xref: /freebsd/sys/i386/i386/vm_machdep.c (revision be8aff81c11e51d25e7a134298b6aa0300f4b7fc)
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 	    roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN);
131 	KASSERT((p % XSAVE_AREA_ALIGN) == 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 	    roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN) -
151 	    sizeof(struct pcb);
152 	return ((struct pcb *)p);
153 }
154 
155 void *
156 alloc_fpusave(int flags)
157 {
158 	void *res;
159 #ifdef CPU_ENABLE_SSE
160 	struct savefpu_ymm *sf;
161 #endif
162 
163 	res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags);
164 #ifdef CPU_ENABLE_SSE
165 	if (use_xsave) {
166 		sf = (struct savefpu_ymm *)res;
167 		bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd));
168 		sf->sv_xstate.sx_hd.xstate_bv = xsave_mask;
169 	}
170 #endif
171 	return (res);
172 }
173 /*
174  * Finish a fork operation, with process p2 nearly set up.
175  * Copy and update the pcb, set up the stack so that the child
176  * ready to run and return to user mode.
177  */
178 void
179 cpu_fork(td1, p2, td2, flags)
180 	register struct thread *td1;
181 	register struct proc *p2;
182 	struct thread *td2;
183 	int flags;
184 {
185 	register struct proc *p1;
186 	struct pcb *pcb2;
187 	struct mdproc *mdp2;
188 
189 	p1 = td1->td_proc;
190 	if ((flags & RFPROC) == 0) {
191 		if ((flags & RFMEM) == 0) {
192 			/* unshare user LDT */
193 			struct mdproc *mdp1 = &p1->p_md;
194 			struct proc_ldt *pldt, *pldt1;
195 
196 			mtx_lock_spin(&dt_lock);
197 			if ((pldt1 = mdp1->md_ldt) != NULL &&
198 			    pldt1->ldt_refcnt > 1) {
199 				pldt = user_ldt_alloc(mdp1, pldt1->ldt_len);
200 				if (pldt == NULL)
201 					panic("could not copy LDT");
202 				mdp1->md_ldt = pldt;
203 				set_user_ldt(mdp1);
204 				user_ldt_deref(pldt1);
205 			} else
206 				mtx_unlock_spin(&dt_lock);
207 		}
208 		return;
209 	}
210 
211 	/* Ensure that td1's pcb is up to date. */
212 	if (td1 == curthread)
213 		td1->td_pcb->pcb_gs = rgs();
214 #ifdef DEV_NPX
215 	critical_enter();
216 	if (PCPU_GET(fpcurthread) == td1)
217 		npxsave(td1->td_pcb->pcb_save);
218 	critical_exit();
219 #endif
220 
221 	/* Point the pcb to the top of the stack */
222 	pcb2 = get_pcb_td(td2);
223 	td2->td_pcb = pcb2;
224 
225 	/* Copy td1's pcb */
226 	bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
227 
228 	/* Properly initialize pcb_save */
229 	pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
230 	bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2),
231 	    cpu_max_ext_state_size);
232 
233 	/* Point mdproc and then copy over td1's contents */
234 	mdp2 = &p2->p_md;
235 	bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
236 
237 	/*
238 	 * Create a new fresh stack for the new process.
239 	 * Copy the trap frame for the return to user mode as if from a
240 	 * syscall.  This copies most of the user mode register values.
241 	 * The -16 is so we can expand the trapframe if we go to vm86.
242 	 */
243 	td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1;
244 	bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
245 
246 	td2->td_frame->tf_eax = 0;		/* Child returns zero */
247 	td2->td_frame->tf_eflags &= ~PSL_C;	/* success */
248 	td2->td_frame->tf_edx = 1;
249 
250 	/*
251 	 * If the parent process has the trap bit set (i.e. a debugger had
252 	 * single stepped the process to the system call), we need to clear
253 	 * the trap flag from the new frame unless the debugger had set PF_FORK
254 	 * on the parent.  Otherwise, the child will receive a (likely
255 	 * unexpected) SIGTRAP when it executes the first instruction after
256 	 * returning  to userland.
257 	 */
258 	if ((p1->p_pfsflags & PF_FORK) == 0)
259 		td2->td_frame->tf_eflags &= ~PSL_T;
260 
261 	/*
262 	 * Set registers for trampoline to user mode.  Leave space for the
263 	 * return address on stack.  These are the kernel mode register values.
264 	 */
265 #if defined(PAE) || defined(PAE_TABLES)
266 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdpt);
267 #else
268 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
269 #endif
270 	pcb2->pcb_edi = 0;
271 	pcb2->pcb_esi = (int)fork_return;	/* fork_trampoline argument */
272 	pcb2->pcb_ebp = 0;
273 	pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
274 	pcb2->pcb_ebx = (int)td2;		/* fork_trampoline argument */
275 	pcb2->pcb_eip = (int)fork_trampoline;
276 	pcb2->pcb_psl = PSL_KERNEL;		/* ints disabled */
277 	/*-
278 	 * pcb2->pcb_dr*:	cloned above.
279 	 * pcb2->pcb_savefpu:	cloned above.
280 	 * pcb2->pcb_flags:	cloned above.
281 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
282 	 * pcb2->pcb_gs:	cloned above.
283 	 * pcb2->pcb_ext:	cleared below.
284 	 */
285 
286 	/*
287 	 * XXX don't copy the i/o pages.  this should probably be fixed.
288 	 */
289 	pcb2->pcb_ext = 0;
290 
291 	/* Copy the LDT, if necessary. */
292 	mtx_lock_spin(&dt_lock);
293 	if (mdp2->md_ldt != NULL) {
294 		if (flags & RFMEM) {
295 			mdp2->md_ldt->ldt_refcnt++;
296 		} else {
297 			mdp2->md_ldt = user_ldt_alloc(mdp2,
298 			    mdp2->md_ldt->ldt_len);
299 			if (mdp2->md_ldt == NULL)
300 				panic("could not copy LDT");
301 		}
302 	}
303 	mtx_unlock_spin(&dt_lock);
304 
305 	/* Setup to release spin count in fork_exit(). */
306 	td2->td_md.md_spinlock_count = 1;
307 	td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
308 
309 	/*
310 	 * Now, cpu_switch() can schedule the new process.
311 	 * pcb_esp is loaded pointing to the cpu_switch() stack frame
312 	 * containing the return address when exiting cpu_switch.
313 	 * This will normally be to fork_trampoline(), which will have
314 	 * %ebx loaded with the new proc's pointer.  fork_trampoline()
315 	 * will set up a stack to call fork_return(p, frame); to complete
316 	 * the return to user-mode.
317 	 */
318 }
319 
320 /*
321  * Intercept the return address from a freshly forked process that has NOT
322  * been scheduled yet.
323  *
324  * This is needed to make kernel threads stay in kernel mode.
325  */
326 void
327 cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg)
328 {
329 	/*
330 	 * Note that the trap frame follows the args, so the function
331 	 * is really called like this:  func(arg, frame);
332 	 */
333 	td->td_pcb->pcb_esi = (int) func;	/* function */
334 	td->td_pcb->pcb_ebx = (int) arg;	/* first arg */
335 }
336 
337 void
338 cpu_exit(struct thread *td)
339 {
340 
341 	/*
342 	 * If this process has a custom LDT, release it.  Reset pc->pcb_gs
343 	 * and %gs before we free it in case they refer to an LDT entry.
344 	 */
345 	mtx_lock_spin(&dt_lock);
346 	if (td->td_proc->p_md.md_ldt) {
347 		td->td_pcb->pcb_gs = _udatasel;
348 		load_gs(_udatasel);
349 		user_ldt_free(td);
350 	} else
351 		mtx_unlock_spin(&dt_lock);
352 }
353 
354 void
355 cpu_thread_exit(struct thread *td)
356 {
357 
358 #ifdef DEV_NPX
359 	critical_enter();
360 	if (td == PCPU_GET(fpcurthread))
361 		npxdrop();
362 	critical_exit();
363 #endif
364 
365 	/* Disable any hardware breakpoints. */
366 	if (td->td_pcb->pcb_flags & PCB_DBREGS) {
367 		reset_dbregs();
368 		td->td_pcb->pcb_flags &= ~PCB_DBREGS;
369 	}
370 }
371 
372 void
373 cpu_thread_clean(struct thread *td)
374 {
375 	struct pcb *pcb;
376 
377 	pcb = td->td_pcb;
378 	if (pcb->pcb_ext != NULL) {
379 		/* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */
380 		/*
381 		 * XXX do we need to move the TSS off the allocated pages
382 		 * before freeing them?  (not done here)
383 		 */
384 		kmem_free(kernel_arena, (vm_offset_t)pcb->pcb_ext,
385 		    ctob(IOPAGES + 1));
386 		pcb->pcb_ext = NULL;
387 	}
388 }
389 
390 void
391 cpu_thread_swapin(struct thread *td)
392 {
393 }
394 
395 void
396 cpu_thread_swapout(struct thread *td)
397 {
398 }
399 
400 void
401 cpu_thread_alloc(struct thread *td)
402 {
403 	struct pcb *pcb;
404 #ifdef CPU_ENABLE_SSE
405 	struct xstate_hdr *xhdr;
406 #endif
407 
408 	td->td_pcb = pcb = get_pcb_td(td);
409 	td->td_frame = (struct trapframe *)((caddr_t)pcb - 16) - 1;
410 	pcb->pcb_ext = NULL;
411 	pcb->pcb_save = get_pcb_user_save_pcb(pcb);
412 #ifdef CPU_ENABLE_SSE
413 	if (use_xsave) {
414 		xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
415 		bzero(xhdr, sizeof(*xhdr));
416 		xhdr->xstate_bv = xsave_mask;
417 	}
418 #endif
419 }
420 
421 void
422 cpu_thread_free(struct thread *td)
423 {
424 
425 	cpu_thread_clean(td);
426 }
427 
428 void
429 cpu_set_syscall_retval(struct thread *td, int error)
430 {
431 
432 	switch (error) {
433 	case 0:
434 		td->td_frame->tf_eax = td->td_retval[0];
435 		td->td_frame->tf_edx = td->td_retval[1];
436 		td->td_frame->tf_eflags &= ~PSL_C;
437 		break;
438 
439 	case ERESTART:
440 		/*
441 		 * Reconstruct pc, assuming lcall $X,y is 7 bytes, int
442 		 * 0x80 is 2 bytes. We saved this in tf_err.
443 		 */
444 		td->td_frame->tf_eip -= td->td_frame->tf_err;
445 		break;
446 
447 	case EJUSTRETURN:
448 		break;
449 
450 	default:
451 		td->td_frame->tf_eax = SV_ABI_ERRNO(td->td_proc, error);
452 		td->td_frame->tf_eflags |= PSL_C;
453 		break;
454 	}
455 }
456 
457 /*
458  * Initialize machine state, mostly pcb and trap frame for a new
459  * thread, about to return to userspace.  Put enough state in the new
460  * thread's PCB to get it to go back to the fork_return(), which
461  * finalizes the thread state and handles peculiarities of the first
462  * return to userspace for the new thread.
463  */
464 void
465 cpu_copy_thread(struct thread *td, struct thread *td0)
466 {
467 	struct pcb *pcb2;
468 
469 	/* Point the pcb to the top of the stack. */
470 	pcb2 = td->td_pcb;
471 
472 	/*
473 	 * Copy the upcall pcb.  This loads kernel regs.
474 	 * Those not loaded individually below get their default
475 	 * values here.
476 	 */
477 	bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
478 	pcb2->pcb_flags &= ~(PCB_NPXINITDONE | PCB_NPXUSERINITDONE |
479 	    PCB_KERNNPX);
480 	pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
481 	bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save,
482 	    cpu_max_ext_state_size);
483 
484 	/*
485 	 * Create a new fresh stack for the new thread.
486 	 */
487 	bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
488 
489 	/* If the current thread has the trap bit set (i.e. a debugger had
490 	 * single stepped the process to the system call), we need to clear
491 	 * the trap flag from the new frame. Otherwise, the new thread will
492 	 * receive a (likely unexpected) SIGTRAP when it executes the first
493 	 * instruction after returning to userland.
494 	 */
495 	td->td_frame->tf_eflags &= ~PSL_T;
496 
497 	/*
498 	 * Set registers for trampoline to user mode.  Leave space for the
499 	 * return address on stack.  These are the kernel mode register values.
500 	 */
501 	pcb2->pcb_edi = 0;
502 	pcb2->pcb_esi = (int)fork_return;		    /* trampoline arg */
503 	pcb2->pcb_ebp = 0;
504 	pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */
505 	pcb2->pcb_ebx = (int)td;			    /* trampoline arg */
506 	pcb2->pcb_eip = (int)fork_trampoline;
507 	pcb2->pcb_psl &= ~(PSL_I);	/* interrupts must be disabled */
508 	pcb2->pcb_gs = rgs();
509 	/*
510 	 * If we didn't copy the pcb, we'd need to do the following registers:
511 	 * pcb2->pcb_cr3:	cloned above.
512 	 * pcb2->pcb_dr*:	cloned above.
513 	 * pcb2->pcb_savefpu:	cloned above.
514 	 * pcb2->pcb_flags:	cloned above.
515 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
516 	 * pcb2->pcb_gs:	cloned above.
517 	 * pcb2->pcb_ext:	cleared below.
518 	 */
519 	pcb2->pcb_ext = NULL;
520 
521 	/* Setup to release spin count in fork_exit(). */
522 	td->td_md.md_spinlock_count = 1;
523 	td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
524 }
525 
526 /*
527  * Set that machine state for performing an upcall that starts
528  * the entry function with the given argument.
529  */
530 void
531 cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg,
532     stack_t *stack)
533 {
534 
535 	/*
536 	 * Do any extra cleaning that needs to be done.
537 	 * The thread may have optional components
538 	 * that are not present in a fresh thread.
539 	 * This may be a recycled thread so make it look
540 	 * as though it's newly allocated.
541 	 */
542 	cpu_thread_clean(td);
543 
544 	/*
545 	 * Set the trap frame to point at the beginning of the entry
546 	 * function.
547 	 */
548 	td->td_frame->tf_ebp = 0;
549 	td->td_frame->tf_esp =
550 	    (((int)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
551 	td->td_frame->tf_eip = (int)entry;
552 
553 	/* Pass the argument to the entry point. */
554 	suword((void *)(td->td_frame->tf_esp + sizeof(void *)),
555 	    (int)arg);
556 }
557 
558 int
559 cpu_set_user_tls(struct thread *td, void *tls_base)
560 {
561 	struct segment_descriptor sd;
562 	uint32_t base;
563 
564 	/*
565 	 * Construct a descriptor and store it in the pcb for
566 	 * the next context switch.  Also store it in the gdt
567 	 * so that the load of tf_fs into %fs will activate it
568 	 * at return to userland.
569 	 */
570 	base = (uint32_t)tls_base;
571 	sd.sd_lobase = base & 0xffffff;
572 	sd.sd_hibase = (base >> 24) & 0xff;
573 	sd.sd_lolimit = 0xffff;	/* 4GB limit, wraps around */
574 	sd.sd_hilimit = 0xf;
575 	sd.sd_type  = SDT_MEMRWA;
576 	sd.sd_dpl   = SEL_UPL;
577 	sd.sd_p     = 1;
578 	sd.sd_xx    = 0;
579 	sd.sd_def32 = 1;
580 	sd.sd_gran  = 1;
581 	critical_enter();
582 	/* set %gs */
583 	td->td_pcb->pcb_gsd = sd;
584 	if (td == curthread) {
585 		PCPU_GET(fsgs_gdt)[1] = sd;
586 		load_gs(GSEL(GUGS_SEL, SEL_UPL));
587 	}
588 	critical_exit();
589 	return (0);
590 }
591 
592 /*
593  * Convert kernel VA to physical address
594  */
595 vm_paddr_t
596 kvtop(void *addr)
597 {
598 	vm_paddr_t pa;
599 
600 	pa = pmap_kextract((vm_offset_t)addr);
601 	if (pa == 0)
602 		panic("kvtop: zero page frame");
603 	return (pa);
604 }
605 
606 #ifdef SMP
607 static void
608 cpu_reset_proxy()
609 {
610 	cpuset_t tcrp;
611 
612 	cpu_reset_proxy_active = 1;
613 	while (cpu_reset_proxy_active == 1)
614 		;	/* Wait for other cpu to see that we've started */
615 	CPU_SETOF(cpu_reset_proxyid, &tcrp);
616 	stop_cpus(tcrp);
617 	printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
618 	DELAY(1000000);
619 	cpu_reset_real();
620 }
621 #endif
622 
623 void
624 cpu_reset()
625 {
626 #ifdef XBOX
627 	if (arch_i386_is_xbox) {
628 		/* Kick the PIC16L, it can reboot the box */
629 		pic16l_reboot();
630 		for (;;);
631 	}
632 #endif
633 
634 #ifdef SMP
635 	cpuset_t map;
636 	u_int cnt;
637 
638 	if (smp_started) {
639 		map = all_cpus;
640 		CPU_CLR(PCPU_GET(cpuid), &map);
641 		CPU_NAND(&map, &stopped_cpus);
642 		if (!CPU_EMPTY(&map)) {
643 			printf("cpu_reset: Stopping other CPUs\n");
644 			stop_cpus(map);
645 		}
646 
647 		if (PCPU_GET(cpuid) != 0) {
648 			cpu_reset_proxyid = PCPU_GET(cpuid);
649 			cpustop_restartfunc = cpu_reset_proxy;
650 			cpu_reset_proxy_active = 0;
651 			printf("cpu_reset: Restarting BSP\n");
652 
653 			/* Restart CPU #0. */
654 			/* XXX: restart_cpus(1 << 0); */
655 			CPU_SETOF(0, &started_cpus);
656 			wmb();
657 
658 			cnt = 0;
659 			while (cpu_reset_proxy_active == 0 && cnt < 10000000)
660 				cnt++;	/* Wait for BSP to announce restart */
661 			if (cpu_reset_proxy_active == 0)
662 				printf("cpu_reset: Failed to restart BSP\n");
663 			enable_intr();
664 			cpu_reset_proxy_active = 2;
665 
666 			while (1);
667 			/* NOTREACHED */
668 		}
669 
670 		DELAY(1000000);
671 	}
672 #endif
673 	cpu_reset_real();
674 	/* NOTREACHED */
675 }
676 
677 static void
678 cpu_reset_real()
679 {
680 	struct region_descriptor null_idt;
681 #ifndef PC98
682 	int b;
683 #endif
684 
685 	disable_intr();
686 #ifdef CPU_ELAN
687 	if (elan_mmcr != NULL)
688 		elan_mmcr->RESCFG = 1;
689 #endif
690 
691 	if (cpu == CPU_GEODE1100) {
692 		/* Attempt Geode's own reset */
693 		outl(0xcf8, 0x80009044ul);
694 		outl(0xcfc, 0xf);
695 	}
696 
697 #ifdef PC98
698 	/*
699 	 * Attempt to do a CPU reset via CPU reset port.
700 	 */
701 	if ((inb(0x35) & 0xa0) != 0xa0) {
702 		outb(0x37, 0x0f);		/* SHUT0 = 0. */
703 		outb(0x37, 0x0b);		/* SHUT1 = 0. */
704 	}
705 	outb(0xf0, 0x00);		/* Reset. */
706 #else
707 #if !defined(BROKEN_KEYBOARD_RESET)
708 	/*
709 	 * Attempt to do a CPU reset via the keyboard controller,
710 	 * do not turn off GateA20, as any machine that fails
711 	 * to do the reset here would then end up in no man's land.
712 	 */
713 	outb(IO_KBD + 4, 0xFE);
714 	DELAY(500000);	/* wait 0.5 sec to see if that did it */
715 #endif
716 
717 	/*
718 	 * Attempt to force a reset via the Reset Control register at
719 	 * I/O port 0xcf9.  Bit 2 forces a system reset when it
720 	 * transitions from 0 to 1.  Bit 1 selects the type of reset
721 	 * to attempt: 0 selects a "soft" reset, and 1 selects a
722 	 * "hard" reset.  We try a "hard" reset.  The first write sets
723 	 * bit 1 to select a "hard" reset and clears bit 2.  The
724 	 * second write forces a 0 -> 1 transition in bit 2 to trigger
725 	 * a reset.
726 	 */
727 	outb(0xcf9, 0x2);
728 	outb(0xcf9, 0x6);
729 	DELAY(500000);  /* wait 0.5 sec to see if that did it */
730 
731 	/*
732 	 * Attempt to force a reset via the Fast A20 and Init register
733 	 * at I/O port 0x92.  Bit 1 serves as an alternate A20 gate.
734 	 * Bit 0 asserts INIT# when set to 1.  We are careful to only
735 	 * preserve bit 1 while setting bit 0.  We also must clear bit
736 	 * 0 before setting it if it isn't already clear.
737 	 */
738 	b = inb(0x92);
739 	if (b != 0xff) {
740 		if ((b & 0x1) != 0)
741 			outb(0x92, b & 0xfe);
742 		outb(0x92, b | 0x1);
743 		DELAY(500000);  /* wait 0.5 sec to see if that did it */
744 	}
745 #endif /* PC98 */
746 
747 	printf("No known reset method worked, attempting CPU shutdown\n");
748 	DELAY(1000000); /* wait 1 sec for printf to complete */
749 
750 	/* Wipe the IDT. */
751 	null_idt.rd_limit = 0;
752 	null_idt.rd_base = 0;
753 	lidt(&null_idt);
754 
755 	/* "good night, sweet prince .... <THUNK!>" */
756 	breakpoint();
757 
758 	/* NOTREACHED */
759 	while(1);
760 }
761 
762 /*
763  * Get an sf_buf from the freelist.  May block if none are available.
764  */
765 void
766 sf_buf_map(struct sf_buf *sf, int flags)
767 {
768 	pt_entry_t opte, *ptep;
769 
770 	/*
771 	 * Update the sf_buf's virtual-to-physical mapping, flushing the
772 	 * virtual address from the TLB.  Since the reference count for
773 	 * the sf_buf's old mapping was zero, that mapping is not
774 	 * currently in use.  Consequently, there is no need to exchange
775 	 * the old and new PTEs atomically, even under PAE.
776 	 */
777 	ptep = vtopte(sf->kva);
778 	opte = *ptep;
779 	*ptep = VM_PAGE_TO_PHYS(sf->m) | pgeflag | PG_RW | PG_V |
780 	    pmap_cache_bits(sf->m->md.pat_mode, 0);
781 
782 	/*
783 	 * Avoid unnecessary TLB invalidations: If the sf_buf's old
784 	 * virtual-to-physical mapping was not used, then any processor
785 	 * that has invalidated the sf_buf's virtual address from its TLB
786 	 * since the last used mapping need not invalidate again.
787 	 */
788 #ifdef SMP
789 	if ((opte & (PG_V | PG_A)) ==  (PG_V | PG_A))
790 		CPU_ZERO(&sf->cpumask);
791 
792 	sf_buf_shootdown(sf, flags);
793 #else
794 	if ((opte & (PG_V | PG_A)) ==  (PG_V | PG_A))
795 		pmap_invalidate_page(kernel_pmap, sf->kva);
796 #endif
797 }
798 
799 #ifdef SMP
800 void
801 sf_buf_shootdown(struct sf_buf *sf, int flags)
802 {
803 	cpuset_t other_cpus;
804 	u_int cpuid;
805 
806 	sched_pin();
807 	cpuid = PCPU_GET(cpuid);
808 	if (!CPU_ISSET(cpuid, &sf->cpumask)) {
809 		CPU_SET(cpuid, &sf->cpumask);
810 		invlpg(sf->kva);
811 	}
812 	if ((flags & SFB_CPUPRIVATE) == 0) {
813 		other_cpus = all_cpus;
814 		CPU_CLR(cpuid, &other_cpus);
815 		CPU_NAND(&other_cpus, &sf->cpumask);
816 		if (!CPU_EMPTY(&other_cpus)) {
817 			CPU_OR(&sf->cpumask, &other_cpus);
818 			smp_masked_invlpg(other_cpus, sf->kva);
819 		}
820 	}
821 	sched_unpin();
822 }
823 #endif
824 
825 /*
826  * MD part of sf_buf_free().
827  */
828 int
829 sf_buf_unmap(struct sf_buf *sf)
830 {
831 
832 	return (0);
833 }
834 
835 static void
836 sf_buf_invalidate(struct sf_buf *sf)
837 {
838 	vm_page_t m = sf->m;
839 
840 	/*
841 	 * Use pmap_qenter to update the pte for
842 	 * existing mapping, in particular, the PAT
843 	 * settings are recalculated.
844 	 */
845 	pmap_qenter(sf->kva, &m, 1);
846 	pmap_invalidate_cache_range(sf->kva, sf->kva + PAGE_SIZE, FALSE);
847 }
848 
849 /*
850  * Invalidate the cache lines that may belong to the page, if
851  * (possibly old) mapping of the page by sf buffer exists.  Returns
852  * TRUE when mapping was found and cache invalidated.
853  */
854 boolean_t
855 sf_buf_invalidate_cache(vm_page_t m)
856 {
857 
858 	return (sf_buf_process_page(m, sf_buf_invalidate));
859 }
860 
861 /*
862  * Software interrupt handler for queued VM system processing.
863  */
864 void
865 swi_vm(void *dummy)
866 {
867 	if (busdma_swi_pending != 0)
868 		busdma_swi();
869 }
870 
871 /*
872  * Tell whether this address is in some physical memory region.
873  * Currently used by the kernel coredump code in order to avoid
874  * dumping the ``ISA memory hole'' which could cause indefinite hangs,
875  * or other unpredictable behaviour.
876  */
877 
878 int
879 is_physical_memory(vm_paddr_t addr)
880 {
881 
882 #ifdef DEV_ISA
883 	/* The ISA ``memory hole''. */
884 	if (addr >= 0xa0000 && addr < 0x100000)
885 		return 0;
886 #endif
887 
888 	/*
889 	 * stuff other tests for known memory-mapped devices (PCI?)
890 	 * here
891 	 */
892 
893 	return 1;
894 }
895