xref: /freebsd/sys/i386/i386/vm_machdep.c (revision 49dae58b287906be26f56ba3e3dc693c3ba8cf37)
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_set_fork_handler(td, func, arg)
328 	struct thread *td;
329 	void (*func)(void *);
330 	void *arg;
331 {
332 	/*
333 	 * Note that the trap frame follows the args, so the function
334 	 * is really called like this:  func(arg, frame);
335 	 */
336 	td->td_pcb->pcb_esi = (int) func;	/* function */
337 	td->td_pcb->pcb_ebx = (int) arg;	/* first arg */
338 }
339 
340 void
341 cpu_exit(struct thread *td)
342 {
343 
344 	/*
345 	 * If this process has a custom LDT, release it.  Reset pc->pcb_gs
346 	 * and %gs before we free it in case they refer to an LDT entry.
347 	 */
348 	mtx_lock_spin(&dt_lock);
349 	if (td->td_proc->p_md.md_ldt) {
350 		td->td_pcb->pcb_gs = _udatasel;
351 		load_gs(_udatasel);
352 		user_ldt_free(td);
353 	} else
354 		mtx_unlock_spin(&dt_lock);
355 }
356 
357 void
358 cpu_thread_exit(struct thread *td)
359 {
360 
361 #ifdef DEV_NPX
362 	critical_enter();
363 	if (td == PCPU_GET(fpcurthread))
364 		npxdrop();
365 	critical_exit();
366 #endif
367 
368 	/* Disable any hardware breakpoints. */
369 	if (td->td_pcb->pcb_flags & PCB_DBREGS) {
370 		reset_dbregs();
371 		td->td_pcb->pcb_flags &= ~PCB_DBREGS;
372 	}
373 }
374 
375 void
376 cpu_thread_clean(struct thread *td)
377 {
378 	struct pcb *pcb;
379 
380 	pcb = td->td_pcb;
381 	if (pcb->pcb_ext != NULL) {
382 		/* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */
383 		/*
384 		 * XXX do we need to move the TSS off the allocated pages
385 		 * before freeing them?  (not done here)
386 		 */
387 		kmem_free(kernel_arena, (vm_offset_t)pcb->pcb_ext,
388 		    ctob(IOPAGES + 1));
389 		pcb->pcb_ext = NULL;
390 	}
391 }
392 
393 void
394 cpu_thread_swapin(struct thread *td)
395 {
396 }
397 
398 void
399 cpu_thread_swapout(struct thread *td)
400 {
401 }
402 
403 void
404 cpu_thread_alloc(struct thread *td)
405 {
406 	struct pcb *pcb;
407 #ifdef CPU_ENABLE_SSE
408 	struct xstate_hdr *xhdr;
409 #endif
410 
411 	td->td_pcb = pcb = get_pcb_td(td);
412 	td->td_frame = (struct trapframe *)((caddr_t)pcb - 16) - 1;
413 	pcb->pcb_ext = NULL;
414 	pcb->pcb_save = get_pcb_user_save_pcb(pcb);
415 #ifdef CPU_ENABLE_SSE
416 	if (use_xsave) {
417 		xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
418 		bzero(xhdr, sizeof(*xhdr));
419 		xhdr->xstate_bv = xsave_mask;
420 	}
421 #endif
422 }
423 
424 void
425 cpu_thread_free(struct thread *td)
426 {
427 
428 	cpu_thread_clean(td);
429 }
430 
431 void
432 cpu_set_syscall_retval(struct thread *td, int error)
433 {
434 
435 	switch (error) {
436 	case 0:
437 		td->td_frame->tf_eax = td->td_retval[0];
438 		td->td_frame->tf_edx = td->td_retval[1];
439 		td->td_frame->tf_eflags &= ~PSL_C;
440 		break;
441 
442 	case ERESTART:
443 		/*
444 		 * Reconstruct pc, assuming lcall $X,y is 7 bytes, int
445 		 * 0x80 is 2 bytes. We saved this in tf_err.
446 		 */
447 		td->td_frame->tf_eip -= td->td_frame->tf_err;
448 		break;
449 
450 	case EJUSTRETURN:
451 		break;
452 
453 	default:
454 		if (td->td_proc->p_sysent->sv_errsize) {
455 			if (error >= td->td_proc->p_sysent->sv_errsize)
456 				error = -1;	/* XXX */
457 			else
458 				error = td->td_proc->p_sysent->sv_errtbl[error];
459 		}
460 		td->td_frame->tf_eax = error;
461 		td->td_frame->tf_eflags |= PSL_C;
462 		break;
463 	}
464 }
465 
466 /*
467  * Initialize machine state (pcb and trap frame) for a new thread about to
468  * upcall. Put enough state in the new thread's PCB to get it to go back
469  * userret(), where we can intercept it again to set the return (upcall)
470  * Address and stack, along with those from upcals that are from other sources
471  * such as those generated in thread_userret() itself.
472  */
473 void
474 cpu_set_upcall(struct thread *td, struct thread *td0)
475 {
476 	struct pcb *pcb2;
477 
478 	/* Point the pcb to the top of the stack. */
479 	pcb2 = td->td_pcb;
480 
481 	/*
482 	 * Copy the upcall pcb.  This loads kernel regs.
483 	 * Those not loaded individually below get their default
484 	 * values here.
485 	 */
486 	bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
487 	pcb2->pcb_flags &= ~(PCB_NPXINITDONE | PCB_NPXUSERINITDONE |
488 	    PCB_KERNNPX);
489 	pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
490 	bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save,
491 	    cpu_max_ext_state_size);
492 
493 	/*
494 	 * Create a new fresh stack for the new thread.
495 	 */
496 	bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
497 
498 	/* If the current thread has the trap bit set (i.e. a debugger had
499 	 * single stepped the process to the system call), we need to clear
500 	 * the trap flag from the new frame. Otherwise, the new thread will
501 	 * receive a (likely unexpected) SIGTRAP when it executes the first
502 	 * instruction after returning to userland.
503 	 */
504 	td->td_frame->tf_eflags &= ~PSL_T;
505 
506 	/*
507 	 * Set registers for trampoline to user mode.  Leave space for the
508 	 * return address on stack.  These are the kernel mode register values.
509 	 */
510 	pcb2->pcb_edi = 0;
511 	pcb2->pcb_esi = (int)fork_return;		    /* trampoline arg */
512 	pcb2->pcb_ebp = 0;
513 	pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */
514 	pcb2->pcb_ebx = (int)td;			    /* trampoline arg */
515 	pcb2->pcb_eip = (int)fork_trampoline;
516 	pcb2->pcb_psl &= ~(PSL_I);	/* interrupts must be disabled */
517 	pcb2->pcb_gs = rgs();
518 	/*
519 	 * If we didn't copy the pcb, we'd need to do the following registers:
520 	 * pcb2->pcb_cr3:	cloned above.
521 	 * pcb2->pcb_dr*:	cloned above.
522 	 * pcb2->pcb_savefpu:	cloned above.
523 	 * pcb2->pcb_flags:	cloned above.
524 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
525 	 * pcb2->pcb_gs:	cloned above.
526 	 * pcb2->pcb_ext:	cleared below.
527 	 */
528 	pcb2->pcb_ext = NULL;
529 
530 	/* Setup to release spin count in fork_exit(). */
531 	td->td_md.md_spinlock_count = 1;
532 	td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
533 }
534 
535 /*
536  * Set that machine state for performing an upcall that has to
537  * be done in thread_userret() so that those upcalls generated
538  * in thread_userret() itself can be done as well.
539  */
540 void
541 cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg,
542 	stack_t *stack)
543 {
544 
545 	/*
546 	 * Do any extra cleaning that needs to be done.
547 	 * The thread may have optional components
548 	 * that are not present in a fresh thread.
549 	 * This may be a recycled thread so make it look
550 	 * as though it's newly allocated.
551 	 */
552 	cpu_thread_clean(td);
553 
554 	/*
555 	 * Set the trap frame to point at the beginning of the uts
556 	 * function.
557 	 */
558 	td->td_frame->tf_ebp = 0;
559 	td->td_frame->tf_esp =
560 	    (((int)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
561 	td->td_frame->tf_eip = (int)entry;
562 
563 	/*
564 	 * Pass the address of the mailbox for this kse to the uts
565 	 * function as a parameter on the stack.
566 	 */
567 	suword((void *)(td->td_frame->tf_esp + sizeof(void *)),
568 	    (int)arg);
569 }
570 
571 int
572 cpu_set_user_tls(struct thread *td, void *tls_base)
573 {
574 	struct segment_descriptor sd;
575 	uint32_t base;
576 
577 	/*
578 	 * Construct a descriptor and store it in the pcb for
579 	 * the next context switch.  Also store it in the gdt
580 	 * so that the load of tf_fs into %fs will activate it
581 	 * at return to userland.
582 	 */
583 	base = (uint32_t)tls_base;
584 	sd.sd_lobase = base & 0xffffff;
585 	sd.sd_hibase = (base >> 24) & 0xff;
586 	sd.sd_lolimit = 0xffff;	/* 4GB limit, wraps around */
587 	sd.sd_hilimit = 0xf;
588 	sd.sd_type  = SDT_MEMRWA;
589 	sd.sd_dpl   = SEL_UPL;
590 	sd.sd_p     = 1;
591 	sd.sd_xx    = 0;
592 	sd.sd_def32 = 1;
593 	sd.sd_gran  = 1;
594 	critical_enter();
595 	/* set %gs */
596 	td->td_pcb->pcb_gsd = sd;
597 	if (td == curthread) {
598 		PCPU_GET(fsgs_gdt)[1] = sd;
599 		load_gs(GSEL(GUGS_SEL, SEL_UPL));
600 	}
601 	critical_exit();
602 	return (0);
603 }
604 
605 /*
606  * Convert kernel VA to physical address
607  */
608 vm_paddr_t
609 kvtop(void *addr)
610 {
611 	vm_paddr_t pa;
612 
613 	pa = pmap_kextract((vm_offset_t)addr);
614 	if (pa == 0)
615 		panic("kvtop: zero page frame");
616 	return (pa);
617 }
618 
619 #ifdef SMP
620 static void
621 cpu_reset_proxy()
622 {
623 	cpuset_t tcrp;
624 
625 	cpu_reset_proxy_active = 1;
626 	while (cpu_reset_proxy_active == 1)
627 		;	/* Wait for other cpu to see that we've started */
628 	CPU_SETOF(cpu_reset_proxyid, &tcrp);
629 	stop_cpus(tcrp);
630 	printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
631 	DELAY(1000000);
632 	cpu_reset_real();
633 }
634 #endif
635 
636 void
637 cpu_reset()
638 {
639 #ifdef XBOX
640 	if (arch_i386_is_xbox) {
641 		/* Kick the PIC16L, it can reboot the box */
642 		pic16l_reboot();
643 		for (;;);
644 	}
645 #endif
646 
647 #ifdef SMP
648 	cpuset_t map;
649 	u_int cnt;
650 
651 	if (smp_started) {
652 		map = all_cpus;
653 		CPU_CLR(PCPU_GET(cpuid), &map);
654 		CPU_NAND(&map, &stopped_cpus);
655 		if (!CPU_EMPTY(&map)) {
656 			printf("cpu_reset: Stopping other CPUs\n");
657 			stop_cpus(map);
658 		}
659 
660 		if (PCPU_GET(cpuid) != 0) {
661 			cpu_reset_proxyid = PCPU_GET(cpuid);
662 			cpustop_restartfunc = cpu_reset_proxy;
663 			cpu_reset_proxy_active = 0;
664 			printf("cpu_reset: Restarting BSP\n");
665 
666 			/* Restart CPU #0. */
667 			/* XXX: restart_cpus(1 << 0); */
668 			CPU_SETOF(0, &started_cpus);
669 			wmb();
670 
671 			cnt = 0;
672 			while (cpu_reset_proxy_active == 0 && cnt < 10000000)
673 				cnt++;	/* Wait for BSP to announce restart */
674 			if (cpu_reset_proxy_active == 0)
675 				printf("cpu_reset: Failed to restart BSP\n");
676 			enable_intr();
677 			cpu_reset_proxy_active = 2;
678 
679 			while (1);
680 			/* NOTREACHED */
681 		}
682 
683 		DELAY(1000000);
684 	}
685 #endif
686 	cpu_reset_real();
687 	/* NOTREACHED */
688 }
689 
690 static void
691 cpu_reset_real()
692 {
693 	struct region_descriptor null_idt;
694 #ifndef PC98
695 	int b;
696 #endif
697 
698 	disable_intr();
699 #ifdef CPU_ELAN
700 	if (elan_mmcr != NULL)
701 		elan_mmcr->RESCFG = 1;
702 #endif
703 
704 	if (cpu == CPU_GEODE1100) {
705 		/* Attempt Geode's own reset */
706 		outl(0xcf8, 0x80009044ul);
707 		outl(0xcfc, 0xf);
708 	}
709 
710 #ifdef PC98
711 	/*
712 	 * Attempt to do a CPU reset via CPU reset port.
713 	 */
714 	if ((inb(0x35) & 0xa0) != 0xa0) {
715 		outb(0x37, 0x0f);		/* SHUT0 = 0. */
716 		outb(0x37, 0x0b);		/* SHUT1 = 0. */
717 	}
718 	outb(0xf0, 0x00);		/* Reset. */
719 #else
720 #if !defined(BROKEN_KEYBOARD_RESET)
721 	/*
722 	 * Attempt to do a CPU reset via the keyboard controller,
723 	 * do not turn off GateA20, as any machine that fails
724 	 * to do the reset here would then end up in no man's land.
725 	 */
726 	outb(IO_KBD + 4, 0xFE);
727 	DELAY(500000);	/* wait 0.5 sec to see if that did it */
728 #endif
729 
730 	/*
731 	 * Attempt to force a reset via the Reset Control register at
732 	 * I/O port 0xcf9.  Bit 2 forces a system reset when it
733 	 * transitions from 0 to 1.  Bit 1 selects the type of reset
734 	 * to attempt: 0 selects a "soft" reset, and 1 selects a
735 	 * "hard" reset.  We try a "hard" reset.  The first write sets
736 	 * bit 1 to select a "hard" reset and clears bit 2.  The
737 	 * second write forces a 0 -> 1 transition in bit 2 to trigger
738 	 * a reset.
739 	 */
740 	outb(0xcf9, 0x2);
741 	outb(0xcf9, 0x6);
742 	DELAY(500000);  /* wait 0.5 sec to see if that did it */
743 
744 	/*
745 	 * Attempt to force a reset via the Fast A20 and Init register
746 	 * at I/O port 0x92.  Bit 1 serves as an alternate A20 gate.
747 	 * Bit 0 asserts INIT# when set to 1.  We are careful to only
748 	 * preserve bit 1 while setting bit 0.  We also must clear bit
749 	 * 0 before setting it if it isn't already clear.
750 	 */
751 	b = inb(0x92);
752 	if (b != 0xff) {
753 		if ((b & 0x1) != 0)
754 			outb(0x92, b & 0xfe);
755 		outb(0x92, b | 0x1);
756 		DELAY(500000);  /* wait 0.5 sec to see if that did it */
757 	}
758 #endif /* PC98 */
759 
760 	printf("No known reset method worked, attempting CPU shutdown\n");
761 	DELAY(1000000); /* wait 1 sec for printf to complete */
762 
763 	/* Wipe the IDT. */
764 	null_idt.rd_limit = 0;
765 	null_idt.rd_base = 0;
766 	lidt(&null_idt);
767 
768 	/* "good night, sweet prince .... <THUNK!>" */
769 	breakpoint();
770 
771 	/* NOTREACHED */
772 	while(1);
773 }
774 
775 /*
776  * Get an sf_buf from the freelist.  May block if none are available.
777  */
778 void
779 sf_buf_map(struct sf_buf *sf, int flags)
780 {
781 	pt_entry_t opte, *ptep;
782 
783 	/*
784 	 * Update the sf_buf's virtual-to-physical mapping, flushing the
785 	 * virtual address from the TLB.  Since the reference count for
786 	 * the sf_buf's old mapping was zero, that mapping is not
787 	 * currently in use.  Consequently, there is no need to exchange
788 	 * the old and new PTEs atomically, even under PAE.
789 	 */
790 	ptep = vtopte(sf->kva);
791 	opte = *ptep;
792 	*ptep = VM_PAGE_TO_PHYS(sf->m) | pgeflag | PG_RW | PG_V |
793 	    pmap_cache_bits(sf->m->md.pat_mode, 0);
794 
795 	/*
796 	 * Avoid unnecessary TLB invalidations: If the sf_buf's old
797 	 * virtual-to-physical mapping was not used, then any processor
798 	 * that has invalidated the sf_buf's virtual address from its TLB
799 	 * since the last used mapping need not invalidate again.
800 	 */
801 #ifdef SMP
802 	if ((opte & (PG_V | PG_A)) ==  (PG_V | PG_A))
803 		CPU_ZERO(&sf->cpumask);
804 
805 	sf_buf_shootdown(sf, flags);
806 #else
807 	if ((opte & (PG_V | PG_A)) ==  (PG_V | PG_A))
808 		pmap_invalidate_page(kernel_pmap, sf->kva);
809 #endif
810 }
811 
812 #ifdef SMP
813 void
814 sf_buf_shootdown(struct sf_buf *sf, int flags)
815 {
816 	cpuset_t other_cpus;
817 	u_int cpuid;
818 
819 	sched_pin();
820 	cpuid = PCPU_GET(cpuid);
821 	if (!CPU_ISSET(cpuid, &sf->cpumask)) {
822 		CPU_SET(cpuid, &sf->cpumask);
823 		invlpg(sf->kva);
824 	}
825 	if ((flags & SFB_CPUPRIVATE) == 0) {
826 		other_cpus = all_cpus;
827 		CPU_CLR(cpuid, &other_cpus);
828 		CPU_NAND(&other_cpus, &sf->cpumask);
829 		if (!CPU_EMPTY(&other_cpus)) {
830 			CPU_OR(&sf->cpumask, &other_cpus);
831 			smp_masked_invlpg(other_cpus, sf->kva);
832 		}
833 	}
834 	sched_unpin();
835 }
836 #endif
837 
838 /*
839  * MD part of sf_buf_free().
840  */
841 int
842 sf_buf_unmap(struct sf_buf *sf)
843 {
844 
845 	return (0);
846 }
847 
848 static void
849 sf_buf_invalidate(struct sf_buf *sf)
850 {
851 	vm_page_t m = sf->m;
852 
853 	/*
854 	 * Use pmap_qenter to update the pte for
855 	 * existing mapping, in particular, the PAT
856 	 * settings are recalculated.
857 	 */
858 	pmap_qenter(sf->kva, &m, 1);
859 	pmap_invalidate_cache_range(sf->kva, sf->kva + PAGE_SIZE, FALSE);
860 }
861 
862 /*
863  * Invalidate the cache lines that may belong to the page, if
864  * (possibly old) mapping of the page by sf buffer exists.  Returns
865  * TRUE when mapping was found and cache invalidated.
866  */
867 boolean_t
868 sf_buf_invalidate_cache(vm_page_t m)
869 {
870 
871 	return (sf_buf_process_page(m, sf_buf_invalidate));
872 }
873 
874 /*
875  * Software interrupt handler for queued VM system processing.
876  */
877 void
878 swi_vm(void *dummy)
879 {
880 	if (busdma_swi_pending != 0)
881 		busdma_swi();
882 }
883 
884 /*
885  * Tell whether this address is in some physical memory region.
886  * Currently used by the kernel coredump code in order to avoid
887  * dumping the ``ISA memory hole'' which could cause indefinite hangs,
888  * or other unpredictable behaviour.
889  */
890 
891 int
892 is_physical_memory(vm_paddr_t addr)
893 {
894 
895 #ifdef DEV_ISA
896 	/* The ISA ``memory hole''. */
897 	if (addr >= 0xa0000 && addr < 0x100000)
898 		return 0;
899 #endif
900 
901 	/*
902 	 * stuff other tests for known memory-mapped devices (PCI?)
903 	 * here
904 	 */
905 
906 	return 1;
907 }
908