xref: /freebsd/sys/i386/i386/vm_machdep.c (revision ff0ba87247820afbdfdc1b307c803f7923d0e4d3)
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 <isa/isareg.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 #if !defined(CPU_DISABLE_SSE) && defined(I686_CPU)
110 #define CPU_ENABLE_SSE
111 #endif
112 
113 _Static_assert(OFFSETOF_CURTHREAD == offsetof(struct pcpu, pc_curthread),
114     "OFFSETOF_CURTHREAD does not correspond with offset of pc_curthread.");
115 _Static_assert(OFFSETOF_CURPCB == offsetof(struct pcpu, pc_curpcb),
116     "OFFSETOF_CURPCB does not correspond with offset of pc_curpcb.");
117 
118 static void	cpu_reset_real(void);
119 #ifdef SMP
120 static void	cpu_reset_proxy(void);
121 static u_int	cpu_reset_proxyid;
122 static volatile u_int	cpu_reset_proxy_active;
123 #endif
124 
125 union savefpu *
126 get_pcb_user_save_td(struct thread *td)
127 {
128 	vm_offset_t p;
129 
130 	p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
131 	    cpu_max_ext_state_size;
132 	KASSERT((p % 64) == 0, ("Unaligned pcb_user_save area"));
133 	return ((union savefpu *)p);
134 }
135 
136 union savefpu *
137 get_pcb_user_save_pcb(struct pcb *pcb)
138 {
139 	vm_offset_t p;
140 
141 	p = (vm_offset_t)(pcb + 1);
142 	return ((union savefpu *)p);
143 }
144 
145 struct pcb *
146 get_pcb_td(struct thread *td)
147 {
148 	vm_offset_t p;
149 
150 	p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
151 	    cpu_max_ext_state_size - sizeof(struct pcb);
152 	return ((struct pcb *)p);
153 }
154 
155 void *
156 alloc_fpusave(int flags)
157 {
158 	void *res;
159 #ifdef CPU_ENABLE_SSE
160 	struct savefpu_ymm *sf;
161 #endif
162 
163 	res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags);
164 #ifdef CPU_ENABLE_SSE
165 	if (use_xsave) {
166 		sf = (struct savefpu_ymm *)res;
167 		bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd));
168 		sf->sv_xstate.sx_hd.xstate_bv = xsave_mask;
169 	}
170 #endif
171 	return (res);
172 }
173 /*
174  * Finish a fork operation, with process p2 nearly set up.
175  * Copy and update the pcb, set up the stack so that the child
176  * ready to run and return to user mode.
177  */
178 void
179 cpu_fork(td1, p2, td2, flags)
180 	register struct thread *td1;
181 	register struct proc *p2;
182 	struct thread *td2;
183 	int flags;
184 {
185 	register struct proc *p1;
186 	struct pcb *pcb2;
187 	struct mdproc *mdp2;
188 
189 	p1 = td1->td_proc;
190 	if ((flags & RFPROC) == 0) {
191 		if ((flags & RFMEM) == 0) {
192 			/* unshare user LDT */
193 			struct mdproc *mdp1 = &p1->p_md;
194 			struct proc_ldt *pldt, *pldt1;
195 
196 			mtx_lock_spin(&dt_lock);
197 			if ((pldt1 = mdp1->md_ldt) != NULL &&
198 			    pldt1->ldt_refcnt > 1) {
199 				pldt = user_ldt_alloc(mdp1, pldt1->ldt_len);
200 				if (pldt == NULL)
201 					panic("could not copy LDT");
202 				mdp1->md_ldt = pldt;
203 				set_user_ldt(mdp1);
204 				user_ldt_deref(pldt1);
205 			} else
206 				mtx_unlock_spin(&dt_lock);
207 		}
208 		return;
209 	}
210 
211 	/* Ensure that td1's pcb is up to date. */
212 	if (td1 == curthread)
213 		td1->td_pcb->pcb_gs = rgs();
214 #ifdef DEV_NPX
215 	critical_enter();
216 	if (PCPU_GET(fpcurthread) == td1)
217 		npxsave(td1->td_pcb->pcb_save);
218 	critical_exit();
219 #endif
220 
221 	/* Point the pcb to the top of the stack */
222 	pcb2 = get_pcb_td(td2);
223 	td2->td_pcb = pcb2;
224 
225 	/* Copy td1's pcb */
226 	bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
227 
228 	/* Properly initialize pcb_save */
229 	pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
230 	bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2),
231 	    cpu_max_ext_state_size);
232 
233 	/* Point mdproc and then copy over td1's contents */
234 	mdp2 = &p2->p_md;
235 	bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
236 
237 	/*
238 	 * Create a new fresh stack for the new process.
239 	 * Copy the trap frame for the return to user mode as if from a
240 	 * syscall.  This copies most of the user mode register values.
241 	 * The -16 is so we can expand the trapframe if we go to vm86.
242 	 */
243 	td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1;
244 	bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
245 
246 	td2->td_frame->tf_eax = 0;		/* Child returns zero */
247 	td2->td_frame->tf_eflags &= ~PSL_C;	/* success */
248 	td2->td_frame->tf_edx = 1;
249 
250 	/*
251 	 * If the parent process has the trap bit set (i.e. a debugger had
252 	 * single stepped the process to the system call), we need to clear
253 	 * the trap flag from the new frame unless the debugger had set PF_FORK
254 	 * on the parent.  Otherwise, the child will receive a (likely
255 	 * unexpected) SIGTRAP when it executes the first instruction after
256 	 * returning  to userland.
257 	 */
258 	if ((p1->p_pfsflags & PF_FORK) == 0)
259 		td2->td_frame->tf_eflags &= ~PSL_T;
260 
261 	/*
262 	 * Set registers for trampoline to user mode.  Leave space for the
263 	 * return address on stack.  These are the kernel mode register values.
264 	 */
265 #ifdef PAE
266 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdpt);
267 #else
268 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
269 #endif
270 	pcb2->pcb_edi = 0;
271 	pcb2->pcb_esi = (int)fork_return;	/* fork_trampoline argument */
272 	pcb2->pcb_ebp = 0;
273 	pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
274 	pcb2->pcb_ebx = (int)td2;		/* fork_trampoline argument */
275 	pcb2->pcb_eip = (int)fork_trampoline;
276 	pcb2->pcb_psl = PSL_KERNEL;		/* ints disabled */
277 	/*-
278 	 * pcb2->pcb_dr*:	cloned above.
279 	 * pcb2->pcb_savefpu:	cloned above.
280 	 * pcb2->pcb_flags:	cloned above.
281 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
282 	 * pcb2->pcb_gs:	cloned above.
283 	 * pcb2->pcb_ext:	cleared below.
284 	 */
285 
286 	/*
287 	 * XXX don't copy the i/o pages.  this should probably be fixed.
288 	 */
289 	pcb2->pcb_ext = 0;
290 
291 	/* Copy the LDT, if necessary. */
292 	mtx_lock_spin(&dt_lock);
293 	if (mdp2->md_ldt != NULL) {
294 		if (flags & RFMEM) {
295 			mdp2->md_ldt->ldt_refcnt++;
296 		} else {
297 			mdp2->md_ldt = user_ldt_alloc(mdp2,
298 			    mdp2->md_ldt->ldt_len);
299 			if (mdp2->md_ldt == NULL)
300 				panic("could not copy LDT");
301 		}
302 	}
303 	mtx_unlock_spin(&dt_lock);
304 
305 	/* Setup to release spin count in fork_exit(). */
306 	td2->td_md.md_spinlock_count = 1;
307 	/*
308 	 * XXX XEN need to check on PSL_USER is handled
309 	 */
310 	td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
311 	/*
312 	 * Now, cpu_switch() can schedule the new process.
313 	 * pcb_esp is loaded pointing to the cpu_switch() stack frame
314 	 * containing the return address when exiting cpu_switch.
315 	 * This will normally be to fork_trampoline(), which will have
316 	 * %ebx loaded with the new proc's pointer.  fork_trampoline()
317 	 * will set up a stack to call fork_return(p, frame); to complete
318 	 * the return to user-mode.
319 	 */
320 }
321 
322 /*
323  * Intercept the return address from a freshly forked process that has NOT
324  * been scheduled yet.
325  *
326  * This is needed to make kernel threads stay in kernel mode.
327  */
328 void
329 cpu_set_fork_handler(td, func, arg)
330 	struct thread *td;
331 	void (*func)(void *);
332 	void *arg;
333 {
334 	/*
335 	 * Note that the trap frame follows the args, so the function
336 	 * is really called like this:  func(arg, frame);
337 	 */
338 	td->td_pcb->pcb_esi = (int) func;	/* function */
339 	td->td_pcb->pcb_ebx = (int) arg;	/* first arg */
340 }
341 
342 void
343 cpu_exit(struct thread *td)
344 {
345 
346 	/*
347 	 * If this process has a custom LDT, release it.  Reset pc->pcb_gs
348 	 * and %gs before we free it in case they refer to an LDT entry.
349 	 */
350 	mtx_lock_spin(&dt_lock);
351 	if (td->td_proc->p_md.md_ldt) {
352 		td->td_pcb->pcb_gs = _udatasel;
353 		load_gs(_udatasel);
354 		user_ldt_free(td);
355 	} else
356 		mtx_unlock_spin(&dt_lock);
357 }
358 
359 void
360 cpu_thread_exit(struct thread *td)
361 {
362 
363 #ifdef DEV_NPX
364 	critical_enter();
365 	if (td == PCPU_GET(fpcurthread))
366 		npxdrop();
367 	critical_exit();
368 #endif
369 
370 	/* Disable any hardware breakpoints. */
371 	if (td->td_pcb->pcb_flags & PCB_DBREGS) {
372 		reset_dbregs();
373 		td->td_pcb->pcb_flags &= ~PCB_DBREGS;
374 	}
375 }
376 
377 void
378 cpu_thread_clean(struct thread *td)
379 {
380 	struct pcb *pcb;
381 
382 	pcb = td->td_pcb;
383 	if (pcb->pcb_ext != NULL) {
384 		/* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */
385 		/*
386 		 * XXX do we need to move the TSS off the allocated pages
387 		 * before freeing them?  (not done here)
388 		 */
389 		kmem_free(kernel_arena, (vm_offset_t)pcb->pcb_ext,
390 		    ctob(IOPAGES + 1));
391 		pcb->pcb_ext = NULL;
392 	}
393 }
394 
395 void
396 cpu_thread_swapin(struct thread *td)
397 {
398 }
399 
400 void
401 cpu_thread_swapout(struct thread *td)
402 {
403 }
404 
405 void
406 cpu_thread_alloc(struct thread *td)
407 {
408 	struct pcb *pcb;
409 #ifdef CPU_ENABLE_SSE
410 	struct xstate_hdr *xhdr;
411 #endif
412 
413 	td->td_pcb = pcb = get_pcb_td(td);
414 	td->td_frame = (struct trapframe *)((caddr_t)pcb - 16) - 1;
415 	pcb->pcb_ext = NULL;
416 	pcb->pcb_save = get_pcb_user_save_pcb(pcb);
417 #ifdef CPU_ENABLE_SSE
418 	if (use_xsave) {
419 		xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
420 		bzero(xhdr, sizeof(*xhdr));
421 		xhdr->xstate_bv = xsave_mask;
422 	}
423 #endif
424 }
425 
426 void
427 cpu_thread_free(struct thread *td)
428 {
429 
430 	cpu_thread_clean(td);
431 }
432 
433 void
434 cpu_set_syscall_retval(struct thread *td, int error)
435 {
436 
437 	switch (error) {
438 	case 0:
439 		td->td_frame->tf_eax = td->td_retval[0];
440 		td->td_frame->tf_edx = td->td_retval[1];
441 		td->td_frame->tf_eflags &= ~PSL_C;
442 		break;
443 
444 	case ERESTART:
445 		/*
446 		 * Reconstruct pc, assuming lcall $X,y is 7 bytes, int
447 		 * 0x80 is 2 bytes. We saved this in tf_err.
448 		 */
449 		td->td_frame->tf_eip -= td->td_frame->tf_err;
450 		break;
451 
452 	case EJUSTRETURN:
453 		break;
454 
455 	default:
456 		if (td->td_proc->p_sysent->sv_errsize) {
457 			if (error >= td->td_proc->p_sysent->sv_errsize)
458 				error = -1;	/* XXX */
459 			else
460 				error = td->td_proc->p_sysent->sv_errtbl[error];
461 		}
462 		td->td_frame->tf_eax = error;
463 		td->td_frame->tf_eflags |= PSL_C;
464 		break;
465 	}
466 }
467 
468 /*
469  * Initialize machine state (pcb and trap frame) for a new thread about to
470  * upcall. Put enough state in the new thread's PCB to get it to go back
471  * userret(), where we can intercept it again to set the return (upcall)
472  * Address and stack, along with those from upcals that are from other sources
473  * such as those generated in thread_userret() itself.
474  */
475 void
476 cpu_set_upcall(struct thread *td, struct thread *td0)
477 {
478 	struct pcb *pcb2;
479 
480 	/* Point the pcb to the top of the stack. */
481 	pcb2 = td->td_pcb;
482 
483 	/*
484 	 * Copy the upcall pcb.  This loads kernel regs.
485 	 * Those not loaded individually below get their default
486 	 * values here.
487 	 */
488 	bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
489 	pcb2->pcb_flags &= ~(PCB_NPXINITDONE | PCB_NPXUSERINITDONE |
490 	    PCB_KERNNPX);
491 	pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
492 	bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save,
493 	    cpu_max_ext_state_size);
494 
495 	/*
496 	 * Create a new fresh stack for the new thread.
497 	 */
498 	bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
499 
500 	/* If the current thread has the trap bit set (i.e. a debugger had
501 	 * single stepped the process to the system call), we need to clear
502 	 * the trap flag from the new frame. Otherwise, the new thread will
503 	 * receive a (likely unexpected) SIGTRAP when it executes the first
504 	 * instruction after returning to userland.
505 	 */
506 	td->td_frame->tf_eflags &= ~PSL_T;
507 
508 	/*
509 	 * Set registers for trampoline to user mode.  Leave space for the
510 	 * return address on stack.  These are the kernel mode register values.
511 	 */
512 	pcb2->pcb_edi = 0;
513 	pcb2->pcb_esi = (int)fork_return;		    /* trampoline arg */
514 	pcb2->pcb_ebp = 0;
515 	pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */
516 	pcb2->pcb_ebx = (int)td;			    /* trampoline arg */
517 	pcb2->pcb_eip = (int)fork_trampoline;
518 	pcb2->pcb_psl &= ~(PSL_I);	/* interrupts must be disabled */
519 	pcb2->pcb_gs = rgs();
520 	/*
521 	 * If we didn't copy the pcb, we'd need to do the following registers:
522 	 * pcb2->pcb_cr3:	cloned above.
523 	 * pcb2->pcb_dr*:	cloned above.
524 	 * pcb2->pcb_savefpu:	cloned above.
525 	 * pcb2->pcb_flags:	cloned above.
526 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
527 	 * pcb2->pcb_gs:	cloned above.
528 	 * pcb2->pcb_ext:	cleared below.
529 	 */
530 	pcb2->pcb_ext = NULL;
531 
532 	/* Setup to release spin count in fork_exit(). */
533 	td->td_md.md_spinlock_count = 1;
534 	td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
535 }
536 
537 /*
538  * Set that machine state for performing an upcall that has to
539  * be done in thread_userret() so that those upcalls generated
540  * in thread_userret() itself can be done as well.
541  */
542 void
543 cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg,
544 	stack_t *stack)
545 {
546 
547 	/*
548 	 * Do any extra cleaning that needs to be done.
549 	 * The thread may have optional components
550 	 * that are not present in a fresh thread.
551 	 * This may be a recycled thread so make it look
552 	 * as though it's newly allocated.
553 	 */
554 	cpu_thread_clean(td);
555 
556 	/*
557 	 * Set the trap frame to point at the beginning of the uts
558 	 * function.
559 	 */
560 	td->td_frame->tf_ebp = 0;
561 	td->td_frame->tf_esp =
562 	    (((int)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
563 	td->td_frame->tf_eip = (int)entry;
564 
565 	/*
566 	 * Pass the address of the mailbox for this kse to the uts
567 	 * function as a parameter on the stack.
568 	 */
569 	suword((void *)(td->td_frame->tf_esp + sizeof(void *)),
570 	    (int)arg);
571 }
572 
573 int
574 cpu_set_user_tls(struct thread *td, void *tls_base)
575 {
576 	struct segment_descriptor sd;
577 	uint32_t base;
578 
579 	/*
580 	 * Construct a descriptor and store it in the pcb for
581 	 * the next context switch.  Also store it in the gdt
582 	 * so that the load of tf_fs into %fs will activate it
583 	 * at return to userland.
584 	 */
585 	base = (uint32_t)tls_base;
586 	sd.sd_lobase = base & 0xffffff;
587 	sd.sd_hibase = (base >> 24) & 0xff;
588 	sd.sd_lolimit = 0xffff;	/* 4GB limit, wraps around */
589 	sd.sd_hilimit = 0xf;
590 	sd.sd_type  = SDT_MEMRWA;
591 	sd.sd_dpl   = SEL_UPL;
592 	sd.sd_p     = 1;
593 	sd.sd_xx    = 0;
594 	sd.sd_def32 = 1;
595 	sd.sd_gran  = 1;
596 	critical_enter();
597 	/* set %gs */
598 	td->td_pcb->pcb_gsd = sd;
599 	if (td == curthread) {
600 		PCPU_GET(fsgs_gdt)[1] = sd;
601 		load_gs(GSEL(GUGS_SEL, SEL_UPL));
602 	}
603 	critical_exit();
604 	return (0);
605 }
606 
607 /*
608  * Convert kernel VA to physical address
609  */
610 vm_paddr_t
611 kvtop(void *addr)
612 {
613 	vm_paddr_t pa;
614 
615 	pa = pmap_kextract((vm_offset_t)addr);
616 	if (pa == 0)
617 		panic("kvtop: zero page frame");
618 	return (pa);
619 }
620 
621 #ifdef SMP
622 static void
623 cpu_reset_proxy()
624 {
625 	cpuset_t tcrp;
626 
627 	cpu_reset_proxy_active = 1;
628 	while (cpu_reset_proxy_active == 1)
629 		;	/* Wait for other cpu to see that we've started */
630 	CPU_SETOF(cpu_reset_proxyid, &tcrp);
631 	stop_cpus(tcrp);
632 	printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
633 	DELAY(1000000);
634 	cpu_reset_real();
635 }
636 #endif
637 
638 void
639 cpu_reset()
640 {
641 #ifdef XBOX
642 	if (arch_i386_is_xbox) {
643 		/* Kick the PIC16L, it can reboot the box */
644 		pic16l_reboot();
645 		for (;;);
646 	}
647 #endif
648 
649 #ifdef SMP
650 	cpuset_t map;
651 	u_int cnt;
652 
653 	if (smp_started) {
654 		map = all_cpus;
655 		CPU_CLR(PCPU_GET(cpuid), &map);
656 		CPU_NAND(&map, &stopped_cpus);
657 		if (!CPU_EMPTY(&map)) {
658 			printf("cpu_reset: Stopping other CPUs\n");
659 			stop_cpus(map);
660 		}
661 
662 		if (PCPU_GET(cpuid) != 0) {
663 			cpu_reset_proxyid = PCPU_GET(cpuid);
664 			cpustop_restartfunc = cpu_reset_proxy;
665 			cpu_reset_proxy_active = 0;
666 			printf("cpu_reset: Restarting BSP\n");
667 
668 			/* Restart CPU #0. */
669 			/* XXX: restart_cpus(1 << 0); */
670 			CPU_SETOF(0, &started_cpus);
671 			wmb();
672 
673 			cnt = 0;
674 			while (cpu_reset_proxy_active == 0 && cnt < 10000000)
675 				cnt++;	/* Wait for BSP to announce restart */
676 			if (cpu_reset_proxy_active == 0)
677 				printf("cpu_reset: Failed to restart BSP\n");
678 			enable_intr();
679 			cpu_reset_proxy_active = 2;
680 
681 			while (1);
682 			/* NOTREACHED */
683 		}
684 
685 		DELAY(1000000);
686 	}
687 #endif
688 	cpu_reset_real();
689 	/* NOTREACHED */
690 }
691 
692 static void
693 cpu_reset_real()
694 {
695 	struct region_descriptor null_idt;
696 #ifndef PC98
697 	int b;
698 #endif
699 
700 	disable_intr();
701 #ifdef XEN
702 	if (smp_processor_id() == 0)
703 		HYPERVISOR_shutdown(SHUTDOWN_reboot);
704 	else
705 		HYPERVISOR_shutdown(SHUTDOWN_poweroff);
706 #endif
707 #ifdef CPU_ELAN
708 	if (elan_mmcr != NULL)
709 		elan_mmcr->RESCFG = 1;
710 #endif
711 
712 	if (cpu == CPU_GEODE1100) {
713 		/* Attempt Geode's own reset */
714 		outl(0xcf8, 0x80009044ul);
715 		outl(0xcfc, 0xf);
716 	}
717 
718 #ifdef PC98
719 	/*
720 	 * Attempt to do a CPU reset via CPU reset port.
721 	 */
722 	if ((inb(0x35) & 0xa0) != 0xa0) {
723 		outb(0x37, 0x0f);		/* SHUT0 = 0. */
724 		outb(0x37, 0x0b);		/* SHUT1 = 0. */
725 	}
726 	outb(0xf0, 0x00);		/* Reset. */
727 #else
728 #if !defined(BROKEN_KEYBOARD_RESET)
729 	/*
730 	 * Attempt to do a CPU reset via the keyboard controller,
731 	 * do not turn off GateA20, as any machine that fails
732 	 * to do the reset here would then end up in no man's land.
733 	 */
734 	outb(IO_KBD + 4, 0xFE);
735 	DELAY(500000);	/* wait 0.5 sec to see if that did it */
736 #endif
737 
738 	/*
739 	 * Attempt to force a reset via the Reset Control register at
740 	 * I/O port 0xcf9.  Bit 2 forces a system reset when it
741 	 * transitions from 0 to 1.  Bit 1 selects the type of reset
742 	 * to attempt: 0 selects a "soft" reset, and 1 selects a
743 	 * "hard" reset.  We try a "hard" reset.  The first write sets
744 	 * bit 1 to select a "hard" reset and clears bit 2.  The
745 	 * second write forces a 0 -> 1 transition in bit 2 to trigger
746 	 * a reset.
747 	 */
748 	outb(0xcf9, 0x2);
749 	outb(0xcf9, 0x6);
750 	DELAY(500000);  /* wait 0.5 sec to see if that did it */
751 
752 	/*
753 	 * Attempt to force a reset via the Fast A20 and Init register
754 	 * at I/O port 0x92.  Bit 1 serves as an alternate A20 gate.
755 	 * Bit 0 asserts INIT# when set to 1.  We are careful to only
756 	 * preserve bit 1 while setting bit 0.  We also must clear bit
757 	 * 0 before setting it if it isn't already clear.
758 	 */
759 	b = inb(0x92);
760 	if (b != 0xff) {
761 		if ((b & 0x1) != 0)
762 			outb(0x92, b & 0xfe);
763 		outb(0x92, b | 0x1);
764 		DELAY(500000);  /* wait 0.5 sec to see if that did it */
765 	}
766 #endif /* PC98 */
767 
768 	printf("No known reset method worked, attempting CPU shutdown\n");
769 	DELAY(1000000); /* wait 1 sec for printf to complete */
770 
771 	/* Wipe the IDT. */
772 	null_idt.rd_limit = 0;
773 	null_idt.rd_base = 0;
774 	lidt(&null_idt);
775 
776 	/* "good night, sweet prince .... <THUNK!>" */
777 	breakpoint();
778 
779 	/* NOTREACHED */
780 	while(1);
781 }
782 
783 /*
784  * Get an sf_buf from the freelist.  May block if none are available.
785  */
786 void
787 sf_buf_map(struct sf_buf *sf, int flags)
788 {
789 	pt_entry_t opte, *ptep;
790 
791 	/*
792 	 * Update the sf_buf's virtual-to-physical mapping, flushing the
793 	 * virtual address from the TLB.  Since the reference count for
794 	 * the sf_buf's old mapping was zero, that mapping is not
795 	 * currently in use.  Consequently, there is no need to exchange
796 	 * the old and new PTEs atomically, even under PAE.
797 	 */
798 	ptep = vtopte(sf->kva);
799 	opte = *ptep;
800 #ifdef XEN
801        PT_SET_MA(sf->kva, xpmap_ptom(VM_PAGE_TO_PHYS(sf->m)) | pgeflag
802 	   | PG_RW | PG_V | pmap_cache_bits(sf->m->md.pat_mode, 0));
803 #else
804 	*ptep = VM_PAGE_TO_PHYS(sf->m) | pgeflag | PG_RW | PG_V |
805 	    pmap_cache_bits(sf->m->md.pat_mode, 0);
806 #endif
807 
808 	/*
809 	 * Avoid unnecessary TLB invalidations: If the sf_buf's old
810 	 * virtual-to-physical mapping was not used, then any processor
811 	 * that has invalidated the sf_buf's virtual address from its TLB
812 	 * since the last used mapping need not invalidate again.
813 	 */
814 #ifdef SMP
815 	if ((opte & (PG_V | PG_A)) ==  (PG_V | PG_A))
816 		CPU_ZERO(&sf->cpumask);
817 
818 	sf_buf_shootdown(sf, flags);
819 #else
820 	if ((opte & (PG_V | PG_A)) ==  (PG_V | PG_A))
821 		pmap_invalidate_page(kernel_pmap, sf->kva);
822 #endif
823 }
824 
825 #ifdef SMP
826 void
827 sf_buf_shootdown(struct sf_buf *sf, int flags)
828 {
829 	cpuset_t other_cpus;
830 	u_int cpuid;
831 
832 	sched_pin();
833 	cpuid = PCPU_GET(cpuid);
834 	if (!CPU_ISSET(cpuid, &sf->cpumask)) {
835 		CPU_SET(cpuid, &sf->cpumask);
836 		invlpg(sf->kva);
837 	}
838 	if ((flags & SFB_CPUPRIVATE) == 0) {
839 		other_cpus = all_cpus;
840 		CPU_CLR(cpuid, &other_cpus);
841 		CPU_NAND(&other_cpus, &sf->cpumask);
842 		if (!CPU_EMPTY(&other_cpus)) {
843 			CPU_OR(&sf->cpumask, &other_cpus);
844 			smp_masked_invlpg(other_cpus, sf->kva);
845 		}
846 	}
847 	sched_unpin();
848 }
849 #endif
850 
851 /*
852  * MD part of sf_buf_free().
853  */
854 int
855 sf_buf_unmap(struct sf_buf *sf)
856 {
857 #ifdef XEN
858 	/*
859 	 * Xen doesn't like having dangling R/W mappings
860 	 */
861 	pmap_qremove(sf->kva, 1);
862 	return (1);
863 #else
864 	return (0);
865 #endif
866 }
867 
868 static void
869 sf_buf_invalidate(struct sf_buf *sf)
870 {
871 	vm_page_t m = sf->m;
872 
873 	/*
874 	 * Use pmap_qenter to update the pte for
875 	 * existing mapping, in particular, the PAT
876 	 * settings are recalculated.
877 	 */
878 	pmap_qenter(sf->kva, &m, 1);
879 	pmap_invalidate_cache_range(sf->kva, sf->kva + PAGE_SIZE, FALSE);
880 }
881 
882 /*
883  * Invalidate the cache lines that may belong to the page, if
884  * (possibly old) mapping of the page by sf buffer exists.  Returns
885  * TRUE when mapping was found and cache invalidated.
886  */
887 boolean_t
888 sf_buf_invalidate_cache(vm_page_t m)
889 {
890 
891 	return (sf_buf_process_page(m, sf_buf_invalidate));
892 }
893 
894 /*
895  * Software interrupt handler for queued VM system processing.
896  */
897 void
898 swi_vm(void *dummy)
899 {
900 	if (busdma_swi_pending != 0)
901 		busdma_swi();
902 }
903 
904 /*
905  * Tell whether this address is in some physical memory region.
906  * Currently used by the kernel coredump code in order to avoid
907  * dumping the ``ISA memory hole'' which could cause indefinite hangs,
908  * or other unpredictable behaviour.
909  */
910 
911 int
912 is_physical_memory(vm_paddr_t addr)
913 {
914 
915 #ifdef DEV_ISA
916 	/* The ISA ``memory hole''. */
917 	if (addr >= 0xa0000 && addr < 0x100000)
918 		return 0;
919 #endif
920 
921 	/*
922 	 * stuff other tests for known memory-mapped devices (PCI?)
923 	 * here
924 	 */
925 
926 	return 1;
927 }
928