xref: /freebsd/sys/i386/i386/vm_machdep.c (revision 6472ac3d8a86336899b6cfb789a4cd9897e3fab5)
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 void	cpu_reset_real(void);
110 #ifdef SMP
111 static void	cpu_reset_proxy(void);
112 static u_int	cpu_reset_proxyid;
113 static volatile u_int	cpu_reset_proxy_active;
114 #endif
115 static void	sf_buf_init(void *arg);
116 SYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL);
117 
118 LIST_HEAD(sf_head, sf_buf);
119 
120 /*
121  * A hash table of active sendfile(2) buffers
122  */
123 static struct sf_head *sf_buf_active;
124 static u_long sf_buf_hashmask;
125 
126 #define	SF_BUF_HASH(m)	(((m) - vm_page_array) & sf_buf_hashmask)
127 
128 static TAILQ_HEAD(, sf_buf) sf_buf_freelist;
129 static u_int	sf_buf_alloc_want;
130 
131 /*
132  * A lock used to synchronize access to the hash table and free list
133  */
134 static struct mtx sf_buf_lock;
135 
136 extern int	_ucodesel, _udatasel;
137 
138 /*
139  * Finish a fork operation, with process p2 nearly set up.
140  * Copy and update the pcb, set up the stack so that the child
141  * ready to run and return to user mode.
142  */
143 void
144 cpu_fork(td1, p2, td2, flags)
145 	register struct thread *td1;
146 	register struct proc *p2;
147 	struct thread *td2;
148 	int flags;
149 {
150 	register struct proc *p1;
151 	struct pcb *pcb2;
152 	struct mdproc *mdp2;
153 
154 	p1 = td1->td_proc;
155 	if ((flags & RFPROC) == 0) {
156 		if ((flags & RFMEM) == 0) {
157 			/* unshare user LDT */
158 			struct mdproc *mdp1 = &p1->p_md;
159 			struct proc_ldt *pldt, *pldt1;
160 
161 			mtx_lock_spin(&dt_lock);
162 			if ((pldt1 = mdp1->md_ldt) != NULL &&
163 			    pldt1->ldt_refcnt > 1) {
164 				pldt = user_ldt_alloc(mdp1, pldt1->ldt_len);
165 				if (pldt == NULL)
166 					panic("could not copy LDT");
167 				mdp1->md_ldt = pldt;
168 				set_user_ldt(mdp1);
169 				user_ldt_deref(pldt1);
170 			} else
171 				mtx_unlock_spin(&dt_lock);
172 		}
173 		return;
174 	}
175 
176 	/* Ensure that td1's pcb is up to date. */
177 	if (td1 == curthread)
178 		td1->td_pcb->pcb_gs = rgs();
179 #ifdef DEV_NPX
180 	critical_enter();
181 	if (PCPU_GET(fpcurthread) == td1)
182 		npxsave(td1->td_pcb->pcb_save);
183 	critical_exit();
184 #endif
185 
186 	/* Point the pcb to the top of the stack */
187 	pcb2 = (struct pcb *)(td2->td_kstack +
188 	    td2->td_kstack_pages * PAGE_SIZE) - 1;
189 	td2->td_pcb = pcb2;
190 
191 	/* Copy td1's pcb */
192 	bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
193 
194 	/* Properly initialize pcb_save */
195 	pcb2->pcb_save = &pcb2->pcb_user_save;
196 
197 	/* Point mdproc and then copy over td1's contents */
198 	mdp2 = &p2->p_md;
199 	bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
200 
201 	/*
202 	 * Create a new fresh stack for the new process.
203 	 * Copy the trap frame for the return to user mode as if from a
204 	 * syscall.  This copies most of the user mode register values.
205 	 * The -16 is so we can expand the trapframe if we go to vm86.
206 	 */
207 	td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1;
208 	bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
209 
210 	td2->td_frame->tf_eax = 0;		/* Child returns zero */
211 	td2->td_frame->tf_eflags &= ~PSL_C;	/* success */
212 	td2->td_frame->tf_edx = 1;
213 
214 	/*
215 	 * If the parent process has the trap bit set (i.e. a debugger had
216 	 * single stepped the process to the system call), we need to clear
217 	 * the trap flag from the new frame unless the debugger had set PF_FORK
218 	 * on the parent.  Otherwise, the child will receive a (likely
219 	 * unexpected) SIGTRAP when it executes the first instruction after
220 	 * returning  to userland.
221 	 */
222 	if ((p1->p_pfsflags & PF_FORK) == 0)
223 		td2->td_frame->tf_eflags &= ~PSL_T;
224 
225 	/*
226 	 * Set registers for trampoline to user mode.  Leave space for the
227 	 * return address on stack.  These are the kernel mode register values.
228 	 */
229 #ifdef PAE
230 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdpt);
231 #else
232 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
233 #endif
234 	pcb2->pcb_edi = 0;
235 	pcb2->pcb_esi = (int)fork_return;	/* fork_trampoline argument */
236 	pcb2->pcb_ebp = 0;
237 	pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
238 	pcb2->pcb_ebx = (int)td2;		/* fork_trampoline argument */
239 	pcb2->pcb_eip = (int)fork_trampoline;
240 	pcb2->pcb_psl = PSL_KERNEL;		/* ints disabled */
241 	/*-
242 	 * pcb2->pcb_dr*:	cloned above.
243 	 * pcb2->pcb_savefpu:	cloned above.
244 	 * pcb2->pcb_flags:	cloned above.
245 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
246 	 * pcb2->pcb_gs:	cloned above.
247 	 * pcb2->pcb_ext:	cleared below.
248 	 */
249 
250 	/*
251 	 * XXX don't copy the i/o pages.  this should probably be fixed.
252 	 */
253 	pcb2->pcb_ext = 0;
254 
255 	/* Copy the LDT, if necessary. */
256 	mtx_lock_spin(&dt_lock);
257 	if (mdp2->md_ldt != NULL) {
258 		if (flags & RFMEM) {
259 			mdp2->md_ldt->ldt_refcnt++;
260 		} else {
261 			mdp2->md_ldt = user_ldt_alloc(mdp2,
262 			    mdp2->md_ldt->ldt_len);
263 			if (mdp2->md_ldt == NULL)
264 				panic("could not copy LDT");
265 		}
266 	}
267 	mtx_unlock_spin(&dt_lock);
268 
269 	/* Setup to release spin count in fork_exit(). */
270 	td2->td_md.md_spinlock_count = 1;
271 	/*
272 	 * XXX XEN need to check on PSL_USER is handled
273 	 */
274 	td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
275 	/*
276 	 * Now, cpu_switch() can schedule the new process.
277 	 * pcb_esp is loaded pointing to the cpu_switch() stack frame
278 	 * containing the return address when exiting cpu_switch.
279 	 * This will normally be to fork_trampoline(), which will have
280 	 * %ebx loaded with the new proc's pointer.  fork_trampoline()
281 	 * will set up a stack to call fork_return(p, frame); to complete
282 	 * the return to user-mode.
283 	 */
284 }
285 
286 /*
287  * Intercept the return address from a freshly forked process that has NOT
288  * been scheduled yet.
289  *
290  * This is needed to make kernel threads stay in kernel mode.
291  */
292 void
293 cpu_set_fork_handler(td, func, arg)
294 	struct thread *td;
295 	void (*func)(void *);
296 	void *arg;
297 {
298 	/*
299 	 * Note that the trap frame follows the args, so the function
300 	 * is really called like this:  func(arg, frame);
301 	 */
302 	td->td_pcb->pcb_esi = (int) func;	/* function */
303 	td->td_pcb->pcb_ebx = (int) arg;	/* first arg */
304 }
305 
306 void
307 cpu_exit(struct thread *td)
308 {
309 
310 	/*
311 	 * If this process has a custom LDT, release it.  Reset pc->pcb_gs
312 	 * and %gs before we free it in case they refer to an LDT entry.
313 	 */
314 	mtx_lock_spin(&dt_lock);
315 	if (td->td_proc->p_md.md_ldt) {
316 		td->td_pcb->pcb_gs = _udatasel;
317 		load_gs(_udatasel);
318 		user_ldt_free(td);
319 	} else
320 		mtx_unlock_spin(&dt_lock);
321 }
322 
323 void
324 cpu_thread_exit(struct thread *td)
325 {
326 
327 #ifdef DEV_NPX
328 	critical_enter();
329 	if (td == PCPU_GET(fpcurthread))
330 		npxdrop();
331 	critical_exit();
332 #endif
333 
334 	/* Disable any hardware breakpoints. */
335 	if (td->td_pcb->pcb_flags & PCB_DBREGS) {
336 		reset_dbregs();
337 		td->td_pcb->pcb_flags &= ~PCB_DBREGS;
338 	}
339 }
340 
341 void
342 cpu_thread_clean(struct thread *td)
343 {
344 	struct pcb *pcb;
345 
346 	pcb = td->td_pcb;
347 	if (pcb->pcb_ext != NULL) {
348 		/* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */
349 		/*
350 		 * XXX do we need to move the TSS off the allocated pages
351 		 * before freeing them?  (not done here)
352 		 */
353 		kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
354 		    ctob(IOPAGES + 1));
355 		pcb->pcb_ext = NULL;
356 	}
357 }
358 
359 void
360 cpu_thread_swapin(struct thread *td)
361 {
362 }
363 
364 void
365 cpu_thread_swapout(struct thread *td)
366 {
367 }
368 
369 void
370 cpu_thread_alloc(struct thread *td)
371 {
372 
373 	td->td_pcb = (struct pcb *)(td->td_kstack +
374 	    td->td_kstack_pages * PAGE_SIZE) - 1;
375 	td->td_frame = (struct trapframe *)((caddr_t)td->td_pcb - 16) - 1;
376 	td->td_pcb->pcb_ext = NULL;
377 	td->td_pcb->pcb_save = &td->td_pcb->pcb_user_save;
378 }
379 
380 void
381 cpu_thread_free(struct thread *td)
382 {
383 
384 	cpu_thread_clean(td);
385 }
386 
387 void
388 cpu_set_syscall_retval(struct thread *td, int error)
389 {
390 
391 	switch (error) {
392 	case 0:
393 		td->td_frame->tf_eax = td->td_retval[0];
394 		td->td_frame->tf_edx = td->td_retval[1];
395 		td->td_frame->tf_eflags &= ~PSL_C;
396 		break;
397 
398 	case ERESTART:
399 		/*
400 		 * Reconstruct pc, assuming lcall $X,y is 7 bytes, int
401 		 * 0x80 is 2 bytes. We saved this in tf_err.
402 		 */
403 		td->td_frame->tf_eip -= td->td_frame->tf_err;
404 		break;
405 
406 	case EJUSTRETURN:
407 		break;
408 
409 	default:
410 		if (td->td_proc->p_sysent->sv_errsize) {
411 			if (error >= td->td_proc->p_sysent->sv_errsize)
412 				error = -1;	/* XXX */
413 			else
414 				error = td->td_proc->p_sysent->sv_errtbl[error];
415 		}
416 		td->td_frame->tf_eax = error;
417 		td->td_frame->tf_eflags |= PSL_C;
418 		break;
419 	}
420 }
421 
422 /*
423  * Initialize machine state (pcb and trap frame) for a new thread about to
424  * upcall. Put enough state in the new thread's PCB to get it to go back
425  * userret(), where we can intercept it again to set the return (upcall)
426  * Address and stack, along with those from upcals that are from other sources
427  * such as those generated in thread_userret() itself.
428  */
429 void
430 cpu_set_upcall(struct thread *td, struct thread *td0)
431 {
432 	struct pcb *pcb2;
433 
434 	/* Point the pcb to the top of the stack. */
435 	pcb2 = td->td_pcb;
436 
437 	/*
438 	 * Copy the upcall pcb.  This loads kernel regs.
439 	 * Those not loaded individually below get their default
440 	 * values here.
441 	 */
442 	bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
443 	pcb2->pcb_flags &= ~(PCB_NPXINITDONE | PCB_NPXUSERINITDONE);
444 	pcb2->pcb_save = &pcb2->pcb_user_save;
445 
446 	/*
447 	 * Create a new fresh stack for the new thread.
448 	 */
449 	bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
450 
451 	/* If the current thread has the trap bit set (i.e. a debugger had
452 	 * single stepped the process to the system call), we need to clear
453 	 * the trap flag from the new frame. Otherwise, the new thread will
454 	 * receive a (likely unexpected) SIGTRAP when it executes the first
455 	 * instruction after returning to userland.
456 	 */
457 	td->td_frame->tf_eflags &= ~PSL_T;
458 
459 	/*
460 	 * Set registers for trampoline to user mode.  Leave space for the
461 	 * return address on stack.  These are the kernel mode register values.
462 	 */
463 	pcb2->pcb_edi = 0;
464 	pcb2->pcb_esi = (int)fork_return;		    /* trampoline arg */
465 	pcb2->pcb_ebp = 0;
466 	pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */
467 	pcb2->pcb_ebx = (int)td;			    /* trampoline arg */
468 	pcb2->pcb_eip = (int)fork_trampoline;
469 	pcb2->pcb_psl &= ~(PSL_I);	/* interrupts must be disabled */
470 	pcb2->pcb_gs = rgs();
471 	/*
472 	 * If we didn't copy the pcb, we'd need to do the following registers:
473 	 * pcb2->pcb_cr3:	cloned above.
474 	 * pcb2->pcb_dr*:	cloned above.
475 	 * pcb2->pcb_savefpu:	cloned above.
476 	 * pcb2->pcb_flags:	cloned above.
477 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
478 	 * pcb2->pcb_gs:	cloned above.
479 	 * pcb2->pcb_ext:	cleared below.
480 	 */
481 	pcb2->pcb_ext = NULL;
482 
483 	/* Setup to release spin count in fork_exit(). */
484 	td->td_md.md_spinlock_count = 1;
485 	td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
486 }
487 
488 /*
489  * Set that machine state for performing an upcall that has to
490  * be done in thread_userret() so that those upcalls generated
491  * in thread_userret() itself can be done as well.
492  */
493 void
494 cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg,
495 	stack_t *stack)
496 {
497 
498 	/*
499 	 * Do any extra cleaning that needs to be done.
500 	 * The thread may have optional components
501 	 * that are not present in a fresh thread.
502 	 * This may be a recycled thread so make it look
503 	 * as though it's newly allocated.
504 	 */
505 	cpu_thread_clean(td);
506 
507 	/*
508 	 * Set the trap frame to point at the beginning of the uts
509 	 * function.
510 	 */
511 	td->td_frame->tf_ebp = 0;
512 	td->td_frame->tf_esp =
513 	    (((int)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
514 	td->td_frame->tf_eip = (int)entry;
515 
516 	/*
517 	 * Pass the address of the mailbox for this kse to the uts
518 	 * function as a parameter on the stack.
519 	 */
520 	suword((void *)(td->td_frame->tf_esp + sizeof(void *)),
521 	    (int)arg);
522 }
523 
524 int
525 cpu_set_user_tls(struct thread *td, void *tls_base)
526 {
527 	struct segment_descriptor sd;
528 	uint32_t base;
529 
530 	/*
531 	 * Construct a descriptor and store it in the pcb for
532 	 * the next context switch.  Also store it in the gdt
533 	 * so that the load of tf_fs into %fs will activate it
534 	 * at return to userland.
535 	 */
536 	base = (uint32_t)tls_base;
537 	sd.sd_lobase = base & 0xffffff;
538 	sd.sd_hibase = (base >> 24) & 0xff;
539 	sd.sd_lolimit = 0xffff;	/* 4GB limit, wraps around */
540 	sd.sd_hilimit = 0xf;
541 	sd.sd_type  = SDT_MEMRWA;
542 	sd.sd_dpl   = SEL_UPL;
543 	sd.sd_p     = 1;
544 	sd.sd_xx    = 0;
545 	sd.sd_def32 = 1;
546 	sd.sd_gran  = 1;
547 	critical_enter();
548 	/* set %gs */
549 	td->td_pcb->pcb_gsd = sd;
550 	if (td == curthread) {
551 		PCPU_GET(fsgs_gdt)[1] = sd;
552 		load_gs(GSEL(GUGS_SEL, SEL_UPL));
553 	}
554 	critical_exit();
555 	return (0);
556 }
557 
558 /*
559  * Convert kernel VA to physical address
560  */
561 vm_paddr_t
562 kvtop(void *addr)
563 {
564 	vm_paddr_t pa;
565 
566 	pa = pmap_kextract((vm_offset_t)addr);
567 	if (pa == 0)
568 		panic("kvtop: zero page frame");
569 	return (pa);
570 }
571 
572 #ifdef SMP
573 static void
574 cpu_reset_proxy()
575 {
576 	cpuset_t tcrp;
577 
578 	cpu_reset_proxy_active = 1;
579 	while (cpu_reset_proxy_active == 1)
580 		;	/* Wait for other cpu to see that we've started */
581 	CPU_SETOF(cpu_reset_proxyid, &tcrp);
582 	stop_cpus(tcrp);
583 	printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
584 	DELAY(1000000);
585 	cpu_reset_real();
586 }
587 #endif
588 
589 void
590 cpu_reset()
591 {
592 #ifdef XBOX
593 	if (arch_i386_is_xbox) {
594 		/* Kick the PIC16L, it can reboot the box */
595 		pic16l_reboot();
596 		for (;;);
597 	}
598 #endif
599 
600 #ifdef SMP
601 	cpuset_t map;
602 	u_int cnt;
603 
604 	if (smp_active) {
605 		map = all_cpus;
606 		CPU_CLR(PCPU_GET(cpuid), &map);
607 		CPU_NAND(&map, &stopped_cpus);
608 		if (!CPU_EMPTY(&map)) {
609 			printf("cpu_reset: Stopping other CPUs\n");
610 			stop_cpus(map);
611 		}
612 
613 		if (PCPU_GET(cpuid) != 0) {
614 			cpu_reset_proxyid = PCPU_GET(cpuid);
615 			cpustop_restartfunc = cpu_reset_proxy;
616 			cpu_reset_proxy_active = 0;
617 			printf("cpu_reset: Restarting BSP\n");
618 
619 			/* Restart CPU #0. */
620 			/* XXX: restart_cpus(1 << 0); */
621 			CPU_SETOF(0, &started_cpus);
622 			wmb();
623 
624 			cnt = 0;
625 			while (cpu_reset_proxy_active == 0 && cnt < 10000000)
626 				cnt++;	/* Wait for BSP to announce restart */
627 			if (cpu_reset_proxy_active == 0)
628 				printf("cpu_reset: Failed to restart BSP\n");
629 			enable_intr();
630 			cpu_reset_proxy_active = 2;
631 
632 			while (1);
633 			/* NOTREACHED */
634 		}
635 
636 		DELAY(1000000);
637 	}
638 #endif
639 	cpu_reset_real();
640 	/* NOTREACHED */
641 }
642 
643 static void
644 cpu_reset_real()
645 {
646 	struct region_descriptor null_idt;
647 #ifndef PC98
648 	int b;
649 #endif
650 
651 	disable_intr();
652 #ifdef XEN
653 	if (smp_processor_id() == 0)
654 		HYPERVISOR_shutdown(SHUTDOWN_reboot);
655 	else
656 		HYPERVISOR_shutdown(SHUTDOWN_poweroff);
657 #endif
658 #ifdef CPU_ELAN
659 	if (elan_mmcr != NULL)
660 		elan_mmcr->RESCFG = 1;
661 #endif
662 
663 	if (cpu == CPU_GEODE1100) {
664 		/* Attempt Geode's own reset */
665 		outl(0xcf8, 0x80009044ul);
666 		outl(0xcfc, 0xf);
667 	}
668 
669 #ifdef PC98
670 	/*
671 	 * Attempt to do a CPU reset via CPU reset port.
672 	 */
673 	if ((inb(0x35) & 0xa0) != 0xa0) {
674 		outb(0x37, 0x0f);		/* SHUT0 = 0. */
675 		outb(0x37, 0x0b);		/* SHUT1 = 0. */
676 	}
677 	outb(0xf0, 0x00);		/* Reset. */
678 #else
679 #if !defined(BROKEN_KEYBOARD_RESET)
680 	/*
681 	 * Attempt to do a CPU reset via the keyboard controller,
682 	 * do not turn off GateA20, as any machine that fails
683 	 * to do the reset here would then end up in no man's land.
684 	 */
685 	outb(IO_KBD + 4, 0xFE);
686 	DELAY(500000);	/* wait 0.5 sec to see if that did it */
687 #endif
688 
689 	/*
690 	 * Attempt to force a reset via the Reset Control register at
691 	 * I/O port 0xcf9.  Bit 2 forces a system reset when it
692 	 * transitions from 0 to 1.  Bit 1 selects the type of reset
693 	 * to attempt: 0 selects a "soft" reset, and 1 selects a
694 	 * "hard" reset.  We try a "hard" reset.  The first write sets
695 	 * bit 1 to select a "hard" reset and clears bit 2.  The
696 	 * second write forces a 0 -> 1 transition in bit 2 to trigger
697 	 * a reset.
698 	 */
699 	outb(0xcf9, 0x2);
700 	outb(0xcf9, 0x6);
701 	DELAY(500000);  /* wait 0.5 sec to see if that did it */
702 
703 	/*
704 	 * Attempt to force a reset via the Fast A20 and Init register
705 	 * at I/O port 0x92.  Bit 1 serves as an alternate A20 gate.
706 	 * Bit 0 asserts INIT# when set to 1.  We are careful to only
707 	 * preserve bit 1 while setting bit 0.  We also must clear bit
708 	 * 0 before setting it if it isn't already clear.
709 	 */
710 	b = inb(0x92);
711 	if (b != 0xff) {
712 		if ((b & 0x1) != 0)
713 			outb(0x92, b & 0xfe);
714 		outb(0x92, b | 0x1);
715 		DELAY(500000);  /* wait 0.5 sec to see if that did it */
716 	}
717 #endif /* PC98 */
718 
719 	printf("No known reset method worked, attempting CPU shutdown\n");
720 	DELAY(1000000); /* wait 1 sec for printf to complete */
721 
722 	/* Wipe the IDT. */
723 	null_idt.rd_limit = 0;
724 	null_idt.rd_base = 0;
725 	lidt(&null_idt);
726 
727 	/* "good night, sweet prince .... <THUNK!>" */
728 	breakpoint();
729 
730 	/* NOTREACHED */
731 	while(1);
732 }
733 
734 /*
735  * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-))
736  */
737 static void
738 sf_buf_init(void *arg)
739 {
740 	struct sf_buf *sf_bufs;
741 	vm_offset_t sf_base;
742 	int i;
743 
744 	nsfbufs = NSFBUFS;
745 	TUNABLE_INT_FETCH("kern.ipc.nsfbufs", &nsfbufs);
746 
747 	sf_buf_active = hashinit(nsfbufs, M_TEMP, &sf_buf_hashmask);
748 	TAILQ_INIT(&sf_buf_freelist);
749 	sf_base = kmem_alloc_nofault(kernel_map, nsfbufs * PAGE_SIZE);
750 	sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP,
751 	    M_NOWAIT | M_ZERO);
752 	for (i = 0; i < nsfbufs; i++) {
753 		sf_bufs[i].kva = sf_base + i * PAGE_SIZE;
754 		TAILQ_INSERT_TAIL(&sf_buf_freelist, &sf_bufs[i], free_entry);
755 	}
756 	sf_buf_alloc_want = 0;
757 	mtx_init(&sf_buf_lock, "sf_buf", NULL, MTX_DEF);
758 }
759 
760 /*
761  * Invalidate the cache lines that may belong to the page, if
762  * (possibly old) mapping of the page by sf buffer exists.  Returns
763  * TRUE when mapping was found and cache invalidated.
764  */
765 boolean_t
766 sf_buf_invalidate_cache(vm_page_t m)
767 {
768 	struct sf_head *hash_list;
769 	struct sf_buf *sf;
770 	boolean_t ret;
771 
772 	hash_list = &sf_buf_active[SF_BUF_HASH(m)];
773 	ret = FALSE;
774 	mtx_lock(&sf_buf_lock);
775 	LIST_FOREACH(sf, hash_list, list_entry) {
776 		if (sf->m == m) {
777 			/*
778 			 * Use pmap_qenter to update the pte for
779 			 * existing mapping, in particular, the PAT
780 			 * settings are recalculated.
781 			 */
782 			pmap_qenter(sf->kva, &m, 1);
783 			pmap_invalidate_cache_range(sf->kva, sf->kva +
784 			    PAGE_SIZE);
785 			ret = TRUE;
786 			break;
787 		}
788 	}
789 	mtx_unlock(&sf_buf_lock);
790 	return (ret);
791 }
792 
793 /*
794  * Get an sf_buf from the freelist.  May block if none are available.
795  */
796 struct sf_buf *
797 sf_buf_alloc(struct vm_page *m, int flags)
798 {
799 	pt_entry_t opte, *ptep;
800 	struct sf_head *hash_list;
801 	struct sf_buf *sf;
802 #ifdef SMP
803 	cpuset_t other_cpus;
804 	u_int cpuid;
805 #endif
806 	int error;
807 
808 	KASSERT(curthread->td_pinned > 0 || (flags & SFB_CPUPRIVATE) == 0,
809 	    ("sf_buf_alloc(SFB_CPUPRIVATE): curthread not pinned"));
810 	hash_list = &sf_buf_active[SF_BUF_HASH(m)];
811 	mtx_lock(&sf_buf_lock);
812 	LIST_FOREACH(sf, hash_list, list_entry) {
813 		if (sf->m == m) {
814 			sf->ref_count++;
815 			if (sf->ref_count == 1) {
816 				TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
817 				nsfbufsused++;
818 				nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
819 			}
820 #ifdef SMP
821 			goto shootdown;
822 #else
823 			goto done;
824 #endif
825 		}
826 	}
827 	while ((sf = TAILQ_FIRST(&sf_buf_freelist)) == NULL) {
828 		if (flags & SFB_NOWAIT)
829 			goto done;
830 		sf_buf_alloc_want++;
831 		mbstat.sf_allocwait++;
832 		error = msleep(&sf_buf_freelist, &sf_buf_lock,
833 		    (flags & SFB_CATCH) ? PCATCH | PVM : PVM, "sfbufa", 0);
834 		sf_buf_alloc_want--;
835 
836 		/*
837 		 * If we got a signal, don't risk going back to sleep.
838 		 */
839 		if (error)
840 			goto done;
841 	}
842 	TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
843 	if (sf->m != NULL)
844 		LIST_REMOVE(sf, list_entry);
845 	LIST_INSERT_HEAD(hash_list, sf, list_entry);
846 	sf->ref_count = 1;
847 	sf->m = m;
848 	nsfbufsused++;
849 	nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
850 
851 	/*
852 	 * Update the sf_buf's virtual-to-physical mapping, flushing the
853 	 * virtual address from the TLB.  Since the reference count for
854 	 * the sf_buf's old mapping was zero, that mapping is not
855 	 * currently in use.  Consequently, there is no need to exchange
856 	 * the old and new PTEs atomically, even under PAE.
857 	 */
858 	ptep = vtopte(sf->kva);
859 	opte = *ptep;
860 #ifdef XEN
861        PT_SET_MA(sf->kva, xpmap_ptom(VM_PAGE_TO_PHYS(m)) | pgeflag
862 	   | PG_RW | PG_V | pmap_cache_bits(m->md.pat_mode, 0));
863 #else
864 	*ptep = VM_PAGE_TO_PHYS(m) | pgeflag | PG_RW | PG_V |
865 	    pmap_cache_bits(m->md.pat_mode, 0);
866 #endif
867 
868 	/*
869 	 * Avoid unnecessary TLB invalidations: If the sf_buf's old
870 	 * virtual-to-physical mapping was not used, then any processor
871 	 * that has invalidated the sf_buf's virtual address from its TLB
872 	 * since the last used mapping need not invalidate again.
873 	 */
874 #ifdef SMP
875 	if ((opte & (PG_V | PG_A)) ==  (PG_V | PG_A))
876 		CPU_ZERO(&sf->cpumask);
877 shootdown:
878 	sched_pin();
879 	cpuid = PCPU_GET(cpuid);
880 	if (!CPU_ISSET(cpuid, &sf->cpumask)) {
881 		CPU_SET(cpuid, &sf->cpumask);
882 		invlpg(sf->kva);
883 	}
884 	if ((flags & SFB_CPUPRIVATE) == 0) {
885 		other_cpus = all_cpus;
886 		CPU_CLR(cpuid, &other_cpus);
887 		CPU_NAND(&other_cpus, &sf->cpumask);
888 		if (!CPU_EMPTY(&other_cpus)) {
889 			CPU_OR(&sf->cpumask, &other_cpus);
890 			smp_masked_invlpg(other_cpus, sf->kva);
891 		}
892 	}
893 	sched_unpin();
894 #else
895 	if ((opte & (PG_V | PG_A)) ==  (PG_V | PG_A))
896 		pmap_invalidate_page(kernel_pmap, sf->kva);
897 #endif
898 done:
899 	mtx_unlock(&sf_buf_lock);
900 	return (sf);
901 }
902 
903 /*
904  * Remove a reference from the given sf_buf, adding it to the free
905  * list when its reference count reaches zero.  A freed sf_buf still,
906  * however, retains its virtual-to-physical mapping until it is
907  * recycled or reactivated by sf_buf_alloc(9).
908  */
909 void
910 sf_buf_free(struct sf_buf *sf)
911 {
912 
913 	mtx_lock(&sf_buf_lock);
914 	sf->ref_count--;
915 	if (sf->ref_count == 0) {
916 		TAILQ_INSERT_TAIL(&sf_buf_freelist, sf, free_entry);
917 		nsfbufsused--;
918 #ifdef XEN
919 /*
920  * Xen doesn't like having dangling R/W mappings
921  */
922 		pmap_qremove(sf->kva, 1);
923 		sf->m = NULL;
924 		LIST_REMOVE(sf, list_entry);
925 #endif
926 		if (sf_buf_alloc_want > 0)
927 			wakeup(&sf_buf_freelist);
928 	}
929 	mtx_unlock(&sf_buf_lock);
930 }
931 
932 /*
933  * Software interrupt handler for queued VM system processing.
934  */
935 void
936 swi_vm(void *dummy)
937 {
938 	if (busdma_swi_pending != 0)
939 		busdma_swi();
940 }
941 
942 /*
943  * Tell whether this address is in some physical memory region.
944  * Currently used by the kernel coredump code in order to avoid
945  * dumping the ``ISA memory hole'' which could cause indefinite hangs,
946  * or other unpredictable behaviour.
947  */
948 
949 int
950 is_physical_memory(vm_paddr_t addr)
951 {
952 
953 #ifdef DEV_ISA
954 	/* The ISA ``memory hole''. */
955 	if (addr >= 0xa0000 && addr < 0x100000)
956 		return 0;
957 #endif
958 
959 	/*
960 	 * stuff other tests for known memory-mapped devices (PCI?)
961 	 * here
962 	 */
963 
964 	return 1;
965 }
966