xref: /freebsd/sys/i386/i386/vm_machdep.c (revision 0a10f22a30d61a6f32777a236a82d461129538cc)
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 XEN
93 #include <xen/hypervisor.h>
94 #endif
95 #ifdef PC98
96 #include <pc98/cbus/cbus.h>
97 #else
98 #include <x86/isa/isa.h>
99 #endif
100 
101 #ifdef XBOX
102 #include <machine/xbox.h>
103 #endif
104 
105 #ifndef NSFBUFS
106 #define	NSFBUFS		(512 + maxusers * 16)
107 #endif
108 
109 _Static_assert(OFFSETOF_CURTHREAD == offsetof(struct pcpu, pc_curthread),
110     "OFFSETOF_CURTHREAD does not correspond with offset of pc_curthread.");
111 _Static_assert(OFFSETOF_CURPCB == offsetof(struct pcpu, pc_curpcb),
112     "OFFSETOF_CURPCB does not correspond with offset of pc_curpcb.");
113 
114 static void	cpu_reset_real(void);
115 #ifdef SMP
116 static void	cpu_reset_proxy(void);
117 static u_int	cpu_reset_proxyid;
118 static volatile u_int	cpu_reset_proxy_active;
119 #endif
120 static void	sf_buf_init(void *arg);
121 SYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL);
122 
123 LIST_HEAD(sf_head, sf_buf);
124 
125 /*
126  * A hash table of active sendfile(2) buffers
127  */
128 static struct sf_head *sf_buf_active;
129 static u_long sf_buf_hashmask;
130 
131 #define	SF_BUF_HASH(m)	(((m) - vm_page_array) & sf_buf_hashmask)
132 
133 static TAILQ_HEAD(, sf_buf) sf_buf_freelist;
134 static u_int	sf_buf_alloc_want;
135 
136 /*
137  * A lock used to synchronize access to the hash table and free list
138  */
139 static struct mtx sf_buf_lock;
140 
141 extern int	_ucodesel, _udatasel;
142 
143 /*
144  * Finish a fork operation, with process p2 nearly set up.
145  * Copy and update the pcb, set up the stack so that the child
146  * ready to run and return to user mode.
147  */
148 void
149 cpu_fork(td1, p2, td2, flags)
150 	register struct thread *td1;
151 	register struct proc *p2;
152 	struct thread *td2;
153 	int flags;
154 {
155 	register struct proc *p1;
156 	struct pcb *pcb2;
157 	struct mdproc *mdp2;
158 
159 	p1 = td1->td_proc;
160 	if ((flags & RFPROC) == 0) {
161 		if ((flags & RFMEM) == 0) {
162 			/* unshare user LDT */
163 			struct mdproc *mdp1 = &p1->p_md;
164 			struct proc_ldt *pldt, *pldt1;
165 
166 			mtx_lock_spin(&dt_lock);
167 			if ((pldt1 = mdp1->md_ldt) != NULL &&
168 			    pldt1->ldt_refcnt > 1) {
169 				pldt = user_ldt_alloc(mdp1, pldt1->ldt_len);
170 				if (pldt == NULL)
171 					panic("could not copy LDT");
172 				mdp1->md_ldt = pldt;
173 				set_user_ldt(mdp1);
174 				user_ldt_deref(pldt1);
175 			} else
176 				mtx_unlock_spin(&dt_lock);
177 		}
178 		return;
179 	}
180 
181 	/* Ensure that td1's pcb is up to date. */
182 	if (td1 == curthread)
183 		td1->td_pcb->pcb_gs = rgs();
184 #ifdef DEV_NPX
185 	critical_enter();
186 	if (PCPU_GET(fpcurthread) == td1)
187 		npxsave(td1->td_pcb->pcb_save);
188 	critical_exit();
189 #endif
190 
191 	/* Point the pcb to the top of the stack */
192 	pcb2 = (struct pcb *)(td2->td_kstack +
193 	    td2->td_kstack_pages * PAGE_SIZE) - 1;
194 	td2->td_pcb = pcb2;
195 
196 	/* Copy td1's pcb */
197 	bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
198 
199 	/* Properly initialize pcb_save */
200 	pcb2->pcb_save = &pcb2->pcb_user_save;
201 
202 	/* Point mdproc and then copy over td1's contents */
203 	mdp2 = &p2->p_md;
204 	bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
205 
206 	/*
207 	 * Create a new fresh stack for the new process.
208 	 * Copy the trap frame for the return to user mode as if from a
209 	 * syscall.  This copies most of the user mode register values.
210 	 * The -16 is so we can expand the trapframe if we go to vm86.
211 	 */
212 	td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1;
213 	bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
214 
215 	td2->td_frame->tf_eax = 0;		/* Child returns zero */
216 	td2->td_frame->tf_eflags &= ~PSL_C;	/* success */
217 	td2->td_frame->tf_edx = 1;
218 
219 	/*
220 	 * If the parent process has the trap bit set (i.e. a debugger had
221 	 * single stepped the process to the system call), we need to clear
222 	 * the trap flag from the new frame unless the debugger had set PF_FORK
223 	 * on the parent.  Otherwise, the child will receive a (likely
224 	 * unexpected) SIGTRAP when it executes the first instruction after
225 	 * returning  to userland.
226 	 */
227 	if ((p1->p_pfsflags & PF_FORK) == 0)
228 		td2->td_frame->tf_eflags &= ~PSL_T;
229 
230 	/*
231 	 * Set registers for trampoline to user mode.  Leave space for the
232 	 * return address on stack.  These are the kernel mode register values.
233 	 */
234 #ifdef PAE
235 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdpt);
236 #else
237 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
238 #endif
239 	pcb2->pcb_edi = 0;
240 	pcb2->pcb_esi = (int)fork_return;	/* fork_trampoline argument */
241 	pcb2->pcb_ebp = 0;
242 	pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
243 	pcb2->pcb_ebx = (int)td2;		/* fork_trampoline argument */
244 	pcb2->pcb_eip = (int)fork_trampoline;
245 	pcb2->pcb_psl = PSL_KERNEL;		/* ints disabled */
246 	/*-
247 	 * pcb2->pcb_dr*:	cloned above.
248 	 * pcb2->pcb_savefpu:	cloned above.
249 	 * pcb2->pcb_flags:	cloned above.
250 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
251 	 * pcb2->pcb_gs:	cloned above.
252 	 * pcb2->pcb_ext:	cleared below.
253 	 */
254 
255 	/*
256 	 * XXX don't copy the i/o pages.  this should probably be fixed.
257 	 */
258 	pcb2->pcb_ext = 0;
259 
260 	/* Copy the LDT, if necessary. */
261 	mtx_lock_spin(&dt_lock);
262 	if (mdp2->md_ldt != NULL) {
263 		if (flags & RFMEM) {
264 			mdp2->md_ldt->ldt_refcnt++;
265 		} else {
266 			mdp2->md_ldt = user_ldt_alloc(mdp2,
267 			    mdp2->md_ldt->ldt_len);
268 			if (mdp2->md_ldt == NULL)
269 				panic("could not copy LDT");
270 		}
271 	}
272 	mtx_unlock_spin(&dt_lock);
273 
274 	/* Setup to release spin count in fork_exit(). */
275 	td2->td_md.md_spinlock_count = 1;
276 	/*
277 	 * XXX XEN need to check on PSL_USER is handled
278 	 */
279 	td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
280 	/*
281 	 * Now, cpu_switch() can schedule the new process.
282 	 * pcb_esp is loaded pointing to the cpu_switch() stack frame
283 	 * containing the return address when exiting cpu_switch.
284 	 * This will normally be to fork_trampoline(), which will have
285 	 * %ebx loaded with the new proc's pointer.  fork_trampoline()
286 	 * will set up a stack to call fork_return(p, frame); to complete
287 	 * the return to user-mode.
288 	 */
289 }
290 
291 /*
292  * Intercept the return address from a freshly forked process that has NOT
293  * been scheduled yet.
294  *
295  * This is needed to make kernel threads stay in kernel mode.
296  */
297 void
298 cpu_set_fork_handler(td, func, arg)
299 	struct thread *td;
300 	void (*func)(void *);
301 	void *arg;
302 {
303 	/*
304 	 * Note that the trap frame follows the args, so the function
305 	 * is really called like this:  func(arg, frame);
306 	 */
307 	td->td_pcb->pcb_esi = (int) func;	/* function */
308 	td->td_pcb->pcb_ebx = (int) arg;	/* first arg */
309 }
310 
311 void
312 cpu_exit(struct thread *td)
313 {
314 
315 	/*
316 	 * If this process has a custom LDT, release it.  Reset pc->pcb_gs
317 	 * and %gs before we free it in case they refer to an LDT entry.
318 	 */
319 	mtx_lock_spin(&dt_lock);
320 	if (td->td_proc->p_md.md_ldt) {
321 		td->td_pcb->pcb_gs = _udatasel;
322 		load_gs(_udatasel);
323 		user_ldt_free(td);
324 	} else
325 		mtx_unlock_spin(&dt_lock);
326 }
327 
328 void
329 cpu_thread_exit(struct thread *td)
330 {
331 
332 #ifdef DEV_NPX
333 	critical_enter();
334 	if (td == PCPU_GET(fpcurthread))
335 		npxdrop();
336 	critical_exit();
337 #endif
338 
339 	/* Disable any hardware breakpoints. */
340 	if (td->td_pcb->pcb_flags & PCB_DBREGS) {
341 		reset_dbregs();
342 		td->td_pcb->pcb_flags &= ~PCB_DBREGS;
343 	}
344 }
345 
346 void
347 cpu_thread_clean(struct thread *td)
348 {
349 	struct pcb *pcb;
350 
351 	pcb = td->td_pcb;
352 	if (pcb->pcb_ext != NULL) {
353 		/* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */
354 		/*
355 		 * XXX do we need to move the TSS off the allocated pages
356 		 * before freeing them?  (not done here)
357 		 */
358 		kva_free((vm_offset_t)pcb->pcb_ext,
359 		    ctob(IOPAGES + 1));
360 		pcb->pcb_ext = NULL;
361 	}
362 }
363 
364 void
365 cpu_thread_swapin(struct thread *td)
366 {
367 }
368 
369 void
370 cpu_thread_swapout(struct thread *td)
371 {
372 }
373 
374 void
375 cpu_thread_alloc(struct thread *td)
376 {
377 
378 	td->td_pcb = (struct pcb *)(td->td_kstack +
379 	    td->td_kstack_pages * PAGE_SIZE) - 1;
380 	td->td_frame = (struct trapframe *)((caddr_t)td->td_pcb - 16) - 1;
381 	td->td_pcb->pcb_ext = NULL;
382 	td->td_pcb->pcb_save = &td->td_pcb->pcb_user_save;
383 }
384 
385 void
386 cpu_thread_free(struct thread *td)
387 {
388 
389 	cpu_thread_clean(td);
390 }
391 
392 void
393 cpu_set_syscall_retval(struct thread *td, int error)
394 {
395 
396 	switch (error) {
397 	case 0:
398 		td->td_frame->tf_eax = td->td_retval[0];
399 		td->td_frame->tf_edx = td->td_retval[1];
400 		td->td_frame->tf_eflags &= ~PSL_C;
401 		break;
402 
403 	case ERESTART:
404 		/*
405 		 * Reconstruct pc, assuming lcall $X,y is 7 bytes, int
406 		 * 0x80 is 2 bytes. We saved this in tf_err.
407 		 */
408 		td->td_frame->tf_eip -= td->td_frame->tf_err;
409 		break;
410 
411 	case EJUSTRETURN:
412 		break;
413 
414 	default:
415 		if (td->td_proc->p_sysent->sv_errsize) {
416 			if (error >= td->td_proc->p_sysent->sv_errsize)
417 				error = -1;	/* XXX */
418 			else
419 				error = td->td_proc->p_sysent->sv_errtbl[error];
420 		}
421 		td->td_frame->tf_eax = error;
422 		td->td_frame->tf_eflags |= PSL_C;
423 		break;
424 	}
425 }
426 
427 /*
428  * Initialize machine state (pcb and trap frame) for a new thread about to
429  * upcall. Put enough state in the new thread's PCB to get it to go back
430  * userret(), where we can intercept it again to set the return (upcall)
431  * Address and stack, along with those from upcals that are from other sources
432  * such as those generated in thread_userret() itself.
433  */
434 void
435 cpu_set_upcall(struct thread *td, struct thread *td0)
436 {
437 	struct pcb *pcb2;
438 
439 	/* Point the pcb to the top of the stack. */
440 	pcb2 = td->td_pcb;
441 
442 	/*
443 	 * Copy the upcall pcb.  This loads kernel regs.
444 	 * Those not loaded individually below get their default
445 	 * values here.
446 	 */
447 	bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
448 	pcb2->pcb_flags &= ~(PCB_NPXINITDONE | PCB_NPXUSERINITDONE);
449 	pcb2->pcb_save = &pcb2->pcb_user_save;
450 
451 	/*
452 	 * Create a new fresh stack for the new thread.
453 	 */
454 	bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
455 
456 	/* If the current thread has the trap bit set (i.e. a debugger had
457 	 * single stepped the process to the system call), we need to clear
458 	 * the trap flag from the new frame. Otherwise, the new thread will
459 	 * receive a (likely unexpected) SIGTRAP when it executes the first
460 	 * instruction after returning to userland.
461 	 */
462 	td->td_frame->tf_eflags &= ~PSL_T;
463 
464 	/*
465 	 * Set registers for trampoline to user mode.  Leave space for the
466 	 * return address on stack.  These are the kernel mode register values.
467 	 */
468 	pcb2->pcb_edi = 0;
469 	pcb2->pcb_esi = (int)fork_return;		    /* trampoline arg */
470 	pcb2->pcb_ebp = 0;
471 	pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */
472 	pcb2->pcb_ebx = (int)td;			    /* trampoline arg */
473 	pcb2->pcb_eip = (int)fork_trampoline;
474 	pcb2->pcb_psl &= ~(PSL_I);	/* interrupts must be disabled */
475 	pcb2->pcb_gs = rgs();
476 	/*
477 	 * If we didn't copy the pcb, we'd need to do the following registers:
478 	 * pcb2->pcb_cr3:	cloned above.
479 	 * pcb2->pcb_dr*:	cloned above.
480 	 * pcb2->pcb_savefpu:	cloned above.
481 	 * pcb2->pcb_flags:	cloned above.
482 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
483 	 * pcb2->pcb_gs:	cloned above.
484 	 * pcb2->pcb_ext:	cleared below.
485 	 */
486 	pcb2->pcb_ext = NULL;
487 
488 	/* Setup to release spin count in fork_exit(). */
489 	td->td_md.md_spinlock_count = 1;
490 	td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
491 }
492 
493 /*
494  * Set that machine state for performing an upcall that has to
495  * be done in thread_userret() so that those upcalls generated
496  * in thread_userret() itself can be done as well.
497  */
498 void
499 cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg,
500 	stack_t *stack)
501 {
502 
503 	/*
504 	 * Do any extra cleaning that needs to be done.
505 	 * The thread may have optional components
506 	 * that are not present in a fresh thread.
507 	 * This may be a recycled thread so make it look
508 	 * as though it's newly allocated.
509 	 */
510 	cpu_thread_clean(td);
511 
512 	/*
513 	 * Set the trap frame to point at the beginning of the uts
514 	 * function.
515 	 */
516 	td->td_frame->tf_ebp = 0;
517 	td->td_frame->tf_esp =
518 	    (((int)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
519 	td->td_frame->tf_eip = (int)entry;
520 
521 	/*
522 	 * Pass the address of the mailbox for this kse to the uts
523 	 * function as a parameter on the stack.
524 	 */
525 	suword((void *)(td->td_frame->tf_esp + sizeof(void *)),
526 	    (int)arg);
527 }
528 
529 int
530 cpu_set_user_tls(struct thread *td, void *tls_base)
531 {
532 	struct segment_descriptor sd;
533 	uint32_t base;
534 
535 	/*
536 	 * Construct a descriptor and store it in the pcb for
537 	 * the next context switch.  Also store it in the gdt
538 	 * so that the load of tf_fs into %fs will activate it
539 	 * at return to userland.
540 	 */
541 	base = (uint32_t)tls_base;
542 	sd.sd_lobase = base & 0xffffff;
543 	sd.sd_hibase = (base >> 24) & 0xff;
544 	sd.sd_lolimit = 0xffff;	/* 4GB limit, wraps around */
545 	sd.sd_hilimit = 0xf;
546 	sd.sd_type  = SDT_MEMRWA;
547 	sd.sd_dpl   = SEL_UPL;
548 	sd.sd_p     = 1;
549 	sd.sd_xx    = 0;
550 	sd.sd_def32 = 1;
551 	sd.sd_gran  = 1;
552 	critical_enter();
553 	/* set %gs */
554 	td->td_pcb->pcb_gsd = sd;
555 	if (td == curthread) {
556 		PCPU_GET(fsgs_gdt)[1] = sd;
557 		load_gs(GSEL(GUGS_SEL, SEL_UPL));
558 	}
559 	critical_exit();
560 	return (0);
561 }
562 
563 /*
564  * Convert kernel VA to physical address
565  */
566 vm_paddr_t
567 kvtop(void *addr)
568 {
569 	vm_paddr_t pa;
570 
571 	pa = pmap_kextract((vm_offset_t)addr);
572 	if (pa == 0)
573 		panic("kvtop: zero page frame");
574 	return (pa);
575 }
576 
577 #ifdef SMP
578 static void
579 cpu_reset_proxy()
580 {
581 	cpuset_t tcrp;
582 
583 	cpu_reset_proxy_active = 1;
584 	while (cpu_reset_proxy_active == 1)
585 		;	/* Wait for other cpu to see that we've started */
586 	CPU_SETOF(cpu_reset_proxyid, &tcrp);
587 	stop_cpus(tcrp);
588 	printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
589 	DELAY(1000000);
590 	cpu_reset_real();
591 }
592 #endif
593 
594 void
595 cpu_reset()
596 {
597 #ifdef XBOX
598 	if (arch_i386_is_xbox) {
599 		/* Kick the PIC16L, it can reboot the box */
600 		pic16l_reboot();
601 		for (;;);
602 	}
603 #endif
604 
605 #ifdef SMP
606 	cpuset_t map;
607 	u_int cnt;
608 
609 	if (smp_active) {
610 		map = all_cpus;
611 		CPU_CLR(PCPU_GET(cpuid), &map);
612 		CPU_NAND(&map, &stopped_cpus);
613 		if (!CPU_EMPTY(&map)) {
614 			printf("cpu_reset: Stopping other CPUs\n");
615 			stop_cpus(map);
616 		}
617 
618 		if (PCPU_GET(cpuid) != 0) {
619 			cpu_reset_proxyid = PCPU_GET(cpuid);
620 			cpustop_restartfunc = cpu_reset_proxy;
621 			cpu_reset_proxy_active = 0;
622 			printf("cpu_reset: Restarting BSP\n");
623 
624 			/* Restart CPU #0. */
625 			/* XXX: restart_cpus(1 << 0); */
626 			CPU_SETOF(0, &started_cpus);
627 			wmb();
628 
629 			cnt = 0;
630 			while (cpu_reset_proxy_active == 0 && cnt < 10000000)
631 				cnt++;	/* Wait for BSP to announce restart */
632 			if (cpu_reset_proxy_active == 0)
633 				printf("cpu_reset: Failed to restart BSP\n");
634 			enable_intr();
635 			cpu_reset_proxy_active = 2;
636 
637 			while (1);
638 			/* NOTREACHED */
639 		}
640 
641 		DELAY(1000000);
642 	}
643 #endif
644 	cpu_reset_real();
645 	/* NOTREACHED */
646 }
647 
648 static void
649 cpu_reset_real()
650 {
651 	struct region_descriptor null_idt;
652 #ifndef PC98
653 	int b;
654 #endif
655 
656 	disable_intr();
657 #ifdef XEN
658 	if (smp_processor_id() == 0)
659 		HYPERVISOR_shutdown(SHUTDOWN_reboot);
660 	else
661 		HYPERVISOR_shutdown(SHUTDOWN_poweroff);
662 #endif
663 #ifdef CPU_ELAN
664 	if (elan_mmcr != NULL)
665 		elan_mmcr->RESCFG = 1;
666 #endif
667 
668 	if (cpu == CPU_GEODE1100) {
669 		/* Attempt Geode's own reset */
670 		outl(0xcf8, 0x80009044ul);
671 		outl(0xcfc, 0xf);
672 	}
673 
674 #ifdef PC98
675 	/*
676 	 * Attempt to do a CPU reset via CPU reset port.
677 	 */
678 	if ((inb(0x35) & 0xa0) != 0xa0) {
679 		outb(0x37, 0x0f);		/* SHUT0 = 0. */
680 		outb(0x37, 0x0b);		/* SHUT1 = 0. */
681 	}
682 	outb(0xf0, 0x00);		/* Reset. */
683 #else
684 #if !defined(BROKEN_KEYBOARD_RESET)
685 	/*
686 	 * Attempt to do a CPU reset via the keyboard controller,
687 	 * do not turn off GateA20, as any machine that fails
688 	 * to do the reset here would then end up in no man's land.
689 	 */
690 	outb(IO_KBD + 4, 0xFE);
691 	DELAY(500000);	/* wait 0.5 sec to see if that did it */
692 #endif
693 
694 	/*
695 	 * Attempt to force a reset via the Reset Control register at
696 	 * I/O port 0xcf9.  Bit 2 forces a system reset when it
697 	 * transitions from 0 to 1.  Bit 1 selects the type of reset
698 	 * to attempt: 0 selects a "soft" reset, and 1 selects a
699 	 * "hard" reset.  We try a "hard" reset.  The first write sets
700 	 * bit 1 to select a "hard" reset and clears bit 2.  The
701 	 * second write forces a 0 -> 1 transition in bit 2 to trigger
702 	 * a reset.
703 	 */
704 	outb(0xcf9, 0x2);
705 	outb(0xcf9, 0x6);
706 	DELAY(500000);  /* wait 0.5 sec to see if that did it */
707 
708 	/*
709 	 * Attempt to force a reset via the Fast A20 and Init register
710 	 * at I/O port 0x92.  Bit 1 serves as an alternate A20 gate.
711 	 * Bit 0 asserts INIT# when set to 1.  We are careful to only
712 	 * preserve bit 1 while setting bit 0.  We also must clear bit
713 	 * 0 before setting it if it isn't already clear.
714 	 */
715 	b = inb(0x92);
716 	if (b != 0xff) {
717 		if ((b & 0x1) != 0)
718 			outb(0x92, b & 0xfe);
719 		outb(0x92, b | 0x1);
720 		DELAY(500000);  /* wait 0.5 sec to see if that did it */
721 	}
722 #endif /* PC98 */
723 
724 	printf("No known reset method worked, attempting CPU shutdown\n");
725 	DELAY(1000000); /* wait 1 sec for printf to complete */
726 
727 	/* Wipe the IDT. */
728 	null_idt.rd_limit = 0;
729 	null_idt.rd_base = 0;
730 	lidt(&null_idt);
731 
732 	/* "good night, sweet prince .... <THUNK!>" */
733 	breakpoint();
734 
735 	/* NOTREACHED */
736 	while(1);
737 }
738 
739 /*
740  * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-))
741  */
742 static void
743 sf_buf_init(void *arg)
744 {
745 	struct sf_buf *sf_bufs;
746 	vm_offset_t sf_base;
747 	int i;
748 
749 	nsfbufs = NSFBUFS;
750 	TUNABLE_INT_FETCH("kern.ipc.nsfbufs", &nsfbufs);
751 
752 	sf_buf_active = hashinit(nsfbufs, M_TEMP, &sf_buf_hashmask);
753 	TAILQ_INIT(&sf_buf_freelist);
754 	sf_base = kva_alloc(nsfbufs * PAGE_SIZE);
755 	sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP,
756 	    M_NOWAIT | M_ZERO);
757 	for (i = 0; i < nsfbufs; i++) {
758 		sf_bufs[i].kva = sf_base + i * PAGE_SIZE;
759 		TAILQ_INSERT_TAIL(&sf_buf_freelist, &sf_bufs[i], free_entry);
760 	}
761 	sf_buf_alloc_want = 0;
762 	mtx_init(&sf_buf_lock, "sf_buf", NULL, MTX_DEF);
763 }
764 
765 /*
766  * Invalidate the cache lines that may belong to the page, if
767  * (possibly old) mapping of the page by sf buffer exists.  Returns
768  * TRUE when mapping was found and cache invalidated.
769  */
770 boolean_t
771 sf_buf_invalidate_cache(vm_page_t m)
772 {
773 	struct sf_head *hash_list;
774 	struct sf_buf *sf;
775 	boolean_t ret;
776 
777 	hash_list = &sf_buf_active[SF_BUF_HASH(m)];
778 	ret = FALSE;
779 	mtx_lock(&sf_buf_lock);
780 	LIST_FOREACH(sf, hash_list, list_entry) {
781 		if (sf->m == m) {
782 			/*
783 			 * Use pmap_qenter to update the pte for
784 			 * existing mapping, in particular, the PAT
785 			 * settings are recalculated.
786 			 */
787 			pmap_qenter(sf->kva, &m, 1);
788 			pmap_invalidate_cache_range(sf->kva, sf->kva +
789 			    PAGE_SIZE);
790 			ret = TRUE;
791 			break;
792 		}
793 	}
794 	mtx_unlock(&sf_buf_lock);
795 	return (ret);
796 }
797 
798 /*
799  * Get an sf_buf from the freelist.  May block if none are available.
800  */
801 struct sf_buf *
802 sf_buf_alloc(struct vm_page *m, int flags)
803 {
804 	pt_entry_t opte, *ptep;
805 	struct sf_head *hash_list;
806 	struct sf_buf *sf;
807 #ifdef SMP
808 	cpuset_t other_cpus;
809 	u_int cpuid;
810 #endif
811 	int error;
812 
813 	KASSERT(curthread->td_pinned > 0 || (flags & SFB_CPUPRIVATE) == 0,
814 	    ("sf_buf_alloc(SFB_CPUPRIVATE): curthread not pinned"));
815 	hash_list = &sf_buf_active[SF_BUF_HASH(m)];
816 	mtx_lock(&sf_buf_lock);
817 	LIST_FOREACH(sf, hash_list, list_entry) {
818 		if (sf->m == m) {
819 			sf->ref_count++;
820 			if (sf->ref_count == 1) {
821 				TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
822 				nsfbufsused++;
823 				nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
824 			}
825 #ifdef SMP
826 			goto shootdown;
827 #else
828 			goto done;
829 #endif
830 		}
831 	}
832 	while ((sf = TAILQ_FIRST(&sf_buf_freelist)) == NULL) {
833 		if (flags & SFB_NOWAIT)
834 			goto done;
835 		sf_buf_alloc_want++;
836 		SFSTAT_INC(sf_allocwait);
837 		error = msleep(&sf_buf_freelist, &sf_buf_lock,
838 		    (flags & SFB_CATCH) ? PCATCH | PVM : PVM, "sfbufa", 0);
839 		sf_buf_alloc_want--;
840 
841 		/*
842 		 * If we got a signal, don't risk going back to sleep.
843 		 */
844 		if (error)
845 			goto done;
846 	}
847 	TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
848 	if (sf->m != NULL)
849 		LIST_REMOVE(sf, list_entry);
850 	LIST_INSERT_HEAD(hash_list, sf, list_entry);
851 	sf->ref_count = 1;
852 	sf->m = m;
853 	nsfbufsused++;
854 	nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
855 
856 	/*
857 	 * Update the sf_buf's virtual-to-physical mapping, flushing the
858 	 * virtual address from the TLB.  Since the reference count for
859 	 * the sf_buf's old mapping was zero, that mapping is not
860 	 * currently in use.  Consequently, there is no need to exchange
861 	 * the old and new PTEs atomically, even under PAE.
862 	 */
863 	ptep = vtopte(sf->kva);
864 	opte = *ptep;
865 #ifdef XEN
866        PT_SET_MA(sf->kva, xpmap_ptom(VM_PAGE_TO_PHYS(m)) | pgeflag
867 	   | PG_RW | PG_V | pmap_cache_bits(m->md.pat_mode, 0));
868 #else
869 	*ptep = VM_PAGE_TO_PHYS(m) | pgeflag | PG_RW | PG_V |
870 	    pmap_cache_bits(m->md.pat_mode, 0);
871 #endif
872 
873 	/*
874 	 * Avoid unnecessary TLB invalidations: If the sf_buf's old
875 	 * virtual-to-physical mapping was not used, then any processor
876 	 * that has invalidated the sf_buf's virtual address from its TLB
877 	 * since the last used mapping need not invalidate again.
878 	 */
879 #ifdef SMP
880 	if ((opte & (PG_V | PG_A)) ==  (PG_V | PG_A))
881 		CPU_ZERO(&sf->cpumask);
882 shootdown:
883 	sched_pin();
884 	cpuid = PCPU_GET(cpuid);
885 	if (!CPU_ISSET(cpuid, &sf->cpumask)) {
886 		CPU_SET(cpuid, &sf->cpumask);
887 		invlpg(sf->kva);
888 	}
889 	if ((flags & SFB_CPUPRIVATE) == 0) {
890 		other_cpus = all_cpus;
891 		CPU_CLR(cpuid, &other_cpus);
892 		CPU_NAND(&other_cpus, &sf->cpumask);
893 		if (!CPU_EMPTY(&other_cpus)) {
894 			CPU_OR(&sf->cpumask, &other_cpus);
895 			smp_masked_invlpg(other_cpus, sf->kva);
896 		}
897 	}
898 	sched_unpin();
899 #else
900 	if ((opte & (PG_V | PG_A)) ==  (PG_V | PG_A))
901 		pmap_invalidate_page(kernel_pmap, sf->kva);
902 #endif
903 done:
904 	mtx_unlock(&sf_buf_lock);
905 	return (sf);
906 }
907 
908 /*
909  * Remove a reference from the given sf_buf, adding it to the free
910  * list when its reference count reaches zero.  A freed sf_buf still,
911  * however, retains its virtual-to-physical mapping until it is
912  * recycled or reactivated by sf_buf_alloc(9).
913  */
914 void
915 sf_buf_free(struct sf_buf *sf)
916 {
917 
918 	mtx_lock(&sf_buf_lock);
919 	sf->ref_count--;
920 	if (sf->ref_count == 0) {
921 		TAILQ_INSERT_TAIL(&sf_buf_freelist, sf, free_entry);
922 		nsfbufsused--;
923 #ifdef XEN
924 /*
925  * Xen doesn't like having dangling R/W mappings
926  */
927 		pmap_qremove(sf->kva, 1);
928 		sf->m = NULL;
929 		LIST_REMOVE(sf, list_entry);
930 #endif
931 		if (sf_buf_alloc_want > 0)
932 			wakeup(&sf_buf_freelist);
933 	}
934 	mtx_unlock(&sf_buf_lock);
935 }
936 
937 /*
938  * Software interrupt handler for queued VM system processing.
939  */
940 void
941 swi_vm(void *dummy)
942 {
943 	if (busdma_swi_pending != 0)
944 		busdma_swi();
945 }
946 
947 /*
948  * Tell whether this address is in some physical memory region.
949  * Currently used by the kernel coredump code in order to avoid
950  * dumping the ``ISA memory hole'' which could cause indefinite hangs,
951  * or other unpredictable behaviour.
952  */
953 
954 int
955 is_physical_memory(vm_paddr_t addr)
956 {
957 
958 #ifdef DEV_ISA
959 	/* The ISA ``memory hole''. */
960 	if (addr >= 0xa0000 && addr < 0x100000)
961 		return 0;
962 #endif
963 
964 	/*
965 	 * stuff other tests for known memory-mapped devices (PCI?)
966 	 * here
967 	 */
968 
969 	return 1;
970 }
971