xref: /freebsd/sys/i386/i386/vm_machdep.c (revision c6a37e84139a1c73d4ef46ce4fdf8598a0ebbf45)
15b81b6b3SRodney W. Grimes /*-
25b81b6b3SRodney W. Grimes  * Copyright (c) 1982, 1986 The Regents of the University of California.
35b81b6b3SRodney W. Grimes  * Copyright (c) 1989, 1990 William Jolitz
41561d038SDavid Greenman  * Copyright (c) 1994 John Dyson
55b81b6b3SRodney W. Grimes  * All rights reserved.
65b81b6b3SRodney W. Grimes  *
75b81b6b3SRodney W. Grimes  * This code is derived from software contributed to Berkeley by
85b81b6b3SRodney W. Grimes  * the Systems Programming Group of the University of Utah Computer
95b81b6b3SRodney W. Grimes  * Science Department, and William Jolitz.
105b81b6b3SRodney W. Grimes  *
115b81b6b3SRodney W. Grimes  * Redistribution and use in source and binary forms, with or without
125b81b6b3SRodney W. Grimes  * modification, are permitted provided that the following conditions
135b81b6b3SRodney W. Grimes  * are met:
145b81b6b3SRodney W. Grimes  * 1. Redistributions of source code must retain the above copyright
155b81b6b3SRodney W. Grimes  *    notice, this list of conditions and the following disclaimer.
165b81b6b3SRodney W. Grimes  * 2. Redistributions in binary form must reproduce the above copyright
175b81b6b3SRodney W. Grimes  *    notice, this list of conditions and the following disclaimer in the
185b81b6b3SRodney W. Grimes  *    documentation and/or other materials provided with the distribution.
195b81b6b3SRodney W. Grimes  * 3. All advertising materials mentioning features or use of this software
205b81b6b3SRodney W. Grimes  *    must display the following acknowledgement:
215b81b6b3SRodney W. Grimes  *	This product includes software developed by the University of
225b81b6b3SRodney W. Grimes  *	California, Berkeley and its contributors.
235b81b6b3SRodney W. Grimes  * 4. Neither the name of the University nor the names of its contributors
245b81b6b3SRodney W. Grimes  *    may be used to endorse or promote products derived from this software
255b81b6b3SRodney W. Grimes  *    without specific prior written permission.
265b81b6b3SRodney W. Grimes  *
275b81b6b3SRodney W. Grimes  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
285b81b6b3SRodney W. Grimes  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
295b81b6b3SRodney W. Grimes  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
305b81b6b3SRodney W. Grimes  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
315b81b6b3SRodney W. Grimes  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
325b81b6b3SRodney W. Grimes  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
335b81b6b3SRodney W. Grimes  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
345b81b6b3SRodney W. Grimes  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
355b81b6b3SRodney W. Grimes  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
365b81b6b3SRodney W. Grimes  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
375b81b6b3SRodney W. Grimes  * SUCH DAMAGE.
385b81b6b3SRodney W. Grimes  *
39960173b9SRodney W. Grimes  *	from: @(#)vm_machdep.c	7.3 (Berkeley) 5/13/91
405b81b6b3SRodney W. Grimes  *	Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
415b81b6b3SRodney W. Grimes  */
425b81b6b3SRodney W. Grimes 
439676a785SDavid E. O'Brien #include <sys/cdefs.h>
449676a785SDavid E. O'Brien __FBSDID("$FreeBSD$");
459676a785SDavid E. O'Brien 
466d7d4649SBruce Evans #include "opt_isa.h"
47558226eaSPeter Wemm #include "opt_npx.h"
4861324207SKATO Takenori #ifdef PC98
4961324207SKATO Takenori #include "opt_pc98.h"
5061324207SKATO Takenori #endif
510d74cc48SPeter Wemm #include "opt_reset.h"
52883bd55aSPoul-Henning Kamp #include "opt_cpu.h"
538890984dSGarrett Wollman 
5426f9a767SRodney W. Grimes #include <sys/param.h>
5526f9a767SRodney W. Grimes #include <sys/systm.h>
569626b608SPoul-Henning Kamp #include <sys/bio.h>
5726f9a767SRodney W. Grimes #include <sys/buf.h>
586d7d4649SBruce Evans #include <sys/kse.h>
5966095752SJohn Dyson #include <sys/kernel.h>
600384fff8SJason Evans #include <sys/ktr.h>
616d7d4649SBruce Evans #include <sys/lock.h>
626d7d4649SBruce Evans #include <sys/malloc.h>
63411d10a6SAlan Cox #include <sys/mbuf.h>
6413b0500fSJohn Baldwin #include <sys/mutex.h>
655ad5504cSKelly Yancey #include <sys/pioctl.h>
666d7d4649SBruce Evans #include <sys/proc.h>
67e45db9b8SAlan Cox #include <sys/sf_buf.h>
686d7d4649SBruce Evans #include <sys/smp.h>
6966095752SJohn Dyson #include <sys/sysctl.h>
7091c28bfdSLuoqi Chen #include <sys/unistd.h>
716d7d4649SBruce Evans #include <sys/vnode.h>
726d7d4649SBruce Evans #include <sys/vmmeter.h>
735b81b6b3SRodney W. Grimes 
74a2a1c95cSPeter Wemm #include <machine/cpu.h>
7541ee9f1cSPoul-Henning Kamp #include <machine/cputypes.h>
761f8745a9SPeter Wemm #include <machine/md_var.h>
7791c28bfdSLuoqi Chen #include <machine/pcb.h>
7891c28bfdSLuoqi Chen #include <machine/pcb_ext.h>
795c0db7c7SAlan Cox #include <machine/smp.h>
8091c28bfdSLuoqi Chen #include <machine/vm86.h>
815b81b6b3SRodney W. Grimes 
82883bd55aSPoul-Henning Kamp #ifdef CPU_ELAN
83883bd55aSPoul-Henning Kamp #include <machine/elan_mmcr.h>
84883bd55aSPoul-Henning Kamp #endif
85883bd55aSPoul-Henning Kamp 
8626f9a767SRodney W. Grimes #include <vm/vm.h>
876d7d4649SBruce Evans #include <vm/vm_extern.h>
8826f9a767SRodney W. Grimes #include <vm/vm_kern.h>
8924a1cce3SDavid Greenman #include <vm/vm_page.h>
90efeaf95aSDavid Greenman #include <vm/vm_map.h>
916d7d4649SBruce Evans #include <vm/vm_param.h>
925b81b6b3SRodney W. Grimes 
93e30f0011SSatoshi Asami #ifdef PC98
94e30f0011SSatoshi Asami #include <pc98/pc98/pc98.h>
95e30f0011SSatoshi Asami #else
962320728fSRodney W. Grimes #include <i386/isa/isa.h>
97e30f0011SSatoshi Asami #endif
982320728fSRodney W. Grimes 
99099a0e58SBosko Milekic #ifndef NSFBUFS
100099a0e58SBosko Milekic #define	NSFBUFS		(512 + maxusers * 16)
101099a0e58SBosko Milekic #endif
102099a0e58SBosko Milekic 
10315fe3067SAlfred Perlstein static void	cpu_reset_real(void);
1044260b00aSNate Lawson #ifdef SMP
1054260b00aSNate Lawson static void	cpu_reset_proxy(void);
1064260b00aSNate Lawson static u_int	cpu_reset_proxyid;
1074260b00aSNate Lawson static volatile u_int	cpu_reset_proxy_active;
1084260b00aSNate Lawson #endif
109411d10a6SAlan Cox static void	sf_buf_init(void *arg);
110411d10a6SAlan Cox SYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL)
111411d10a6SAlan Cox 
1120543fa53SAlan Cox LIST_HEAD(sf_head, sf_buf);
113411d10a6SAlan Cox 
1140543fa53SAlan Cox /*
1150543fa53SAlan Cox  * A hash table of active sendfile(2) buffers
1160543fa53SAlan Cox  */
1170543fa53SAlan Cox static struct sf_head *sf_buf_active;
1180543fa53SAlan Cox static u_long sf_buf_hashmask;
1190543fa53SAlan Cox 
1200543fa53SAlan Cox #define	SF_BUF_HASH(m)	(((m) - vm_page_array) & sf_buf_hashmask)
1210543fa53SAlan Cox 
122a5819cb5SAlan Cox static TAILQ_HEAD(, sf_buf) sf_buf_freelist;
123411d10a6SAlan Cox static u_int	sf_buf_alloc_want;
124411d10a6SAlan Cox 
1250543fa53SAlan Cox /*
1260543fa53SAlan Cox  * A lock used to synchronize access to the hash table and free list
1270543fa53SAlan Cox  */
1280543fa53SAlan Cox static struct mtx sf_buf_lock;
1290543fa53SAlan Cox 
13037b087a6SPeter Wemm extern int	_ucodesel, _udatasel;
1312f1e7069STor Egge 
132a4f7a4c9SDavid Greenman /*
1335b81b6b3SRodney W. Grimes  * Finish a fork operation, with process p2 nearly set up.
134a2a1c95cSPeter Wemm  * Copy and update the pcb, set up the stack so that the child
135a2a1c95cSPeter Wemm  * ready to run and return to user mode.
1365b81b6b3SRodney W. Grimes  */
137a2a1c95cSPeter Wemm void
138079b7badSJulian Elischer cpu_fork(td1, p2, td2, flags)
139b40ce416SJulian Elischer 	register struct thread *td1;
140b40ce416SJulian Elischer 	register struct proc *p2;
141079b7badSJulian Elischer 	struct thread *td2;
14291c28bfdSLuoqi Chen 	int flags;
1435b81b6b3SRodney W. Grimes {
144b40ce416SJulian Elischer 	register struct proc *p1;
14591c28bfdSLuoqi Chen 	struct pcb *pcb2;
14624db0459SJohn Baldwin 	struct mdproc *mdp2;
147a4b8c657SBruce Evans #ifdef DEV_NPX
148ba74981eSWarner Losh 	register_t savecrit;
149a4b8c657SBruce Evans #endif
15091c28bfdSLuoqi Chen 
151b40ce416SJulian Elischer 	p1 = td1->td_proc;
15291c28bfdSLuoqi Chen 	if ((flags & RFPROC) == 0) {
15391c28bfdSLuoqi Chen 		if ((flags & RFMEM) == 0) {
15491c28bfdSLuoqi Chen 			/* unshare user LDT */
15545449f84SJulian Elischer 			struct mdproc *mdp1 = &p1->p_md;
15624db0459SJohn Baldwin 			struct proc_ldt *pldt = mdp1->md_ldt;
15724db0459SJohn Baldwin 			if (pldt && pldt->ldt_refcnt > 1) {
15824db0459SJohn Baldwin 				pldt = user_ldt_alloc(mdp1, pldt->ldt_len);
15924db0459SJohn Baldwin 				if (pldt == NULL)
1601acf256dSJohn Baldwin 					panic("could not copy LDT");
16124db0459SJohn Baldwin 				mdp1->md_ldt = pldt;
16224db0459SJohn Baldwin 				set_user_ldt(mdp1);
16324db0459SJohn Baldwin 				user_ldt_free(td1);
16491c28bfdSLuoqi Chen 			}
16591c28bfdSLuoqi Chen 		}
16691c28bfdSLuoqi Chen 		return;
16791c28bfdSLuoqi Chen 	}
1685b81b6b3SRodney W. Grimes 
1691f8745a9SPeter Wemm 	/* Ensure that p1's pcb is up to date. */
170a4b8c657SBruce Evans #ifdef DEV_NPX
171b40ce416SJulian Elischer 	if (td1 == curthread)
172b40ce416SJulian Elischer 		td1->td_pcb->pcb_gs = rgs();
173ba74981eSWarner Losh 	savecrit = intr_disable();
1740bbc8826SJohn Baldwin 	if (PCPU_GET(fpcurthread) == td1)
175b40ce416SJulian Elischer 		npxsave(&td1->td_pcb->pcb_save);
176ba74981eSWarner Losh 	intr_restore(savecrit);
1779f449d2aSBruce Evans #endif
1781f8745a9SPeter Wemm 
179b40ce416SJulian Elischer 	/* Point the pcb to the top of the stack */
1806041f1a5SBill Paul 	pcb2 = (struct pcb *)(td2->td_kstack +
1816041f1a5SBill Paul 	    td2->td_kstack_pages * PAGE_SIZE) - 1;
182b40ce416SJulian Elischer 	td2->td_pcb = pcb2;
183b40ce416SJulian Elischer 
18424db0459SJohn Baldwin 	/* Copy p1's pcb */
185b40ce416SJulian Elischer 	bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
186a2a1c95cSPeter Wemm 
18724db0459SJohn Baldwin 	/* Point mdproc and then copy over td1's contents */
18845449f84SJulian Elischer 	mdp2 = &p2->p_md;
18945449f84SJulian Elischer 	bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
19024db0459SJohn Baldwin 
191a2a1c95cSPeter Wemm 	/*
192a2a1c95cSPeter Wemm 	 * Create a new fresh stack for the new process.
1931f8745a9SPeter Wemm 	 * Copy the trap frame for the return to user mode as if from a
194a4b8c657SBruce Evans 	 * syscall.  This copies most of the user mode register values.
195e5a860ebSPeter Wemm 	 * The -16 is so we can expand the trapframe if we go to vm86.
196a2a1c95cSPeter Wemm 	 */
197e5a860ebSPeter Wemm 	td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1;
198b40ce416SJulian Elischer 	bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
199a2a1c95cSPeter Wemm 
200b40ce416SJulian Elischer 	td2->td_frame->tf_eax = 0;		/* Child returns zero */
201b40ce416SJulian Elischer 	td2->td_frame->tf_eflags &= ~PSL_C;	/* success */
202b40ce416SJulian Elischer 	td2->td_frame->tf_edx = 1;
2030d54b9d1SJohn Baldwin 
204a2a1c95cSPeter Wemm 	/*
2055ad5504cSKelly Yancey 	 * If the parent process has the trap bit set (i.e. a debugger had
2065ad5504cSKelly Yancey 	 * single stepped the process to the system call), we need to clear
2075ad5504cSKelly Yancey 	 * the trap flag from the new frame unless the debugger had set PF_FORK
2085ad5504cSKelly Yancey 	 * on the parent.  Otherwise, the child will receive a (likely
2095ad5504cSKelly Yancey 	 * unexpected) SIGTRAP when it executes the first instruction after
2105ad5504cSKelly Yancey 	 * returning  to userland.
2115ad5504cSKelly Yancey 	 */
2125ad5504cSKelly Yancey 	if ((p1->p_pfsflags & PF_FORK) == 0)
2135ad5504cSKelly Yancey 		td2->td_frame->tf_eflags &= ~PSL_T;
2145ad5504cSKelly Yancey 
2155ad5504cSKelly Yancey 	/*
216a2a1c95cSPeter Wemm 	 * Set registers for trampoline to user mode.  Leave space for the
217a2a1c95cSPeter Wemm 	 * return address on stack.  These are the kernel mode register values.
218a2a1c95cSPeter Wemm 	 */
2197ab9b220SJake Burkholder #ifdef PAE
2207ab9b220SJake Burkholder 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdpt);
2217ab9b220SJake Burkholder #else
222b1028ad1SLuoqi Chen 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
2237ab9b220SJake Burkholder #endif
22437b087a6SPeter Wemm 	pcb2->pcb_edi = 0;
22537b087a6SPeter Wemm 	pcb2->pcb_esi = (int)fork_return;	/* fork_trampoline argument */
22637b087a6SPeter Wemm 	pcb2->pcb_ebp = 0;
227b40ce416SJulian Elischer 	pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
228b40ce416SJulian Elischer 	pcb2->pcb_ebx = (int)td2;		/* fork_trampoline argument */
2291f8745a9SPeter Wemm 	pcb2->pcb_eip = (int)fork_trampoline;
230a93b6bf5SThomas Moestl 	pcb2->pcb_psl = PSL_KERNEL;		/* ints disabled */
2312257a44fSDavid Xu 	pcb2->pcb_gs = rgs();
232a4b8c657SBruce Evans 	/*-
233a4b8c657SBruce Evans 	 * pcb2->pcb_dr*:	cloned above.
2341f8745a9SPeter Wemm 	 * pcb2->pcb_savefpu:	cloned above.
235a4b8c657SBruce Evans 	 * pcb2->pcb_flags:	cloned above.
2361f8745a9SPeter Wemm 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
237a4b8c657SBruce Evans 	 * pcb2->pcb_gs:	cloned above.
238a4b8c657SBruce Evans 	 * pcb2->pcb_ext:	cleared below.
2391f8745a9SPeter Wemm 	 */
2405b81b6b3SRodney W. Grimes 
24148a09cf2SJohn Dyson 	/*
24248a09cf2SJohn Dyson 	 * XXX don't copy the i/o pages.  this should probably be fixed.
24348a09cf2SJohn Dyson 	 */
24448a09cf2SJohn Dyson 	pcb2->pcb_ext = 0;
24548a09cf2SJohn Dyson 
2468c39a127SStefan Eßer 	/* Copy the LDT, if necessary. */
247df4d012bSJohn Baldwin 	mtx_lock_spin(&sched_lock);
248753d1af1SJohn Baldwin 	if (mdp2->md_ldt != NULL) {
24991c28bfdSLuoqi Chen 		if (flags & RFMEM) {
25024db0459SJohn Baldwin 			mdp2->md_ldt->ldt_refcnt++;
25191c28bfdSLuoqi Chen 		} else {
25224db0459SJohn Baldwin 			mdp2->md_ldt = user_ldt_alloc(mdp2,
25324db0459SJohn Baldwin 			    mdp2->md_ldt->ldt_len);
25424db0459SJohn Baldwin 			if (mdp2->md_ldt == NULL)
255df4d012bSJohn Baldwin 				panic("could not copy LDT");
25691c28bfdSLuoqi Chen 		}
2578c39a127SStefan Eßer 	}
258df4d012bSJohn Baldwin 	mtx_unlock_spin(&sched_lock);
2598c39a127SStefan Eßer 
260c6a37e84SJohn Baldwin 	/* Setup to release sched_lock in fork_exit(). */
261c6a37e84SJohn Baldwin 	td2->td_md.md_spinlock_count = 1;
262c6a37e84SJohn Baldwin 	td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
263c6a37e84SJohn Baldwin 
264a2a1c95cSPeter Wemm 	/*
265a2a1c95cSPeter Wemm 	 * Now, cpu_switch() can schedule the new process.
266a2a1c95cSPeter Wemm 	 * pcb_esp is loaded pointing to the cpu_switch() stack frame
267a2a1c95cSPeter Wemm 	 * containing the return address when exiting cpu_switch.
26837b087a6SPeter Wemm 	 * This will normally be to fork_trampoline(), which will have
26937b087a6SPeter Wemm 	 * %ebx loaded with the new proc's pointer.  fork_trampoline()
270a2a1c95cSPeter Wemm 	 * will set up a stack to call fork_return(p, frame); to complete
271a2a1c95cSPeter Wemm 	 * the return to user-mode.
272a2a1c95cSPeter Wemm 	 */
273a2a1c95cSPeter Wemm }
274a2a1c95cSPeter Wemm 
275a2a1c95cSPeter Wemm /*
276a2a1c95cSPeter Wemm  * Intercept the return address from a freshly forked process that has NOT
277a2a1c95cSPeter Wemm  * been scheduled yet.
278a2a1c95cSPeter Wemm  *
279a2a1c95cSPeter Wemm  * This is needed to make kernel threads stay in kernel mode.
280a2a1c95cSPeter Wemm  */
281a2a1c95cSPeter Wemm void
282b40ce416SJulian Elischer cpu_set_fork_handler(td, func, arg)
283b40ce416SJulian Elischer 	struct thread *td;
28415fe3067SAlfred Perlstein 	void (*func)(void *);
2859c8b8baaSPeter Wemm 	void *arg;
286a2a1c95cSPeter Wemm {
287a2a1c95cSPeter Wemm 	/*
288a2a1c95cSPeter Wemm 	 * Note that the trap frame follows the args, so the function
289a2a1c95cSPeter Wemm 	 * is really called like this:  func(arg, frame);
290a2a1c95cSPeter Wemm 	 */
291b40ce416SJulian Elischer 	td->td_pcb->pcb_esi = (int) func;	/* function */
292b40ce416SJulian Elischer 	td->td_pcb->pcb_ebx = (int) arg;	/* first arg */
2935b81b6b3SRodney W. Grimes }
2945b81b6b3SRodney W. Grimes 
2957c2b54e8SNate Williams void
296e602ba25SJulian Elischer cpu_exit(struct thread *td)
297e602ba25SJulian Elischer {
298e602ba25SJulian Elischer 	struct mdproc *mdp;
299e602ba25SJulian Elischer 
30066250360SDavid Xu 	/* Reset pc->pcb_gs and %gs before possibly invalidating it. */
301e602ba25SJulian Elischer 	mdp = &td->td_proc->p_md;
30266250360SDavid Xu 	if (mdp->md_ldt) {
30366250360SDavid Xu 		td->td_pcb->pcb_gs = _udatasel;
30466250360SDavid Xu 		load_gs(_udatasel);
305e602ba25SJulian Elischer 		user_ldt_free(td);
30666250360SDavid Xu 	}
307e602ba25SJulian Elischer }
308e602ba25SJulian Elischer 
309e602ba25SJulian Elischer void
310e602ba25SJulian Elischer cpu_thread_exit(struct thread *td)
3115b81b6b3SRodney W. Grimes {
312b40ce416SJulian Elischer 	struct pcb *pcb = td->td_pcb;
3134cc99cf6SJohn Baldwin 
314558226eaSPeter Wemm #ifdef DEV_NPX
315d5e1f581SDavid Xu 	if (td == PCPU_GET(fpcurthread))
316d5e1f581SDavid Xu 		npxdrop();
317558226eaSPeter Wemm #endif
318b19d9defSMaxime Henrion 	if (pcb->pcb_flags & PCB_DBREGS) {
3196745e78cSBrian S. Dean 		/* disable all hardware breakpoints */
320b19d9defSMaxime Henrion 		reset_dbregs();
321b19d9defSMaxime Henrion 		pcb->pcb_flags &= ~PCB_DBREGS;
322b19d9defSMaxime Henrion 	}
323b19d9defSMaxime Henrion }
324b19d9defSMaxime Henrion 
325b19d9defSMaxime Henrion void
326696058c3SJulian Elischer cpu_thread_clean(struct thread *td)
327b19d9defSMaxime Henrion {
328b19d9defSMaxime Henrion 	struct pcb *pcb;
329b19d9defSMaxime Henrion 
330b19d9defSMaxime Henrion 	pcb = td->td_pcb;
33148a09cf2SJohn Dyson 	if (pcb->pcb_ext != 0) {
332e602ba25SJulian Elischer 		/* XXXKSE  XXXSMP  not SMP SAFE.. what locks do we have? */
333e602ba25SJulian Elischer 		/* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */
33448a09cf2SJohn Dyson 		/*
33548a09cf2SJohn Dyson 		 * XXX do we need to move the TSS off the allocated pages
33648a09cf2SJohn Dyson 		 * before freeing them?  (not done here)
33748a09cf2SJohn Dyson 		 */
33848a09cf2SJohn Dyson 		kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
33948a09cf2SJohn Dyson 		    ctob(IOPAGES + 1));
34048a09cf2SJohn Dyson 		pcb->pcb_ext = 0;
34148a09cf2SJohn Dyson 	}
3425b81b6b3SRodney W. Grimes }
3435b81b6b3SRodney W. Grimes 
344381fe1aaSGarrett Wollman void
345710338e9SMarcel Moolenaar cpu_thread_swapin(struct thread *td)
346710338e9SMarcel Moolenaar {
347710338e9SMarcel Moolenaar }
348710338e9SMarcel Moolenaar 
349710338e9SMarcel Moolenaar void
350710338e9SMarcel Moolenaar cpu_thread_swapout(struct thread *td)
351710338e9SMarcel Moolenaar {
352710338e9SMarcel Moolenaar }
353710338e9SMarcel Moolenaar 
354710338e9SMarcel Moolenaar void
355e602ba25SJulian Elischer cpu_thread_setup(struct thread *td)
356e602ba25SJulian Elischer {
357e602ba25SJulian Elischer 
3586041f1a5SBill Paul 	td->td_pcb = (struct pcb *)(td->td_kstack +
3596041f1a5SBill Paul 	    td->td_kstack_pages * PAGE_SIZE) - 1;
360e602ba25SJulian Elischer 	td->td_frame = (struct trapframe *)((caddr_t)td->td_pcb - 16) - 1;
361c7251aedSTor Egge 	td->td_pcb->pcb_ext = NULL;
362e602ba25SJulian Elischer }
363e602ba25SJulian Elischer 
364e602ba25SJulian Elischer /*
365575525a0SJonathan Mini  * Initialize machine state (pcb and trap frame) for a new thread about to
36648bfcdddSJulian Elischer  * upcall. Put enough state in the new thread's PCB to get it to go back
36748bfcdddSJulian Elischer  * userret(), where we can intercept it again to set the return (upcall)
36848bfcdddSJulian Elischer  * Address and stack, along with those from upcals that are from other sources
36948bfcdddSJulian Elischer  * such as those generated in thread_userret() itself.
370e602ba25SJulian Elischer  */
371e602ba25SJulian Elischer void
37211e0f8e1SMarcel Moolenaar cpu_set_upcall(struct thread *td, struct thread *td0)
373e602ba25SJulian Elischer {
374e602ba25SJulian Elischer 	struct pcb *pcb2;
375e602ba25SJulian Elischer 
376e602ba25SJulian Elischer 	/* Point the pcb to the top of the stack. */
377e602ba25SJulian Elischer 	pcb2 = td->td_pcb;
378e602ba25SJulian Elischer 
379e602ba25SJulian Elischer 	/*
380e602ba25SJulian Elischer 	 * Copy the upcall pcb.  This loads kernel regs.
381e602ba25SJulian Elischer 	 * Those not loaded individually below get their default
382e602ba25SJulian Elischer 	 * values here.
383e602ba25SJulian Elischer 	 *
384e602ba25SJulian Elischer 	 * XXXKSE It might be a good idea to simply skip this as
385e602ba25SJulian Elischer 	 * the values of the other registers may be unimportant.
386e602ba25SJulian Elischer 	 * This would remove any requirement for knowing the KSE
387e602ba25SJulian Elischer 	 * at this time (see the matching comment below for
388e602ba25SJulian Elischer 	 * more analysis) (need a good safe default).
389e602ba25SJulian Elischer 	 */
39011e0f8e1SMarcel Moolenaar 	bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
391d5e1f581SDavid Xu 	pcb2->pcb_flags &= ~(PCB_NPXTRAP|PCB_NPXINITDONE);
392e602ba25SJulian Elischer 
393e602ba25SJulian Elischer 	/*
394e602ba25SJulian Elischer 	 * Create a new fresh stack for the new thread.
395e602ba25SJulian Elischer 	 * The -16 is so we can expand the trapframe if we go to vm86.
396e602ba25SJulian Elischer 	 * Don't forget to set this stack value into whatever supplies
397e602ba25SJulian Elischer 	 * the address for the fault handlers.
398e602ba25SJulian Elischer 	 * The contexts are filled in at the time we actually DO the
399e602ba25SJulian Elischer 	 * upcall as only then do we know which KSE we got.
400e602ba25SJulian Elischer 	 */
40111e0f8e1SMarcel Moolenaar 	bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
402e602ba25SJulian Elischer 
403e602ba25SJulian Elischer 	/*
404e602ba25SJulian Elischer 	 * Set registers for trampoline to user mode.  Leave space for the
405e602ba25SJulian Elischer 	 * return address on stack.  These are the kernel mode register values.
406e602ba25SJulian Elischer 	 */
4077ab9b220SJake Burkholder #ifdef PAE
4087ab9b220SJake Burkholder 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pdpt);
4097ab9b220SJake Burkholder #else
410e602ba25SJulian Elischer 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pdir);
4117ab9b220SJake Burkholder #endif
412e602ba25SJulian Elischer 	pcb2->pcb_edi = 0;
413e602ba25SJulian Elischer 	pcb2->pcb_esi = (int)fork_return;		    /* trampoline arg */
414e602ba25SJulian Elischer 	pcb2->pcb_ebp = 0;
415e602ba25SJulian Elischer 	pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */
416e602ba25SJulian Elischer 	pcb2->pcb_ebx = (int)td;			    /* trampoline arg */
417e602ba25SJulian Elischer 	pcb2->pcb_eip = (int)fork_trampoline;
418e602ba25SJulian Elischer 	pcb2->pcb_psl &= ~(PSL_I);	/* interrupts must be disabled */
4192257a44fSDavid Xu 	pcb2->pcb_gs = rgs();
420e602ba25SJulian Elischer 	/*
421e602ba25SJulian Elischer 	 * If we didn't copy the pcb, we'd need to do the following registers:
422e602ba25SJulian Elischer 	 * pcb2->pcb_dr*:	cloned above.
423e602ba25SJulian Elischer 	 * pcb2->pcb_savefpu:	cloned above.
424e602ba25SJulian Elischer 	 * pcb2->pcb_flags:	cloned above.
425e602ba25SJulian Elischer 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
426e602ba25SJulian Elischer 	 * pcb2->pcb_gs:	cloned above.  XXXKSE ???
427e602ba25SJulian Elischer 	 * pcb2->pcb_ext:	cleared below.
428e602ba25SJulian Elischer 	 */
429e602ba25SJulian Elischer 	pcb2->pcb_ext = NULL;
430c6a37e84SJohn Baldwin 
431c6a37e84SJohn Baldwin 	/* Setup to release sched_lock in fork_exit(). */
432c6a37e84SJohn Baldwin 	td->td_md.md_spinlock_count = 1;
433c6a37e84SJohn Baldwin 	td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
434e602ba25SJulian Elischer }
435e602ba25SJulian Elischer 
436e602ba25SJulian Elischer /*
43748bfcdddSJulian Elischer  * Set that machine state for performing an upcall that has to
43848bfcdddSJulian Elischer  * be done in thread_userret() so that those upcalls generated
43948bfcdddSJulian Elischer  * in thread_userret() itself can be done as well.
440e602ba25SJulian Elischer  */
441575525a0SJonathan Mini void
4425215b187SJeff Roberson cpu_set_upcall_kse(struct thread *td, struct kse_upcall *ku)
443575525a0SJonathan Mini {
444575525a0SJonathan Mini 
445575525a0SJonathan Mini 	/*
446696058c3SJulian Elischer 	 * Do any extra cleaning that needs to be done.
447696058c3SJulian Elischer 	 * The thread may have optional components
448696058c3SJulian Elischer 	 * that are not present in a fresh thread.
449696058c3SJulian Elischer 	 * This may be a recycled thread so make it look
450696058c3SJulian Elischer 	 * as though it's newly allocated.
451696058c3SJulian Elischer 	 */
452696058c3SJulian Elischer 	cpu_thread_clean(td);
453696058c3SJulian Elischer 
454696058c3SJulian Elischer 	/*
455575525a0SJonathan Mini 	 * Set the trap frame to point at the beginning of the uts
456575525a0SJonathan Mini 	 * function.
457575525a0SJonathan Mini 	 */
4582396628bSDavid Xu 	td->td_frame->tf_ebp = 0;
459575525a0SJonathan Mini 	td->td_frame->tf_esp =
4605215b187SJeff Roberson 	    (int)ku->ku_stack.ss_sp + ku->ku_stack.ss_size - 16;
4615215b187SJeff Roberson 	td->td_frame->tf_eip = (int)ku->ku_func;
462575525a0SJonathan Mini 
463575525a0SJonathan Mini 	/*
464575525a0SJonathan Mini 	 * Pass the address of the mailbox for this kse to the uts
465575525a0SJonathan Mini 	 * function as a parameter on the stack.
466575525a0SJonathan Mini 	 */
467575525a0SJonathan Mini 	suword((void *)(td->td_frame->tf_esp + sizeof(void *)),
4685215b187SJeff Roberson 	    (int)ku->ku_mailbox);
469e602ba25SJulian Elischer }
470e602ba25SJulian Elischer 
4715b81b6b3SRodney W. Grimes /*
4725b81b6b3SRodney W. Grimes  * Convert kernel VA to physical address
4735b81b6b3SRodney W. Grimes  */
474227f9a1cSJake Burkholder vm_paddr_t
4757f8cb368SDavid Greenman kvtop(void *addr)
4765b81b6b3SRodney W. Grimes {
477227f9a1cSJake Burkholder 	vm_paddr_t pa;
4785b81b6b3SRodney W. Grimes 
479227f9a1cSJake Burkholder 	pa = pmap_kextract((vm_offset_t)addr);
480227f9a1cSJake Burkholder 	if (pa == 0)
4815b81b6b3SRodney W. Grimes 		panic("kvtop: zero page frame");
482227f9a1cSJake Burkholder 	return (pa);
4835b81b6b3SRodney W. Grimes }
4845b81b6b3SRodney W. Grimes 
4854260b00aSNate Lawson #ifdef SMP
4864260b00aSNate Lawson static void
4874260b00aSNate Lawson cpu_reset_proxy()
4884260b00aSNate Lawson {
4894260b00aSNate Lawson 
4904260b00aSNate Lawson 	cpu_reset_proxy_active = 1;
4914260b00aSNate Lawson 	while (cpu_reset_proxy_active == 1)
4924260b00aSNate Lawson 		;	/* Wait for other cpu to see that we've started */
4934260b00aSNate Lawson 	stop_cpus((1<<cpu_reset_proxyid));
4944260b00aSNate Lawson 	printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
4954260b00aSNate Lawson 	DELAY(1000000);
4964260b00aSNate Lawson 	cpu_reset_real();
4974260b00aSNate Lawson }
4984260b00aSNate Lawson #endif
4994260b00aSNate Lawson 
5007f8cb368SDavid Greenman void
501d447dbeeSBruce Evans cpu_reset()
502d447dbeeSBruce Evans {
5032f1e7069STor Egge #ifdef SMP
5044260b00aSNate Lawson 	u_int cnt, map;
5052f1e7069STor Egge 
50663a6daf6SNate Lawson 	if (smp_active) {
507ef73ae4bSJake Burkholder 		map = PCPU_GET(other_cpus) & ~stopped_cpus;
5082f1e7069STor Egge 		if (map != 0) {
5092f1e7069STor Egge 			printf("cpu_reset: Stopping other CPUs\n");
51063a6daf6SNate Lawson 			stop_cpus(map);
5112f1e7069STor Egge 		}
5124260b00aSNate Lawson 
5134260b00aSNate Lawson 		if (PCPU_GET(cpuid) != 0) {
5144260b00aSNate Lawson 			cpu_reset_proxyid = PCPU_GET(cpuid);
5154260b00aSNate Lawson 			cpustop_restartfunc = cpu_reset_proxy;
5164260b00aSNate Lawson 			cpu_reset_proxy_active = 0;
5174260b00aSNate Lawson 			printf("cpu_reset: Restarting BSP\n");
5184260b00aSNate Lawson 			started_cpus = (1<<0);		/* Restart CPU #0 */
5194260b00aSNate Lawson 
5204260b00aSNate Lawson 			cnt = 0;
5214260b00aSNate Lawson 			while (cpu_reset_proxy_active == 0 && cnt < 10000000)
5224260b00aSNate Lawson 				cnt++;	/* Wait for BSP to announce restart */
5234260b00aSNate Lawson 			if (cpu_reset_proxy_active == 0)
5244260b00aSNate Lawson 				printf("cpu_reset: Failed to restart BSP\n");
5254260b00aSNate Lawson 			enable_intr();
5264260b00aSNate Lawson 			cpu_reset_proxy_active = 2;
5274260b00aSNate Lawson 
5284260b00aSNate Lawson 			while (1);
5294260b00aSNate Lawson 			/* NOTREACHED */
5304260b00aSNate Lawson 		}
5314260b00aSNate Lawson 
5322f1e7069STor Egge 		DELAY(1000000);
5332f1e7069STor Egge 	}
5342f1e7069STor Egge #endif
53563a6daf6SNate Lawson 	cpu_reset_real();
53663a6daf6SNate Lawson 	/* NOTREACHED */
5372f1e7069STor Egge }
5382f1e7069STor Egge 
5392f1e7069STor Egge static void
5402f1e7069STor Egge cpu_reset_real()
5412f1e7069STor Egge {
542d447dbeeSBruce Evans 
543883bd55aSPoul-Henning Kamp #ifdef CPU_ELAN
544883bd55aSPoul-Henning Kamp 	if (elan_mmcr != NULL)
545883bd55aSPoul-Henning Kamp 		elan_mmcr->RESCFG = 1;
546883bd55aSPoul-Henning Kamp #endif
547883bd55aSPoul-Henning Kamp 
54802d12d93SPoul-Henning Kamp 	if (cpu == CPU_GEODE1100) {
54902d12d93SPoul-Henning Kamp 		/* Attempt Geode's own reset */
55002d12d93SPoul-Henning Kamp 		outl(0xcf8, 0x80009044ul);
55102d12d93SPoul-Henning Kamp 		outl(0xcfc, 0xf);
55202d12d93SPoul-Henning Kamp 	}
55302d12d93SPoul-Henning Kamp 
55403245f09SKATO Takenori #ifdef PC98
55503245f09SKATO Takenori 	/*
55603245f09SKATO Takenori 	 * Attempt to do a CPU reset via CPU reset port.
55703245f09SKATO Takenori 	 */
558d447dbeeSBruce Evans 	disable_intr();
5598e929d65SKATO Takenori 	if ((inb(0x35) & 0xa0) != 0xa0) {
56003245f09SKATO Takenori 		outb(0x37, 0x0f);		/* SHUT0 = 0. */
56103245f09SKATO Takenori 		outb(0x37, 0x0b);		/* SHUT1 = 0. */
5628e929d65SKATO Takenori 	}
56303245f09SKATO Takenori 	outb(0xf0, 0x00);		/* Reset. */
56403245f09SKATO Takenori #else
56563a6daf6SNate Lawson #if !defined(BROKEN_KEYBOARD_RESET)
5662320728fSRodney W. Grimes 	/*
5672320728fSRodney W. Grimes 	 * Attempt to do a CPU reset via the keyboard controller,
56863a6daf6SNate Lawson 	 * do not turn off GateA20, as any machine that fails
5692320728fSRodney W. Grimes 	 * to do the reset here would then end up in no man's land.
5702320728fSRodney W. Grimes 	 */
5712320728fSRodney W. Grimes 	outb(IO_KBD + 4, 0xFE);
5722320728fSRodney W. Grimes 	DELAY(500000);	/* wait 0.5 sec to see if that did it */
5732320728fSRodney W. Grimes 	printf("Keyboard reset did not work, attempting CPU shutdown\n");
5742320728fSRodney W. Grimes 	DELAY(1000000);	/* wait 1 sec for printf to complete */
5755eb46edfSDavid Greenman #endif
57603245f09SKATO Takenori #endif /* PC98 */
57763a6daf6SNate Lawson 
57863a6daf6SNate Lawson 	/* Force a shutdown by unmapping entire address space. */
579910548deSJake Burkholder 	bzero((caddr_t)PTD, NBPTD);
5805b81b6b3SRodney W. Grimes 
5815b81b6b3SRodney W. Grimes 	/* "good night, sweet prince .... <THUNK!>" */
58227e9b35eSJohn Dyson 	invltlb();
5835b81b6b3SRodney W. Grimes 	/* NOTREACHED */
5847f8cb368SDavid Greenman 	while(1);
5855b81b6b3SRodney W. Grimes }
586b9d60b3fSDavid Greenman 
587e0b78e19SJoerg Wunsch /*
588411d10a6SAlan Cox  * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-))
589411d10a6SAlan Cox  */
590411d10a6SAlan Cox static void
591411d10a6SAlan Cox sf_buf_init(void *arg)
592411d10a6SAlan Cox {
593411d10a6SAlan Cox 	struct sf_buf *sf_bufs;
594411d10a6SAlan Cox 	vm_offset_t sf_base;
595411d10a6SAlan Cox 	int i;
596411d10a6SAlan Cox 
597099a0e58SBosko Milekic 	nsfbufs = NSFBUFS;
598099a0e58SBosko Milekic 	TUNABLE_INT_FETCH("kern.ipc.nsfbufs", &nsfbufs);
599099a0e58SBosko Milekic 
6000543fa53SAlan Cox 	sf_buf_active = hashinit(nsfbufs, M_TEMP, &sf_buf_hashmask);
601a5819cb5SAlan Cox 	TAILQ_INIT(&sf_buf_freelist);
602411d10a6SAlan Cox 	sf_base = kmem_alloc_nofault(kernel_map, nsfbufs * PAGE_SIZE);
603411d10a6SAlan Cox 	sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP,
604411d10a6SAlan Cox 	    M_NOWAIT | M_ZERO);
605411d10a6SAlan Cox 	for (i = 0; i < nsfbufs; i++) {
606411d10a6SAlan Cox 		sf_bufs[i].kva = sf_base + i * PAGE_SIZE;
607a5819cb5SAlan Cox 		TAILQ_INSERT_TAIL(&sf_buf_freelist, &sf_bufs[i], free_entry);
608411d10a6SAlan Cox 	}
609411d10a6SAlan Cox 	sf_buf_alloc_want = 0;
6100543fa53SAlan Cox 	mtx_init(&sf_buf_lock, "sf_buf", NULL, MTX_DEF);
611411d10a6SAlan Cox }
612411d10a6SAlan Cox 
613411d10a6SAlan Cox /*
614411d10a6SAlan Cox  * Get an sf_buf from the freelist. Will block if none are available.
615411d10a6SAlan Cox  */
616411d10a6SAlan Cox struct sf_buf *
617d3cb0d99SAlan Cox sf_buf_alloc(struct vm_page *m, int flags)
618411d10a6SAlan Cox {
61920351fafSAlan Cox 	pt_entry_t opte, *ptep;
6200543fa53SAlan Cox 	struct sf_head *hash_list;
621411d10a6SAlan Cox 	struct sf_buf *sf;
6225c0db7c7SAlan Cox #ifdef SMP
6235c0db7c7SAlan Cox 	cpumask_t cpumask, other_cpus;
6245c0db7c7SAlan Cox #endif
625411d10a6SAlan Cox 	int error;
626411d10a6SAlan Cox 
6275c0db7c7SAlan Cox 	KASSERT(curthread->td_pinned > 0 || (flags & SFB_CPUPRIVATE) == 0,
6285c0db7c7SAlan Cox 	    ("sf_buf_alloc(SFB_CPUPRIVATE): curthread not pinned"));
6290543fa53SAlan Cox 	hash_list = &sf_buf_active[SF_BUF_HASH(m)];
6300543fa53SAlan Cox 	mtx_lock(&sf_buf_lock);
6310543fa53SAlan Cox 	LIST_FOREACH(sf, hash_list, list_entry) {
6320543fa53SAlan Cox 		if (sf->m == m) {
6330543fa53SAlan Cox 			sf->ref_count++;
6345eda9873SMike Silbersack 			if (sf->ref_count == 1) {
6359a63fc0dSAlan Cox 				TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
6365eda9873SMike Silbersack 				nsfbufsused++;
6375caf2b00SMike Silbersack 				nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
6385eda9873SMike Silbersack 			}
6395c0db7c7SAlan Cox #ifdef SMP
6405c0db7c7SAlan Cox 			cpumask = PCPU_GET(cpumask);
6415c0db7c7SAlan Cox 			if ((sf->cpumask & cpumask) == 0) {
6425c0db7c7SAlan Cox 				sf->cpumask |= cpumask;
6435c0db7c7SAlan Cox 				invlpg(sf->kva);
6445c0db7c7SAlan Cox 			}
6455c0db7c7SAlan Cox 			if ((flags & SFB_CPUPRIVATE) == 0) {
6465c0db7c7SAlan Cox 				other_cpus = PCPU_GET(other_cpus) & ~sf->cpumask;
6475c0db7c7SAlan Cox 				if (other_cpus != 0) {
6485c0db7c7SAlan Cox 					sf->cpumask |= other_cpus;
6495c0db7c7SAlan Cox 					mtx_lock_spin(&smp_ipi_mtx);
6505c0db7c7SAlan Cox 					smp_masked_invlpg(other_cpus, sf->kva);
6515c0db7c7SAlan Cox 					mtx_unlock_spin(&smp_ipi_mtx);
6525c0db7c7SAlan Cox 				}
6535c0db7c7SAlan Cox 			}
6545c0db7c7SAlan Cox #endif
6550543fa53SAlan Cox 			goto done;
6560543fa53SAlan Cox 		}
6570543fa53SAlan Cox 	}
658a5819cb5SAlan Cox 	while ((sf = TAILQ_FIRST(&sf_buf_freelist)) == NULL) {
659d3cb0d99SAlan Cox 		if (flags & SFB_NOWAIT)
660d3cb0d99SAlan Cox 			goto done;
661411d10a6SAlan Cox 		sf_buf_alloc_want++;
662ddeb5b24SMike Silbersack 		mbstat.sf_allocwait++;
663d3cb0d99SAlan Cox 		error = msleep(&sf_buf_freelist, &sf_buf_lock,
664d3cb0d99SAlan Cox 		    (flags & SFB_CATCH) ? PCATCH | PVM : PVM, "sfbufa", 0);
665411d10a6SAlan Cox 		sf_buf_alloc_want--;
666411d10a6SAlan Cox 
667411d10a6SAlan Cox 		/*
668411d10a6SAlan Cox 		 * If we got a signal, don't risk going back to sleep.
669411d10a6SAlan Cox 		 */
670411d10a6SAlan Cox 		if (error)
6710543fa53SAlan Cox 			goto done;
672411d10a6SAlan Cox 	}
673a5819cb5SAlan Cox 	TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
674a5819cb5SAlan Cox 	if (sf->m != NULL)
6750543fa53SAlan Cox 		LIST_REMOVE(sf, list_entry);
6760543fa53SAlan Cox 	LIST_INSERT_HEAD(hash_list, sf, list_entry);
6770543fa53SAlan Cox 	sf->ref_count = 1;
678411d10a6SAlan Cox 	sf->m = m;
6795eda9873SMike Silbersack 	nsfbufsused++;
6805caf2b00SMike Silbersack 	nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
68120351fafSAlan Cox 
68220351fafSAlan Cox 	/*
68320351fafSAlan Cox 	 * Update the sf_buf's virtual-to-physical mapping, flushing the
68420351fafSAlan Cox 	 * virtual address from the TLB only if the PTE implies that the old
68520351fafSAlan Cox 	 * mapping has been used.  Since the reference count for the sf_buf's
68620351fafSAlan Cox 	 * old mapping was zero, that mapping is not currently in use.
68720351fafSAlan Cox 	 * Consequently, there is no need to exchange the old and new PTEs
68820351fafSAlan Cox 	 * atomically, even under PAE.
68920351fafSAlan Cox 	 */
69020351fafSAlan Cox 	ptep = vtopte(sf->kva);
69120351fafSAlan Cox 	opte = *ptep;
69220351fafSAlan Cox 	*ptep = VM_PAGE_TO_PHYS(m) | pgeflag | PG_RW | PG_V;
6935c0db7c7SAlan Cox #ifdef SMP
6945c0db7c7SAlan Cox 	if (flags & SFB_CPUPRIVATE) {
6955c0db7c7SAlan Cox 		if ((opte & (PG_A | PG_V)) == (PG_A | PG_V)) {
6965c0db7c7SAlan Cox 			sf->cpumask = PCPU_GET(cpumask);
6975c0db7c7SAlan Cox 			invlpg(sf->kva);
6985c0db7c7SAlan Cox 		} else
6995c0db7c7SAlan Cox 			sf->cpumask = all_cpus;
7005c0db7c7SAlan Cox 		goto done;
7015c0db7c7SAlan Cox 	} else
7025c0db7c7SAlan Cox 		sf->cpumask = all_cpus;
7035c0db7c7SAlan Cox #endif
70420351fafSAlan Cox 	if ((opte & (PG_A | PG_V)) == (PG_A | PG_V))
70520351fafSAlan Cox 		pmap_invalidate_page(kernel_pmap, sf->kva);
7060543fa53SAlan Cox done:
7070543fa53SAlan Cox 	mtx_unlock(&sf_buf_lock);
708411d10a6SAlan Cox 	return (sf);
709411d10a6SAlan Cox }
710411d10a6SAlan Cox 
711411d10a6SAlan Cox /*
71290ecfebdSAlan Cox  * Remove a reference from the given sf_buf, adding it to the free
71390ecfebdSAlan Cox  * list when its reference count reaches zero.  A freed sf_buf still,
71490ecfebdSAlan Cox  * however, retains its virtual-to-physical mapping until it is
71590ecfebdSAlan Cox  * recycled or reactivated by sf_buf_alloc(9).
716411d10a6SAlan Cox  */
717411d10a6SAlan Cox void
71890ecfebdSAlan Cox sf_buf_free(struct sf_buf *sf)
719411d10a6SAlan Cox {
720411d10a6SAlan Cox 
7210543fa53SAlan Cox 	mtx_lock(&sf_buf_lock);
7220543fa53SAlan Cox 	sf->ref_count--;
7230543fa53SAlan Cox 	if (sf->ref_count == 0) {
724a5819cb5SAlan Cox 		TAILQ_INSERT_TAIL(&sf_buf_freelist, sf, free_entry);
72590ecfebdSAlan Cox 		nsfbufsused--;
7260543fa53SAlan Cox 		if (sf_buf_alloc_want > 0)
7270543fa53SAlan Cox 			wakeup_one(&sf_buf_freelist);
7280543fa53SAlan Cox 	}
7290543fa53SAlan Cox 	mtx_unlock(&sf_buf_lock);
730411d10a6SAlan Cox }
731411d10a6SAlan Cox 
732411d10a6SAlan Cox /*
73357d7d7b3SJustin T. Gibbs  * Software interrupt handler for queued VM system processing.
73457d7d7b3SJustin T. Gibbs  */
73557d7d7b3SJustin T. Gibbs void
7368088699fSJohn Baldwin swi_vm(void *dummy)
73757d7d7b3SJustin T. Gibbs {
73857d7d7b3SJustin T. Gibbs 	if (busdma_swi_pending != 0)
73957d7d7b3SJustin T. Gibbs 		busdma_swi();
74057d7d7b3SJustin T. Gibbs }
74157d7d7b3SJustin T. Gibbs 
74257d7d7b3SJustin T. Gibbs /*
743cae6f73aSJoerg Wunsch  * Tell whether this address is in some physical memory region.
744e0b78e19SJoerg Wunsch  * Currently used by the kernel coredump code in order to avoid
745e0b78e19SJoerg Wunsch  * dumping the ``ISA memory hole'' which could cause indefinite hangs,
746e0b78e19SJoerg Wunsch  * or other unpredictable behaviour.
747e0b78e19SJoerg Wunsch  */
748e0b78e19SJoerg Wunsch 
749e0b78e19SJoerg Wunsch int
7502c38d78eSAlan Cox is_physical_memory(vm_paddr_t addr)
751e0b78e19SJoerg Wunsch {
752e0b78e19SJoerg Wunsch 
75303927d3cSPeter Wemm #ifdef DEV_ISA
754e0b78e19SJoerg Wunsch 	/* The ISA ``memory hole''. */
755e0b78e19SJoerg Wunsch 	if (addr >= 0xa0000 && addr < 0x100000)
756cae6f73aSJoerg Wunsch 		return 0;
757e0b78e19SJoerg Wunsch #endif
758e0b78e19SJoerg Wunsch 
759e0b78e19SJoerg Wunsch 	/*
760e0b78e19SJoerg Wunsch 	 * stuff other tests for known memory-mapped devices (PCI?)
761e0b78e19SJoerg Wunsch 	 * here
762e0b78e19SJoerg Wunsch 	 */
763e0b78e19SJoerg Wunsch 
764cae6f73aSJoerg Wunsch 	return 1;
765e0b78e19SJoerg Wunsch }
766