647	DELAY(500000);	/* wait 0.5 sec to see if that did it */
648
649	/*
650	 * Attempt to force a reset via the Reset Control register at
651	 * I/O port 0xcf9.  Bit 2 forces a system reset when it
652	 * transitions from 0 to 1.  Bit 1 selects the type of reset
653	 * to attempt: 0 selects a "soft" reset, and 1 selects a
654	 * "hard" reset.  We try a "hard" reset.  The first write sets
655	 * bit 1 to select a "hard" reset and clears bit 2.  The
656	 * second write forces a 0 -> 1 transition in bit 2 to trigger
657	 * a reset.
658	 */
659	outb(0xcf9, 0x2);
660	outb(0xcf9, 0x6);
661	DELAY(500000);  /* wait 0.5 sec to see if that did it */
662
663	/*
664	 * Attempt to force a reset via the Fast A20 and Init register
665	 * at I/O port 0x92.  Bit 1 serves as an alternate A20 gate.
666	 * Bit 0 asserts INIT# when set to 1.  We are careful to only
667	 * preserve bit 1 while setting bit 0.  We also must clear bit
668	 * 0 before setting it if it isn't already clear.
669	 */
670	b = inb(0x92);
671	if (b != 0xff) {
672		if ((b & 0x1) != 0)
673			outb(0x92, b & 0xfe);
674		outb(0x92, b | 0x1);
675		DELAY(500000);  /* wait 0.5 sec to see if that did it */
676	}
677
678	printf("No known reset method worked, attempting CPU shutdown\n");
679	DELAY(1000000);	/* wait 1 sec for printf to complete */
680
681	/* Wipe the IDT. */
682	null_idt.rd_limit = 0;
683	null_idt.rd_base = 0;
684	lidt(&null_idt);
685
686	/* "good night, sweet prince .... <THUNK!>" */
687	breakpoint();
688
689	/* NOTREACHED */
690	while(1);
691}
692
693/*
694 * Software interrupt handler for queued VM system processing.
695 */
696void
697swi_vm(void *dummy)
698{
699	if (busdma_swi_pending != 0)
700		busdma_swi();
701}
702
703/*
704 * Tell whether this address is in some physical memory region.
705 * Currently used by the kernel coredump code in order to avoid
706 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
707 * or other unpredictable behaviour.
708 */
709
710int
711is_physical_memory(vm_paddr_t addr)
712{
713
714#ifdef DEV_ISA
715	/* The ISA ``memory hole''. */
716	if (addr >= 0xa0000 && addr < 0x100000)
717		return 0;
718#endif
719
720	/*
721	 * stuff other tests for known memory-mapped devices (PCI?)
722	 * here
723	 */
724
725	return 1;
726}
727