xref: /freebsd/sys/i386/i386/vm_machdep.c (revision 63f9a4cb2684a303e3eb2ffed39c03a2e2b28ae0)
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 #ifdef PC98
49 #include "opt_pc98.h"
50 #endif
51 #include "opt_reset.h"
52 #include "opt_cpu.h"
53 
54 #include <sys/param.h>
55 #include <sys/systm.h>
56 #include <sys/bio.h>
57 #include <sys/buf.h>
58 #include <sys/kse.h>
59 #include <sys/kernel.h>
60 #include <sys/ktr.h>
61 #include <sys/lock.h>
62 #include <sys/malloc.h>
63 #include <sys/mbuf.h>
64 #include <sys/mutex.h>
65 #include <sys/proc.h>
66 #include <sys/sf_buf.h>
67 #include <sys/smp.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/vm86.h>
79 
80 #ifdef CPU_ELAN
81 #include <machine/elan_mmcr.h>
82 #endif
83 
84 #include <vm/vm.h>
85 #include <vm/vm_extern.h>
86 #include <vm/vm_kern.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_param.h>
90 
91 #ifdef PC98
92 #include <pc98/pc98/pc98.h>
93 #else
94 #include <i386/isa/isa.h>
95 #endif
96 
97 #ifndef NSFBUFS
98 #define	NSFBUFS		(512 + maxusers * 16)
99 #endif
100 
101 static void	cpu_reset_real(void);
102 static void	sf_buf_init(void *arg);
103 SYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL)
104 
105 LIST_HEAD(sf_head, sf_buf);
106 
107 /*
108  * A hash table of active sendfile(2) buffers
109  */
110 static struct sf_head *sf_buf_active;
111 static u_long sf_buf_hashmask;
112 
113 #define	SF_BUF_HASH(m)	(((m) - vm_page_array) & sf_buf_hashmask)
114 
115 static TAILQ_HEAD(, sf_buf) sf_buf_freelist;
116 static u_int	sf_buf_alloc_want;
117 
118 /*
119  * A lock used to synchronize access to the hash table and free list
120  */
121 static struct mtx sf_buf_lock;
122 
123 extern int	_ucodesel, _udatasel;
124 
125 /*
126  * Finish a fork operation, with process p2 nearly set up.
127  * Copy and update the pcb, set up the stack so that the child
128  * ready to run and return to user mode.
129  */
130 void
131 cpu_fork(td1, p2, td2, flags)
132 	register struct thread *td1;
133 	register struct proc *p2;
134 	struct thread *td2;
135 	int flags;
136 {
137 	register struct proc *p1;
138 	struct pcb *pcb2;
139 	struct mdproc *mdp2;
140 #ifdef DEV_NPX
141 	register_t savecrit;
142 #endif
143 
144 	p1 = td1->td_proc;
145 	if ((flags & RFPROC) == 0) {
146 		if ((flags & RFMEM) == 0) {
147 			/* unshare user LDT */
148 			struct mdproc *mdp1 = &p1->p_md;
149 			struct proc_ldt *pldt = mdp1->md_ldt;
150 			if (pldt && pldt->ldt_refcnt > 1) {
151 				pldt = user_ldt_alloc(mdp1, pldt->ldt_len);
152 				if (pldt == NULL)
153 					panic("could not copy LDT");
154 				mdp1->md_ldt = pldt;
155 				set_user_ldt(mdp1);
156 				user_ldt_free(td1);
157 			}
158 		}
159 		return;
160 	}
161 
162 	/* Ensure that p1's pcb is up to date. */
163 #ifdef DEV_NPX
164 	if (td1 == curthread)
165 		td1->td_pcb->pcb_gs = rgs();
166 	savecrit = intr_disable();
167 	if (PCPU_GET(fpcurthread) == td1)
168 		npxsave(&td1->td_pcb->pcb_save);
169 	intr_restore(savecrit);
170 #endif
171 
172 	/* Point the pcb to the top of the stack */
173 	pcb2 = (struct pcb *)(td2->td_kstack +
174 	    td2->td_kstack_pages * PAGE_SIZE) - 1;
175 	td2->td_pcb = pcb2;
176 
177 	/* Copy p1's pcb */
178 	bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
179 
180 	/* Point mdproc and then copy over td1's contents */
181 	mdp2 = &p2->p_md;
182 	bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
183 
184 	/*
185 	 * Create a new fresh stack for the new process.
186 	 * Copy the trap frame for the return to user mode as if from a
187 	 * syscall.  This copies most of the user mode register values.
188 	 * The -16 is so we can expand the trapframe if we go to vm86.
189 	 */
190 	td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1;
191 	bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
192 
193 	td2->td_frame->tf_eax = 0;		/* Child returns zero */
194 	td2->td_frame->tf_eflags &= ~PSL_C;	/* success */
195 	td2->td_frame->tf_edx = 1;
196 
197 	/*
198 	 * Set registers for trampoline to user mode.  Leave space for the
199 	 * return address on stack.  These are the kernel mode register values.
200 	 */
201 #ifdef PAE
202 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdpt);
203 #else
204 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
205 #endif
206 	pcb2->pcb_edi = 0;
207 	pcb2->pcb_esi = (int)fork_return;	/* fork_trampoline argument */
208 	pcb2->pcb_ebp = 0;
209 	pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
210 	pcb2->pcb_ebx = (int)td2;		/* fork_trampoline argument */
211 	pcb2->pcb_eip = (int)fork_trampoline;
212 	pcb2->pcb_psl = PSL_KERNEL;		/* ints disabled */
213 	pcb2->pcb_gs = rgs();
214 	/*-
215 	 * pcb2->pcb_dr*:	cloned above.
216 	 * pcb2->pcb_savefpu:	cloned above.
217 	 * pcb2->pcb_flags:	cloned above.
218 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
219 	 * pcb2->pcb_gs:	cloned above.
220 	 * pcb2->pcb_ext:	cleared below.
221 	 */
222 
223 	/*
224 	 * XXX don't copy the i/o pages.  this should probably be fixed.
225 	 */
226 	pcb2->pcb_ext = 0;
227 
228         /* Copy the LDT, if necessary. */
229 	mtx_lock_spin(&sched_lock);
230         if (mdp2->md_ldt != 0) {
231 		if (flags & RFMEM) {
232 			mdp2->md_ldt->ldt_refcnt++;
233 		} else {
234 			mdp2->md_ldt = user_ldt_alloc(mdp2,
235 			    mdp2->md_ldt->ldt_len);
236 			if (mdp2->md_ldt == NULL)
237 				panic("could not copy LDT");
238 		}
239         }
240 	mtx_unlock_spin(&sched_lock);
241 
242 	/*
243 	 * Now, cpu_switch() can schedule the new process.
244 	 * pcb_esp is loaded pointing to the cpu_switch() stack frame
245 	 * containing the return address when exiting cpu_switch.
246 	 * This will normally be to fork_trampoline(), which will have
247 	 * %ebx loaded with the new proc's pointer.  fork_trampoline()
248 	 * will set up a stack to call fork_return(p, frame); to complete
249 	 * the return to user-mode.
250 	 */
251 }
252 
253 /*
254  * Intercept the return address from a freshly forked process that has NOT
255  * been scheduled yet.
256  *
257  * This is needed to make kernel threads stay in kernel mode.
258  */
259 void
260 cpu_set_fork_handler(td, func, arg)
261 	struct thread *td;
262 	void (*func)(void *);
263 	void *arg;
264 {
265 	/*
266 	 * Note that the trap frame follows the args, so the function
267 	 * is really called like this:  func(arg, frame);
268 	 */
269 	td->td_pcb->pcb_esi = (int) func;	/* function */
270 	td->td_pcb->pcb_ebx = (int) arg;	/* first arg */
271 }
272 
273 void
274 cpu_exit(struct thread *td)
275 {
276 	struct mdproc *mdp;
277 	struct pcb *pcb = td->td_pcb;
278 
279 
280 	/* Reset pc->pcb_gs and %gs before possibly invalidating it. */
281 	mdp = &td->td_proc->p_md;
282 	if (mdp->md_ldt) {
283 		td->td_pcb->pcb_gs = _udatasel;
284 		load_gs(_udatasel);
285 		user_ldt_free(td);
286 	}
287 	if (pcb->pcb_flags & PCB_DBREGS) {
288 		/* disable all hardware breakpoints */
289 		reset_dbregs();
290 		pcb->pcb_flags &= ~PCB_DBREGS;
291 	}
292 }
293 
294 void
295 cpu_thread_exit(struct thread *td)
296 {
297 	struct pcb *pcb = td->td_pcb;
298 #ifdef DEV_NPX
299 	if (td == PCPU_GET(fpcurthread))
300 		npxdrop();
301 #endif
302         if (pcb->pcb_flags & PCB_DBREGS) {
303 		/* disable all hardware breakpoints */
304                 reset_dbregs();
305                 pcb->pcb_flags &= ~PCB_DBREGS;
306         }
307 }
308 
309 void
310 cpu_thread_clean(struct thread *td)
311 {
312 	struct pcb *pcb;
313 
314 	pcb = td->td_pcb;
315 	if (pcb->pcb_ext != 0) {
316 		/* XXXKSE  XXXSMP  not SMP SAFE.. what locks do we have? */
317 		/* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */
318 		/*
319 		 * XXX do we need to move the TSS off the allocated pages
320 		 * before freeing them?  (not done here)
321 		 */
322 		kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
323 		    ctob(IOPAGES + 1));
324 		pcb->pcb_ext = 0;
325 	}
326 }
327 
328 void
329 cpu_thread_swapin(struct thread *td)
330 {
331 }
332 
333 void
334 cpu_thread_swapout(struct thread *td)
335 {
336 }
337 
338 void
339 cpu_thread_setup(struct thread *td)
340 {
341 
342 	td->td_pcb = (struct pcb *)(td->td_kstack +
343 	    td->td_kstack_pages * PAGE_SIZE) - 1;
344 	td->td_frame = (struct trapframe *)((caddr_t)td->td_pcb - 16) - 1;
345 	td->td_pcb->pcb_ext = NULL;
346 }
347 
348 /*
349  * Initialize machine state (pcb and trap frame) for a new thread about to
350  * upcall. Pu t enough state in the new thread's PCB to get it to go back
351  * userret(), where we can intercept it again to set the return (upcall)
352  * Address and stack, along with those from upcals that are from other sources
353  * such as those generated in thread_userret() itself.
354  */
355 void
356 cpu_set_upcall(struct thread *td, struct thread *td0)
357 {
358 	struct pcb *pcb2;
359 
360 	/* Point the pcb to the top of the stack. */
361 	pcb2 = td->td_pcb;
362 
363 	/*
364 	 * Copy the upcall pcb.  This loads kernel regs.
365 	 * Those not loaded individually below get their default
366 	 * values here.
367 	 *
368 	 * XXXKSE It might be a good idea to simply skip this as
369 	 * the values of the other registers may be unimportant.
370 	 * This would remove any requirement for knowing the KSE
371 	 * at this time (see the matching comment below for
372 	 * more analysis) (need a good safe default).
373 	 */
374 	bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
375 	pcb2->pcb_flags &= ~(PCB_NPXTRAP|PCB_NPXINITDONE);
376 
377 	/*
378 	 * Create a new fresh stack for the new thread.
379 	 * The -16 is so we can expand the trapframe if we go to vm86.
380 	 * Don't forget to set this stack value into whatever supplies
381 	 * the address for the fault handlers.
382 	 * The contexts are filled in at the time we actually DO the
383 	 * upcall as only then do we know which KSE we got.
384 	 */
385 	bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
386 
387 	/*
388 	 * Set registers for trampoline to user mode.  Leave space for the
389 	 * return address on stack.  These are the kernel mode register values.
390 	 */
391 #ifdef PAE
392 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pdpt);
393 #else
394 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pdir);
395 #endif
396 	pcb2->pcb_edi = 0;
397 	pcb2->pcb_esi = (int)fork_return;		    /* trampoline arg */
398 	pcb2->pcb_ebp = 0;
399 	pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */
400 	pcb2->pcb_ebx = (int)td;			    /* trampoline arg */
401 	pcb2->pcb_eip = (int)fork_trampoline;
402 	pcb2->pcb_psl &= ~(PSL_I);	/* interrupts must be disabled */
403 	pcb2->pcb_gs = rgs();
404 	/*
405 	 * If we didn't copy the pcb, we'd need to do the following registers:
406 	 * pcb2->pcb_dr*:	cloned above.
407 	 * pcb2->pcb_savefpu:	cloned above.
408 	 * pcb2->pcb_flags:	cloned above.
409 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
410 	 * pcb2->pcb_gs:	cloned above.  XXXKSE ???
411 	 * pcb2->pcb_ext:	cleared below.
412 	 */
413 	 pcb2->pcb_ext = NULL;
414 }
415 
416 /*
417  * Set that machine state for performing an upcall that has to
418  * be done in thread_userret() so that those upcalls generated
419  * in thread_userret() itself can be done as well.
420  */
421 void
422 cpu_set_upcall_kse(struct thread *td, struct kse_upcall *ku)
423 {
424 
425 	/*
426 	 * Do any extra cleaning that needs to be done.
427 	 * The thread may have optional components
428 	 * that are not present in a fresh thread.
429 	 * This may be a recycled thread so make it look
430 	 * as though it's newly allocated.
431 	 */
432 	cpu_thread_clean(td);
433 
434 	/*
435 	 * Set the trap frame to point at the beginning of the uts
436 	 * function.
437 	 */
438 	td->td_frame->tf_ebp = 0;
439 	td->td_frame->tf_esp =
440 	    (int)ku->ku_stack.ss_sp + ku->ku_stack.ss_size - 16;
441 	td->td_frame->tf_eip = (int)ku->ku_func;
442 
443 	/*
444 	 * Pass the address of the mailbox for this kse to the uts
445 	 * function as a parameter on the stack.
446 	 */
447 	suword((void *)(td->td_frame->tf_esp + sizeof(void *)),
448 	    (int)ku->ku_mailbox);
449 }
450 
451 /*
452  * Convert kernel VA to physical address
453  */
454 vm_paddr_t
455 kvtop(void *addr)
456 {
457 	vm_paddr_t pa;
458 
459 	pa = pmap_kextract((vm_offset_t)addr);
460 	if (pa == 0)
461 		panic("kvtop: zero page frame");
462 	return (pa);
463 }
464 
465 void
466 cpu_reset()
467 {
468 #ifdef SMP
469 	u_int map;
470 
471 	if (smp_active) {
472 		map = PCPU_GET(other_cpus) & ~stopped_cpus;
473 		if (map != 0) {
474 			printf("cpu_reset: Stopping other CPUs\n");
475 			stop_cpus(map);
476 		}
477 		DELAY(1000000);
478 	}
479 #endif
480 	cpu_reset_real();
481 	/* NOTREACHED */
482 }
483 
484 static void
485 cpu_reset_real()
486 {
487 
488 #ifdef CPU_ELAN
489 	if (elan_mmcr != NULL)
490 		elan_mmcr->RESCFG = 1;
491 #endif
492 
493 	if (cpu == CPU_GEODE1100) {
494 		/* Attempt Geode's own reset */
495 		outl(0xcf8, 0x80009044ul);
496 		outl(0xcfc, 0xf);
497 	}
498 
499 #ifdef PC98
500 	/*
501 	 * Attempt to do a CPU reset via CPU reset port.
502 	 */
503 	disable_intr();
504 	if ((inb(0x35) & 0xa0) != 0xa0) {
505 		outb(0x37, 0x0f);		/* SHUT0 = 0. */
506 		outb(0x37, 0x0b);		/* SHUT1 = 0. */
507 	}
508 	outb(0xf0, 0x00);		/* Reset. */
509 #else
510 #if !defined(BROKEN_KEYBOARD_RESET)
511 	/*
512 	 * Attempt to do a CPU reset via the keyboard controller,
513 	 * do not turn off GateA20, as any machine that fails
514 	 * to do the reset here would then end up in no man's land.
515 	 */
516 	outb(IO_KBD + 4, 0xFE);
517 	DELAY(500000);	/* wait 0.5 sec to see if that did it */
518 	printf("Keyboard reset did not work, attempting CPU shutdown\n");
519 	DELAY(1000000);	/* wait 1 sec for printf to complete */
520 #endif
521 #endif /* PC98 */
522 
523 	/* Force a shutdown by unmapping entire address space. */
524 	bzero((caddr_t)PTD, NBPTD);
525 
526 	/* "good night, sweet prince .... <THUNK!>" */
527 	invltlb();
528 	/* NOTREACHED */
529 	while(1);
530 }
531 
532 /*
533  * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-))
534  */
535 static void
536 sf_buf_init(void *arg)
537 {
538 	struct sf_buf *sf_bufs;
539 	vm_offset_t sf_base;
540 	int i;
541 
542 	nsfbufs = NSFBUFS;
543 	TUNABLE_INT_FETCH("kern.ipc.nsfbufs", &nsfbufs);
544 
545 	sf_buf_active = hashinit(nsfbufs, M_TEMP, &sf_buf_hashmask);
546 	TAILQ_INIT(&sf_buf_freelist);
547 	sf_base = kmem_alloc_nofault(kernel_map, nsfbufs * PAGE_SIZE);
548 	sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP,
549 	    M_NOWAIT | M_ZERO);
550 	for (i = 0; i < nsfbufs; i++) {
551 		sf_bufs[i].kva = sf_base + i * PAGE_SIZE;
552 		TAILQ_INSERT_TAIL(&sf_buf_freelist, &sf_bufs[i], free_entry);
553 	}
554 	sf_buf_alloc_want = 0;
555 	mtx_init(&sf_buf_lock, "sf_buf", NULL, MTX_DEF);
556 }
557 
558 /*
559  * Get an sf_buf from the freelist. Will block if none are available.
560  */
561 struct sf_buf *
562 sf_buf_alloc(struct vm_page *m, int flags)
563 {
564 	pt_entry_t opte, *ptep;
565 	struct sf_head *hash_list;
566 	struct sf_buf *sf;
567 	int error;
568 
569 	hash_list = &sf_buf_active[SF_BUF_HASH(m)];
570 	mtx_lock(&sf_buf_lock);
571 	LIST_FOREACH(sf, hash_list, list_entry) {
572 		if (sf->m == m) {
573 			sf->ref_count++;
574 			if (sf->ref_count == 1) {
575 				TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
576 				nsfbufsused++;
577 				nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
578 			}
579 			goto done;
580 		}
581 	}
582 	while ((sf = TAILQ_FIRST(&sf_buf_freelist)) == NULL) {
583 		if (flags & SFB_NOWAIT)
584 			goto done;
585 		sf_buf_alloc_want++;
586 		mbstat.sf_allocwait++;
587 		error = msleep(&sf_buf_freelist, &sf_buf_lock,
588 		    (flags & SFB_CATCH) ? PCATCH | PVM : PVM, "sfbufa", 0);
589 		sf_buf_alloc_want--;
590 
591 		/*
592 		 * If we got a signal, don't risk going back to sleep.
593 		 */
594 		if (error)
595 			goto done;
596 	}
597 	TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
598 	if (sf->m != NULL)
599 		LIST_REMOVE(sf, list_entry);
600 	LIST_INSERT_HEAD(hash_list, sf, list_entry);
601 	sf->ref_count = 1;
602 	sf->m = m;
603 	nsfbufsused++;
604 	nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
605 
606 	/*
607 	 * Update the sf_buf's virtual-to-physical mapping, flushing the
608 	 * virtual address from the TLB only if the PTE implies that the old
609 	 * mapping has been used.  Since the reference count for the sf_buf's
610 	 * old mapping was zero, that mapping is not currently in use.
611 	 * Consequently, there is no need to exchange the old and new PTEs
612 	 * atomically, even under PAE.
613 	 */
614 	ptep = vtopte(sf->kva);
615 	opte = *ptep;
616 	*ptep = VM_PAGE_TO_PHYS(m) | pgeflag | PG_RW | PG_V;
617 	if ((opte & (PG_A | PG_V)) == (PG_A | PG_V))
618 		pmap_invalidate_page(kernel_pmap, sf->kva);
619 done:
620 	mtx_unlock(&sf_buf_lock);
621 	return (sf);
622 }
623 
624 /*
625  * Remove a reference from the given sf_buf, adding it to the free
626  * list when its reference count reaches zero.  A freed sf_buf still,
627  * however, retains its virtual-to-physical mapping until it is
628  * recycled or reactivated by sf_buf_alloc(9).
629  */
630 void
631 sf_buf_free(struct sf_buf *sf)
632 {
633 
634 	mtx_lock(&sf_buf_lock);
635 	sf->ref_count--;
636 	if (sf->ref_count == 0) {
637 		TAILQ_INSERT_TAIL(&sf_buf_freelist, sf, free_entry);
638 		nsfbufsused--;
639 		if (sf_buf_alloc_want > 0)
640 			wakeup_one(&sf_buf_freelist);
641 	}
642 	mtx_unlock(&sf_buf_lock);
643 }
644 
645 /*
646  * Software interrupt handler for queued VM system processing.
647  */
648 void
649 swi_vm(void *dummy)
650 {
651 	if (busdma_swi_pending != 0)
652 		busdma_swi();
653 }
654 
655 /*
656  * Tell whether this address is in some physical memory region.
657  * Currently used by the kernel coredump code in order to avoid
658  * dumping the ``ISA memory hole'' which could cause indefinite hangs,
659  * or other unpredictable behaviour.
660  */
661 
662 int
663 is_physical_memory(vm_paddr_t addr)
664 {
665 
666 #ifdef DEV_ISA
667 	/* The ISA ``memory hole''. */
668 	if (addr >= 0xa0000 && addr < 0x100000)
669 		return 0;
670 #endif
671 
672 	/*
673 	 * stuff other tests for known memory-mapped devices (PCI?)
674 	 * here
675 	 */
676 
677 	return 1;
678 }
679