xref: /freebsd/sys/i386/i386/vm_machdep.c (revision f4b37ed0f8b307b1f3f0f630ca725d68f1dff30d)
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_npx.h"
48 #include "opt_reset.h"
49 #include "opt_cpu.h"
50 #include "opt_xbox.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/sysent.h>
65 #include <sys/sf_buf.h>
66 #include <sys/smp.h>
67 #include <sys/sched.h>
68 #include <sys/sysctl.h>
69 #include <sys/unistd.h>
70 #include <sys/vnode.h>
71 #include <sys/vmmeter.h>
72 
73 #include <machine/cpu.h>
74 #include <machine/cputypes.h>
75 #include <machine/md_var.h>
76 #include <machine/pcb.h>
77 #include <machine/pcb_ext.h>
78 #include <machine/smp.h>
79 #include <machine/vm86.h>
80 
81 #ifdef CPU_ELAN
82 #include <machine/elan_mmcr.h>
83 #endif
84 
85 #include <vm/vm.h>
86 #include <vm/vm_extern.h>
87 #include <vm/vm_kern.h>
88 #include <vm/vm_page.h>
89 #include <vm/vm_map.h>
90 #include <vm/vm_param.h>
91 
92 #ifdef PC98
93 #include <pc98/cbus/cbus.h>
94 #else
95 #include <isa/isareg.h>
96 #endif
97 
98 #ifdef XBOX
99 #include <machine/xbox.h>
100 #endif
101 
102 #ifndef NSFBUFS
103 #define	NSFBUFS		(512 + maxusers * 16)
104 #endif
105 
106 #if !defined(CPU_DISABLE_SSE) && defined(I686_CPU)
107 #define CPU_ENABLE_SSE
108 #endif
109 
110 _Static_assert(OFFSETOF_CURTHREAD == offsetof(struct pcpu, pc_curthread),
111     "OFFSETOF_CURTHREAD does not correspond with offset of pc_curthread.");
112 _Static_assert(OFFSETOF_CURPCB == offsetof(struct pcpu, pc_curpcb),
113     "OFFSETOF_CURPCB does not correspond with offset of pc_curpcb.");
114 _Static_assert(OFFSETOF_MONITORBUF == offsetof(struct pcpu, pc_monitorbuf),
115     "OFFSETOF_MONINORBUF does not correspond with offset of pc_monitorbuf.");
116 
117 static void	cpu_reset_real(void);
118 #ifdef SMP
119 static void	cpu_reset_proxy(void);
120 static u_int	cpu_reset_proxyid;
121 static volatile u_int	cpu_reset_proxy_active;
122 #endif
123 
124 union savefpu *
125 get_pcb_user_save_td(struct thread *td)
126 {
127 	vm_offset_t p;
128 
129 	p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
130 	    cpu_max_ext_state_size;
131 	KASSERT((p % 64) == 0, ("Unaligned pcb_user_save area"));
132 	return ((union savefpu *)p);
133 }
134 
135 union savefpu *
136 get_pcb_user_save_pcb(struct pcb *pcb)
137 {
138 	vm_offset_t p;
139 
140 	p = (vm_offset_t)(pcb + 1);
141 	return ((union savefpu *)p);
142 }
143 
144 struct pcb *
145 get_pcb_td(struct thread *td)
146 {
147 	vm_offset_t p;
148 
149 	p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
150 	    cpu_max_ext_state_size - sizeof(struct pcb);
151 	return ((struct pcb *)p);
152 }
153 
154 void *
155 alloc_fpusave(int flags)
156 {
157 	void *res;
158 #ifdef CPU_ENABLE_SSE
159 	struct savefpu_ymm *sf;
160 #endif
161 
162 	res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags);
163 #ifdef CPU_ENABLE_SSE
164 	if (use_xsave) {
165 		sf = (struct savefpu_ymm *)res;
166 		bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd));
167 		sf->sv_xstate.sx_hd.xstate_bv = xsave_mask;
168 	}
169 #endif
170 	return (res);
171 }
172 /*
173  * Finish a fork operation, with process p2 nearly set up.
174  * Copy and update the pcb, set up the stack so that the child
175  * ready to run and return to user mode.
176  */
177 void
178 cpu_fork(td1, p2, td2, flags)
179 	register struct thread *td1;
180 	register struct proc *p2;
181 	struct thread *td2;
182 	int flags;
183 {
184 	register struct proc *p1;
185 	struct pcb *pcb2;
186 	struct mdproc *mdp2;
187 
188 	p1 = td1->td_proc;
189 	if ((flags & RFPROC) == 0) {
190 		if ((flags & RFMEM) == 0) {
191 			/* unshare user LDT */
192 			struct mdproc *mdp1 = &p1->p_md;
193 			struct proc_ldt *pldt, *pldt1;
194 
195 			mtx_lock_spin(&dt_lock);
196 			if ((pldt1 = mdp1->md_ldt) != NULL &&
197 			    pldt1->ldt_refcnt > 1) {
198 				pldt = user_ldt_alloc(mdp1, pldt1->ldt_len);
199 				if (pldt == NULL)
200 					panic("could not copy LDT");
201 				mdp1->md_ldt = pldt;
202 				set_user_ldt(mdp1);
203 				user_ldt_deref(pldt1);
204 			} else
205 				mtx_unlock_spin(&dt_lock);
206 		}
207 		return;
208 	}
209 
210 	/* Ensure that td1's pcb is up to date. */
211 	if (td1 == curthread)
212 		td1->td_pcb->pcb_gs = rgs();
213 #ifdef DEV_NPX
214 	critical_enter();
215 	if (PCPU_GET(fpcurthread) == td1)
216 		npxsave(td1->td_pcb->pcb_save);
217 	critical_exit();
218 #endif
219 
220 	/* Point the pcb to the top of the stack */
221 	pcb2 = get_pcb_td(td2);
222 	td2->td_pcb = pcb2;
223 
224 	/* Copy td1's pcb */
225 	bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
226 
227 	/* Properly initialize pcb_save */
228 	pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
229 	bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2),
230 	    cpu_max_ext_state_size);
231 
232 	/* Point mdproc and then copy over td1's contents */
233 	mdp2 = &p2->p_md;
234 	bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
235 
236 	/*
237 	 * Create a new fresh stack for the new process.
238 	 * Copy the trap frame for the return to user mode as if from a
239 	 * syscall.  This copies most of the user mode register values.
240 	 * The -16 is so we can expand the trapframe if we go to vm86.
241 	 */
242 	td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1;
243 	bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
244 
245 	td2->td_frame->tf_eax = 0;		/* Child returns zero */
246 	td2->td_frame->tf_eflags &= ~PSL_C;	/* success */
247 	td2->td_frame->tf_edx = 1;
248 
249 	/*
250 	 * If the parent process has the trap bit set (i.e. a debugger had
251 	 * single stepped the process to the system call), we need to clear
252 	 * the trap flag from the new frame unless the debugger had set PF_FORK
253 	 * on the parent.  Otherwise, the child will receive a (likely
254 	 * unexpected) SIGTRAP when it executes the first instruction after
255 	 * returning  to userland.
256 	 */
257 	if ((p1->p_pfsflags & PF_FORK) == 0)
258 		td2->td_frame->tf_eflags &= ~PSL_T;
259 
260 	/*
261 	 * Set registers for trampoline to user mode.  Leave space for the
262 	 * return address on stack.  These are the kernel mode register values.
263 	 */
264 #if defined(PAE) || defined(PAE_TABLES)
265 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdpt);
266 #else
267 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
268 #endif
269 	pcb2->pcb_edi = 0;
270 	pcb2->pcb_esi = (int)fork_return;	/* fork_trampoline argument */
271 	pcb2->pcb_ebp = 0;
272 	pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
273 	pcb2->pcb_ebx = (int)td2;		/* fork_trampoline argument */
274 	pcb2->pcb_eip = (int)fork_trampoline;
275 	pcb2->pcb_psl = PSL_KERNEL;		/* ints disabled */
276 	/*-
277 	 * pcb2->pcb_dr*:	cloned above.
278 	 * pcb2->pcb_savefpu:	cloned above.
279 	 * pcb2->pcb_flags:	cloned above.
280 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
281 	 * pcb2->pcb_gs:	cloned above.
282 	 * pcb2->pcb_ext:	cleared below.
283 	 */
284 
285 	/*
286 	 * XXX don't copy the i/o pages.  this should probably be fixed.
287 	 */
288 	pcb2->pcb_ext = 0;
289 
290 	/* Copy the LDT, if necessary. */
291 	mtx_lock_spin(&dt_lock);
292 	if (mdp2->md_ldt != NULL) {
293 		if (flags & RFMEM) {
294 			mdp2->md_ldt->ldt_refcnt++;
295 		} else {
296 			mdp2->md_ldt = user_ldt_alloc(mdp2,
297 			    mdp2->md_ldt->ldt_len);
298 			if (mdp2->md_ldt == NULL)
299 				panic("could not copy LDT");
300 		}
301 	}
302 	mtx_unlock_spin(&dt_lock);
303 
304 	/* Setup to release spin count in fork_exit(). */
305 	td2->td_md.md_spinlock_count = 1;
306 	td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
307 
308 	/*
309 	 * Now, cpu_switch() can schedule the new process.
310 	 * pcb_esp is loaded pointing to the cpu_switch() stack frame
311 	 * containing the return address when exiting cpu_switch.
312 	 * This will normally be to fork_trampoline(), which will have
313 	 * %ebx loaded with the new proc's pointer.  fork_trampoline()
314 	 * will set up a stack to call fork_return(p, frame); to complete
315 	 * the return to user-mode.
316 	 */
317 }
318 
319 /*
320  * Intercept the return address from a freshly forked process that has NOT
321  * been scheduled yet.
322  *
323  * This is needed to make kernel threads stay in kernel mode.
324  */
325 void
326 cpu_set_fork_handler(td, func, arg)
327 	struct thread *td;
328 	void (*func)(void *);
329 	void *arg;
330 {
331 	/*
332 	 * Note that the trap frame follows the args, so the function
333 	 * is really called like this:  func(arg, frame);
334 	 */
335 	td->td_pcb->pcb_esi = (int) func;	/* function */
336 	td->td_pcb->pcb_ebx = (int) arg;	/* first arg */
337 }
338 
339 void
340 cpu_exit(struct thread *td)
341 {
342 
343 	/*
344 	 * If this process has a custom LDT, release it.  Reset pc->pcb_gs
345 	 * and %gs before we free it in case they refer to an LDT entry.
346 	 */
347 	mtx_lock_spin(&dt_lock);
348 	if (td->td_proc->p_md.md_ldt) {
349 		td->td_pcb->pcb_gs = _udatasel;
350 		load_gs(_udatasel);
351 		user_ldt_free(td);
352 	} else
353 		mtx_unlock_spin(&dt_lock);
354 }
355 
356 void
357 cpu_thread_exit(struct thread *td)
358 {
359 
360 #ifdef DEV_NPX
361 	critical_enter();
362 	if (td == PCPU_GET(fpcurthread))
363 		npxdrop();
364 	critical_exit();
365 #endif
366 
367 	/* Disable any hardware breakpoints. */
368 	if (td->td_pcb->pcb_flags & PCB_DBREGS) {
369 		reset_dbregs();
370 		td->td_pcb->pcb_flags &= ~PCB_DBREGS;
371 	}
372 }
373 
374 void
375 cpu_thread_clean(struct thread *td)
376 {
377 	struct pcb *pcb;
378 
379 	pcb = td->td_pcb;
380 	if (pcb->pcb_ext != NULL) {
381 		/* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */
382 		/*
383 		 * XXX do we need to move the TSS off the allocated pages
384 		 * before freeing them?  (not done here)
385 		 */
386 		kmem_free(kernel_arena, (vm_offset_t)pcb->pcb_ext,
387 		    ctob(IOPAGES + 1));
388 		pcb->pcb_ext = NULL;
389 	}
390 }
391 
392 void
393 cpu_thread_swapin(struct thread *td)
394 {
395 }
396 
397 void
398 cpu_thread_swapout(struct thread *td)
399 {
400 }
401 
402 void
403 cpu_thread_alloc(struct thread *td)
404 {
405 	struct pcb *pcb;
406 #ifdef CPU_ENABLE_SSE
407 	struct xstate_hdr *xhdr;
408 #endif
409 
410 	td->td_pcb = pcb = get_pcb_td(td);
411 	td->td_frame = (struct trapframe *)((caddr_t)pcb - 16) - 1;
412 	pcb->pcb_ext = NULL;
413 	pcb->pcb_save = get_pcb_user_save_pcb(pcb);
414 #ifdef CPU_ENABLE_SSE
415 	if (use_xsave) {
416 		xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
417 		bzero(xhdr, sizeof(*xhdr));
418 		xhdr->xstate_bv = xsave_mask;
419 	}
420 #endif
421 }
422 
423 void
424 cpu_thread_free(struct thread *td)
425 {
426 
427 	cpu_thread_clean(td);
428 }
429 
430 void
431 cpu_set_syscall_retval(struct thread *td, int error)
432 {
433 
434 	switch (error) {
435 	case 0:
436 		td->td_frame->tf_eax = td->td_retval[0];
437 		td->td_frame->tf_edx = td->td_retval[1];
438 		td->td_frame->tf_eflags &= ~PSL_C;
439 		break;
440 
441 	case ERESTART:
442 		/*
443 		 * Reconstruct pc, assuming lcall $X,y is 7 bytes, int
444 		 * 0x80 is 2 bytes. We saved this in tf_err.
445 		 */
446 		td->td_frame->tf_eip -= td->td_frame->tf_err;
447 		break;
448 
449 	case EJUSTRETURN:
450 		break;
451 
452 	default:
453 		if (td->td_proc->p_sysent->sv_errsize) {
454 			if (error >= td->td_proc->p_sysent->sv_errsize)
455 				error = -1;	/* XXX */
456 			else
457 				error = td->td_proc->p_sysent->sv_errtbl[error];
458 		}
459 		td->td_frame->tf_eax = error;
460 		td->td_frame->tf_eflags |= PSL_C;
461 		break;
462 	}
463 }
464 
465 /*
466  * Initialize machine state (pcb and trap frame) for a new thread about to
467  * upcall. Put enough state in the new thread's PCB to get it to go back
468  * userret(), where we can intercept it again to set the return (upcall)
469  * Address and stack, along with those from upcals that are from other sources
470  * such as those generated in thread_userret() itself.
471  */
472 void
473 cpu_set_upcall(struct thread *td, struct thread *td0)
474 {
475 	struct pcb *pcb2;
476 
477 	/* Point the pcb to the top of the stack. */
478 	pcb2 = td->td_pcb;
479 
480 	/*
481 	 * Copy the upcall pcb.  This loads kernel regs.
482 	 * Those not loaded individually below get their default
483 	 * values here.
484 	 */
485 	bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
486 	pcb2->pcb_flags &= ~(PCB_NPXINITDONE | PCB_NPXUSERINITDONE |
487 	    PCB_KERNNPX);
488 	pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
489 	bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save,
490 	    cpu_max_ext_state_size);
491 
492 	/*
493 	 * Create a new fresh stack for the new thread.
494 	 */
495 	bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
496 
497 	/* If the current thread has the trap bit set (i.e. a debugger had
498 	 * single stepped the process to the system call), we need to clear
499 	 * the trap flag from the new frame. Otherwise, the new thread will
500 	 * receive a (likely unexpected) SIGTRAP when it executes the first
501 	 * instruction after returning to userland.
502 	 */
503 	td->td_frame->tf_eflags &= ~PSL_T;
504 
505 	/*
506 	 * Set registers for trampoline to user mode.  Leave space for the
507 	 * return address on stack.  These are the kernel mode register values.
508 	 */
509 	pcb2->pcb_edi = 0;
510 	pcb2->pcb_esi = (int)fork_return;		    /* trampoline arg */
511 	pcb2->pcb_ebp = 0;
512 	pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */
513 	pcb2->pcb_ebx = (int)td;			    /* trampoline arg */
514 	pcb2->pcb_eip = (int)fork_trampoline;
515 	pcb2->pcb_psl &= ~(PSL_I);	/* interrupts must be disabled */
516 	pcb2->pcb_gs = rgs();
517 	/*
518 	 * If we didn't copy the pcb, we'd need to do the following registers:
519 	 * pcb2->pcb_cr3:	cloned above.
520 	 * pcb2->pcb_dr*:	cloned above.
521 	 * pcb2->pcb_savefpu:	cloned above.
522 	 * pcb2->pcb_flags:	cloned above.
523 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
524 	 * pcb2->pcb_gs:	cloned above.
525 	 * pcb2->pcb_ext:	cleared below.
526 	 */
527 	pcb2->pcb_ext = NULL;
528 
529 	/* Setup to release spin count in fork_exit(). */
530 	td->td_md.md_spinlock_count = 1;
531 	td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
532 }
533 
534 /*
535  * Set that machine state for performing an upcall that has to
536  * be done in thread_userret() so that those upcalls generated
537  * in thread_userret() itself can be done as well.
538  */
539 void
540 cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg,
541 	stack_t *stack)
542 {
543 
544 	/*
545 	 * Do any extra cleaning that needs to be done.
546 	 * The thread may have optional components
547 	 * that are not present in a fresh thread.
548 	 * This may be a recycled thread so make it look
549 	 * as though it's newly allocated.
550 	 */
551 	cpu_thread_clean(td);
552 
553 	/*
554 	 * Set the trap frame to point at the beginning of the uts
555 	 * function.
556 	 */
557 	td->td_frame->tf_ebp = 0;
558 	td->td_frame->tf_esp =
559 	    (((int)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
560 	td->td_frame->tf_eip = (int)entry;
561 
562 	/*
563 	 * Pass the address of the mailbox for this kse to the uts
564 	 * function as a parameter on the stack.
565 	 */
566 	suword((void *)(td->td_frame->tf_esp + sizeof(void *)),
567 	    (int)arg);
568 }
569 
570 int
571 cpu_set_user_tls(struct thread *td, void *tls_base)
572 {
573 	struct segment_descriptor sd;
574 	uint32_t base;
575 
576 	/*
577 	 * Construct a descriptor and store it in the pcb for
578 	 * the next context switch.  Also store it in the gdt
579 	 * so that the load of tf_fs into %fs will activate it
580 	 * at return to userland.
581 	 */
582 	base = (uint32_t)tls_base;
583 	sd.sd_lobase = base & 0xffffff;
584 	sd.sd_hibase = (base >> 24) & 0xff;
585 	sd.sd_lolimit = 0xffff;	/* 4GB limit, wraps around */
586 	sd.sd_hilimit = 0xf;
587 	sd.sd_type  = SDT_MEMRWA;
588 	sd.sd_dpl   = SEL_UPL;
589 	sd.sd_p     = 1;
590 	sd.sd_xx    = 0;
591 	sd.sd_def32 = 1;
592 	sd.sd_gran  = 1;
593 	critical_enter();
594 	/* set %gs */
595 	td->td_pcb->pcb_gsd = sd;
596 	if (td == curthread) {
597 		PCPU_GET(fsgs_gdt)[1] = sd;
598 		load_gs(GSEL(GUGS_SEL, SEL_UPL));
599 	}
600 	critical_exit();
601 	return (0);
602 }
603 
604 /*
605  * Convert kernel VA to physical address
606  */
607 vm_paddr_t
608 kvtop(void *addr)
609 {
610 	vm_paddr_t pa;
611 
612 	pa = pmap_kextract((vm_offset_t)addr);
613 	if (pa == 0)
614 		panic("kvtop: zero page frame");
615 	return (pa);
616 }
617 
618 #ifdef SMP
619 static void
620 cpu_reset_proxy()
621 {
622 	cpuset_t tcrp;
623 
624 	cpu_reset_proxy_active = 1;
625 	while (cpu_reset_proxy_active == 1)
626 		;	/* Wait for other cpu to see that we've started */
627 	CPU_SETOF(cpu_reset_proxyid, &tcrp);
628 	stop_cpus(tcrp);
629 	printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
630 	DELAY(1000000);
631 	cpu_reset_real();
632 }
633 #endif
634 
635 void
636 cpu_reset()
637 {
638 #ifdef XBOX
639 	if (arch_i386_is_xbox) {
640 		/* Kick the PIC16L, it can reboot the box */
641 		pic16l_reboot();
642 		for (;;);
643 	}
644 #endif
645 
646 #ifdef SMP
647 	cpuset_t map;
648 	u_int cnt;
649 
650 	if (smp_started) {
651 		map = all_cpus;
652 		CPU_CLR(PCPU_GET(cpuid), &map);
653 		CPU_NAND(&map, &stopped_cpus);
654 		if (!CPU_EMPTY(&map)) {
655 			printf("cpu_reset: Stopping other CPUs\n");
656 			stop_cpus(map);
657 		}
658 
659 		if (PCPU_GET(cpuid) != 0) {
660 			cpu_reset_proxyid = PCPU_GET(cpuid);
661 			cpustop_restartfunc = cpu_reset_proxy;
662 			cpu_reset_proxy_active = 0;
663 			printf("cpu_reset: Restarting BSP\n");
664 
665 			/* Restart CPU #0. */
666 			/* XXX: restart_cpus(1 << 0); */
667 			CPU_SETOF(0, &started_cpus);
668 			wmb();
669 
670 			cnt = 0;
671 			while (cpu_reset_proxy_active == 0 && cnt < 10000000)
672 				cnt++;	/* Wait for BSP to announce restart */
673 			if (cpu_reset_proxy_active == 0)
674 				printf("cpu_reset: Failed to restart BSP\n");
675 			enable_intr();
676 			cpu_reset_proxy_active = 2;
677 
678 			while (1);
679 			/* NOTREACHED */
680 		}
681 
682 		DELAY(1000000);
683 	}
684 #endif
685 	cpu_reset_real();
686 	/* NOTREACHED */
687 }
688 
689 static void
690 cpu_reset_real()
691 {
692 	struct region_descriptor null_idt;
693 #ifndef PC98
694 	int b;
695 #endif
696 
697 	disable_intr();
698 #ifdef CPU_ELAN
699 	if (elan_mmcr != NULL)
700 		elan_mmcr->RESCFG = 1;
701 #endif
702 
703 	if (cpu == CPU_GEODE1100) {
704 		/* Attempt Geode's own reset */
705 		outl(0xcf8, 0x80009044ul);
706 		outl(0xcfc, 0xf);
707 	}
708 
709 #ifdef PC98
710 	/*
711 	 * Attempt to do a CPU reset via CPU reset port.
712 	 */
713 	if ((inb(0x35) & 0xa0) != 0xa0) {
714 		outb(0x37, 0x0f);		/* SHUT0 = 0. */
715 		outb(0x37, 0x0b);		/* SHUT1 = 0. */
716 	}
717 	outb(0xf0, 0x00);		/* Reset. */
718 #else
719 #if !defined(BROKEN_KEYBOARD_RESET)
720 	/*
721 	 * Attempt to do a CPU reset via the keyboard controller,
722 	 * do not turn off GateA20, as any machine that fails
723 	 * to do the reset here would then end up in no man's land.
724 	 */
725 	outb(IO_KBD + 4, 0xFE);
726 	DELAY(500000);	/* wait 0.5 sec to see if that did it */
727 #endif
728 
729 	/*
730 	 * Attempt to force a reset via the Reset Control register at
731 	 * I/O port 0xcf9.  Bit 2 forces a system reset when it
732 	 * transitions from 0 to 1.  Bit 1 selects the type of reset
733 	 * to attempt: 0 selects a "soft" reset, and 1 selects a
734 	 * "hard" reset.  We try a "hard" reset.  The first write sets
735 	 * bit 1 to select a "hard" reset and clears bit 2.  The
736 	 * second write forces a 0 -> 1 transition in bit 2 to trigger
737 	 * a reset.
738 	 */
739 	outb(0xcf9, 0x2);
740 	outb(0xcf9, 0x6);
741 	DELAY(500000);  /* wait 0.5 sec to see if that did it */
742 
743 	/*
744 	 * Attempt to force a reset via the Fast A20 and Init register
745 	 * at I/O port 0x92.  Bit 1 serves as an alternate A20 gate.
746 	 * Bit 0 asserts INIT# when set to 1.  We are careful to only
747 	 * preserve bit 1 while setting bit 0.  We also must clear bit
748 	 * 0 before setting it if it isn't already clear.
749 	 */
750 	b = inb(0x92);
751 	if (b != 0xff) {
752 		if ((b & 0x1) != 0)
753 			outb(0x92, b & 0xfe);
754 		outb(0x92, b | 0x1);
755 		DELAY(500000);  /* wait 0.5 sec to see if that did it */
756 	}
757 #endif /* PC98 */
758 
759 	printf("No known reset method worked, attempting CPU shutdown\n");
760 	DELAY(1000000); /* wait 1 sec for printf to complete */
761 
762 	/* Wipe the IDT. */
763 	null_idt.rd_limit = 0;
764 	null_idt.rd_base = 0;
765 	lidt(&null_idt);
766 
767 	/* "good night, sweet prince .... <THUNK!>" */
768 	breakpoint();
769 
770 	/* NOTREACHED */
771 	while(1);
772 }
773 
774 /*
775  * Get an sf_buf from the freelist.  May block if none are available.
776  */
777 void
778 sf_buf_map(struct sf_buf *sf, int flags)
779 {
780 	pt_entry_t opte, *ptep;
781 
782 	/*
783 	 * Update the sf_buf's virtual-to-physical mapping, flushing the
784 	 * virtual address from the TLB.  Since the reference count for
785 	 * the sf_buf's old mapping was zero, that mapping is not
786 	 * currently in use.  Consequently, there is no need to exchange
787 	 * the old and new PTEs atomically, even under PAE.
788 	 */
789 	ptep = vtopte(sf->kva);
790 	opte = *ptep;
791 	*ptep = VM_PAGE_TO_PHYS(sf->m) | pgeflag | PG_RW | PG_V |
792 	    pmap_cache_bits(sf->m->md.pat_mode, 0);
793 
794 	/*
795 	 * Avoid unnecessary TLB invalidations: If the sf_buf's old
796 	 * virtual-to-physical mapping was not used, then any processor
797 	 * that has invalidated the sf_buf's virtual address from its TLB
798 	 * since the last used mapping need not invalidate again.
799 	 */
800 #ifdef SMP
801 	if ((opte & (PG_V | PG_A)) ==  (PG_V | PG_A))
802 		CPU_ZERO(&sf->cpumask);
803 
804 	sf_buf_shootdown(sf, flags);
805 #else
806 	if ((opte & (PG_V | PG_A)) ==  (PG_V | PG_A))
807 		pmap_invalidate_page(kernel_pmap, sf->kva);
808 #endif
809 }
810 
811 #ifdef SMP
812 void
813 sf_buf_shootdown(struct sf_buf *sf, int flags)
814 {
815 	cpuset_t other_cpus;
816 	u_int cpuid;
817 
818 	sched_pin();
819 	cpuid = PCPU_GET(cpuid);
820 	if (!CPU_ISSET(cpuid, &sf->cpumask)) {
821 		CPU_SET(cpuid, &sf->cpumask);
822 		invlpg(sf->kva);
823 	}
824 	if ((flags & SFB_CPUPRIVATE) == 0) {
825 		other_cpus = all_cpus;
826 		CPU_CLR(cpuid, &other_cpus);
827 		CPU_NAND(&other_cpus, &sf->cpumask);
828 		if (!CPU_EMPTY(&other_cpus)) {
829 			CPU_OR(&sf->cpumask, &other_cpus);
830 			smp_masked_invlpg(other_cpus, sf->kva);
831 		}
832 	}
833 	sched_unpin();
834 }
835 #endif
836 
837 /*
838  * MD part of sf_buf_free().
839  */
840 int
841 sf_buf_unmap(struct sf_buf *sf)
842 {
843 
844 	return (0);
845 }
846 
847 static void
848 sf_buf_invalidate(struct sf_buf *sf)
849 {
850 	vm_page_t m = sf->m;
851 
852 	/*
853 	 * Use pmap_qenter to update the pte for
854 	 * existing mapping, in particular, the PAT
855 	 * settings are recalculated.
856 	 */
857 	pmap_qenter(sf->kva, &m, 1);
858 	pmap_invalidate_cache_range(sf->kva, sf->kva + PAGE_SIZE, FALSE);
859 }
860 
861 /*
862  * Invalidate the cache lines that may belong to the page, if
863  * (possibly old) mapping of the page by sf buffer exists.  Returns
864  * TRUE when mapping was found and cache invalidated.
865  */
866 boolean_t
867 sf_buf_invalidate_cache(vm_page_t m)
868 {
869 
870 	return (sf_buf_process_page(m, sf_buf_invalidate));
871 }
872 
873 /*
874  * Software interrupt handler for queued VM system processing.
875  */
876 void
877 swi_vm(void *dummy)
878 {
879 	if (busdma_swi_pending != 0)
880 		busdma_swi();
881 }
882 
883 /*
884  * Tell whether this address is in some physical memory region.
885  * Currently used by the kernel coredump code in order to avoid
886  * dumping the ``ISA memory hole'' which could cause indefinite hangs,
887  * or other unpredictable behaviour.
888  */
889 
890 int
891 is_physical_memory(vm_paddr_t addr)
892 {
893 
894 #ifdef DEV_ISA
895 	/* The ISA ``memory hole''. */
896 	if (addr >= 0xa0000 && addr < 0x100000)
897 		return 0;
898 #endif
899 
900 	/*
901 	 * stuff other tests for known memory-mapped devices (PCI?)
902 	 * here
903 	 */
904 
905 	return 1;
906 }
907