xref: /freebsd/sys/i386/i386/vm_machdep.c (revision 23f282aa31e9b6fceacd449020e936e98d6f2298)
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  * $FreeBSD$
42  */
43 
44 #include "npx.h"
45 #include "opt_user_ldt.h"
46 #ifdef PC98
47 #include "opt_pc98.h"
48 #endif
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/malloc.h>
53 #include <sys/proc.h>
54 #include <sys/buf.h>
55 #include <sys/vnode.h>
56 #include <sys/vmmeter.h>
57 #include <sys/kernel.h>
58 #include <sys/sysctl.h>
59 #include <sys/unistd.h>
60 
61 #include <machine/clock.h>
62 #include <machine/cpu.h>
63 #include <machine/md_var.h>
64 #ifdef SMP
65 #include <machine/smp.h>
66 #endif
67 #include <machine/pcb.h>
68 #include <machine/pcb_ext.h>
69 #include <machine/vm86.h>
70 
71 #include <vm/vm.h>
72 #include <vm/vm_param.h>
73 #include <sys/lock.h>
74 #include <vm/vm_kern.h>
75 #include <vm/vm_page.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_extern.h>
78 
79 #include <sys/user.h>
80 
81 #ifdef PC98
82 #include <pc98/pc98/pc98.h>
83 #else
84 #include <i386/isa/isa.h>
85 #endif
86 
87 static void	cpu_reset_real __P((void));
88 #ifdef SMP
89 static void	cpu_reset_proxy __P((void));
90 static u_int	cpu_reset_proxyid;
91 static volatile u_int	cpu_reset_proxy_active;
92 #endif
93 
94 /*
95  * quick version of vm_fault
96  */
97 int
98 vm_fault_quick(v, prot)
99 	caddr_t v;
100 	int prot;
101 {
102 	int r;
103 
104 	if (prot & VM_PROT_WRITE)
105 		r = subyte(v, fubyte(v));
106 	else
107 		r = fubyte(v);
108 	return(r);
109 }
110 
111 /*
112  * Finish a fork operation, with process p2 nearly set up.
113  * Copy and update the pcb, set up the stack so that the child
114  * ready to run and return to user mode.
115  */
116 void
117 cpu_fork(p1, p2, flags)
118 	register struct proc *p1, *p2;
119 	int flags;
120 {
121 	struct pcb *pcb2;
122 
123 	if ((flags & RFPROC) == 0) {
124 #ifdef USER_LDT
125 		if ((flags & RFMEM) == 0) {
126 			/* unshare user LDT */
127 			struct pcb *pcb1 = &p1->p_addr->u_pcb;
128 			struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt;
129 			if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) {
130 				pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len);
131 				user_ldt_free(pcb1);
132 				pcb1->pcb_ldt = pcb_ldt;
133 				set_user_ldt(pcb1);
134 			}
135 		}
136 #endif
137 		return;
138 	}
139 
140 #if NNPX > 0
141 	/* Ensure that p1's pcb is up to date. */
142 	if (npxproc == p1)
143 		npxsave(&p1->p_addr->u_pcb.pcb_savefpu);
144 #endif
145 
146 	/* Copy p1's pcb. */
147 	p2->p_addr->u_pcb = p1->p_addr->u_pcb;
148 	pcb2 = &p2->p_addr->u_pcb;
149 
150 	/*
151 	 * Create a new fresh stack for the new process.
152 	 * Copy the trap frame for the return to user mode as if from a
153 	 * syscall.  This copies the user mode register values.
154 	 */
155 	p2->p_md.md_regs = (struct trapframe *)
156 			   ((int)p2->p_addr + UPAGES * PAGE_SIZE - 16) - 1;
157 	*p2->p_md.md_regs = *p1->p_md.md_regs;
158 
159 	/*
160 	 * Set registers for trampoline to user mode.  Leave space for the
161 	 * return address on stack.  These are the kernel mode register values.
162 	 */
163 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
164 	pcb2->pcb_edi = p2->p_md.md_regs->tf_edi;
165 	pcb2->pcb_esi = (int)fork_return;
166 	pcb2->pcb_ebp = p2->p_md.md_regs->tf_ebp;
167 	pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *);
168 	pcb2->pcb_ebx = (int)p2;
169 	pcb2->pcb_eip = (int)fork_trampoline;
170 	/*
171 	 * pcb2->pcb_ldt:	duplicated below, if necessary.
172 	 * pcb2->pcb_savefpu:	cloned above.
173 	 * pcb2->pcb_flags:	cloned above (always 0 here?).
174 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
175 	 */
176 
177 #ifdef SMP
178 	pcb2->pcb_mpnest = 1;
179 #endif
180 	/*
181 	 * XXX don't copy the i/o pages.  this should probably be fixed.
182 	 */
183 	pcb2->pcb_ext = 0;
184 
185 #ifdef USER_LDT
186         /* Copy the LDT, if necessary. */
187         if (pcb2->pcb_ldt != 0) {
188 		if (flags & RFMEM) {
189 			pcb2->pcb_ldt->ldt_refcnt++;
190 		} else {
191 			pcb2->pcb_ldt = user_ldt_alloc(pcb2,
192 				pcb2->pcb_ldt->ldt_len);
193 		}
194         }
195 #endif
196 
197 	/*
198 	 * Now, cpu_switch() can schedule the new process.
199 	 * pcb_esp is loaded pointing to the cpu_switch() stack frame
200 	 * containing the return address when exiting cpu_switch.
201 	 * This will normally be to proc_trampoline(), which will have
202 	 * %ebx loaded with the new proc's pointer.  proc_trampoline()
203 	 * will set up a stack to call fork_return(p, frame); to complete
204 	 * the return to user-mode.
205 	 */
206 }
207 
208 /*
209  * Intercept the return address from a freshly forked process that has NOT
210  * been scheduled yet.
211  *
212  * This is needed to make kernel threads stay in kernel mode.
213  */
214 void
215 cpu_set_fork_handler(p, func, arg)
216 	struct proc *p;
217 	void (*func) __P((void *));
218 	void *arg;
219 {
220 	/*
221 	 * Note that the trap frame follows the args, so the function
222 	 * is really called like this:  func(arg, frame);
223 	 */
224 	p->p_addr->u_pcb.pcb_esi = (int) func;	/* function */
225 	p->p_addr->u_pcb.pcb_ebx = (int) arg;	/* first arg */
226 }
227 
228 void
229 cpu_exit(p)
230 	register struct proc *p;
231 {
232 	struct pcb *pcb = &p->p_addr->u_pcb;
233 
234 #if NNPX > 0
235 	npxexit(p);
236 #endif	/* NNPX */
237 	if (pcb->pcb_ext != 0) {
238 	        /*
239 		 * XXX do we need to move the TSS off the allocated pages
240 		 * before freeing them?  (not done here)
241 		 */
242 		kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
243 		    ctob(IOPAGES + 1));
244 		pcb->pcb_ext = 0;
245 	}
246 #ifdef USER_LDT
247 	user_ldt_free(pcb);
248 #endif
249         if (pcb->pcb_flags & PCB_DBREGS) {
250                 /*
251                  * disable all hardware breakpoints
252                  */
253                 reset_dbregs();
254                 pcb->pcb_flags &= ~PCB_DBREGS;
255         }
256 	cnt.v_swtch++;
257 	cpu_switch(p);
258 	panic("cpu_exit");
259 }
260 
261 void
262 cpu_wait(p)
263 	struct proc *p;
264 {
265 	/* drop per-process resources */
266 	pmap_dispose_proc(p);
267 
268 	/* and clean-out the vmspace */
269 	vmspace_free(p->p_vmspace);
270 }
271 
272 /*
273  * Dump the machine specific header information at the start of a core dump.
274  */
275 int
276 cpu_coredump(p, vp, cred)
277 	struct proc *p;
278 	struct vnode *vp;
279 	struct ucred *cred;
280 {
281 	int error;
282 	caddr_t tempuser;
283 
284 	tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK);
285 	if (!tempuser)
286 		return EINVAL;
287 
288 	bzero(tempuser, ctob(UPAGES));
289 	bcopy(p->p_addr, tempuser, sizeof(struct user));
290 	bcopy(p->p_md.md_regs,
291 	      tempuser + ((caddr_t) p->p_md.md_regs - (caddr_t) p->p_addr),
292 	      sizeof(struct trapframe));
293 
294 	error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser,
295 			ctob(UPAGES),
296 			(off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT,
297 			cred, (int *)NULL, p);
298 
299 	free(tempuser, M_TEMP);
300 
301 	return error;
302 }
303 
304 #ifdef notyet
305 static void
306 setredzone(pte, vaddr)
307 	u_short *pte;
308 	caddr_t vaddr;
309 {
310 /* eventually do this by setting up an expand-down stack segment
311    for ss0: selector, allowing stack access down to top of u.
312    this means though that protection violations need to be handled
313    thru a double fault exception that must do an integral task
314    switch to a known good context, within which a dump can be
315    taken. a sensible scheme might be to save the initial context
316    used by sched (that has physical memory mapped 1:1 at bottom)
317    and take the dump while still in mapped mode */
318 }
319 #endif
320 
321 /*
322  * Convert kernel VA to physical address
323  */
324 u_long
325 kvtop(void *addr)
326 {
327 	vm_offset_t va;
328 
329 	va = pmap_kextract((vm_offset_t)addr);
330 	if (va == 0)
331 		panic("kvtop: zero page frame");
332 	return((int)va);
333 }
334 
335 /*
336  * Map an IO request into kernel virtual address space.
337  *
338  * All requests are (re)mapped into kernel VA space.
339  * Notice that we use b_bufsize for the size of the buffer
340  * to be mapped.  b_bcount might be modified by the driver.
341  */
342 void
343 vmapbuf(bp)
344 	register struct buf *bp;
345 {
346 	register caddr_t addr, v, kva;
347 	vm_offset_t pa;
348 
349 	if ((bp->b_flags & B_PHYS) == 0)
350 		panic("vmapbuf");
351 
352 	for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data);
353 	    addr < bp->b_data + bp->b_bufsize;
354 	    addr += PAGE_SIZE, v += PAGE_SIZE) {
355 		/*
356 		 * Do the vm_fault if needed; do the copy-on-write thing
357 		 * when reading stuff off device into memory.
358 		 */
359 		vm_fault_quick(addr,
360 			(bp->b_iocmd == BIO_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ);
361 		pa = trunc_page(pmap_kextract((vm_offset_t) addr));
362 		if (pa == 0)
363 			panic("vmapbuf: page not present");
364 		vm_page_hold(PHYS_TO_VM_PAGE(pa));
365 		pmap_kenter((vm_offset_t) v, pa);
366 	}
367 
368 	kva = bp->b_saveaddr;
369 	bp->b_saveaddr = bp->b_data;
370 	bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK);
371 }
372 
373 /*
374  * Free the io map PTEs associated with this IO operation.
375  * We also invalidate the TLB entries and restore the original b_addr.
376  */
377 void
378 vunmapbuf(bp)
379 	register struct buf *bp;
380 {
381 	register caddr_t addr;
382 	vm_offset_t pa;
383 
384 	if ((bp->b_flags & B_PHYS) == 0)
385 		panic("vunmapbuf");
386 
387 	for (addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data);
388 	    addr < bp->b_data + bp->b_bufsize;
389 	    addr += PAGE_SIZE) {
390 		pa = trunc_page(pmap_kextract((vm_offset_t) addr));
391 		pmap_kremove((vm_offset_t) addr);
392 		vm_page_unhold(PHYS_TO_VM_PAGE(pa));
393 	}
394 
395 	bp->b_data = bp->b_saveaddr;
396 }
397 
398 /*
399  * Force reset the processor by invalidating the entire address space!
400  */
401 
402 #ifdef SMP
403 static void
404 cpu_reset_proxy()
405 {
406 	u_int saved_mp_lock;
407 
408 	cpu_reset_proxy_active = 1;
409 	while (cpu_reset_proxy_active == 1)
410 		;	 /* Wait for other cpu to disable interupts */
411 	saved_mp_lock = mp_lock;
412 	mp_lock = 1;
413 	printf("cpu_reset_proxy: Grabbed mp lock for BSP\n");
414 	cpu_reset_proxy_active = 3;
415 	while (cpu_reset_proxy_active == 3)
416 		;	/* Wait for other cpu to enable interrupts */
417 	stop_cpus((1<<cpu_reset_proxyid));
418 	printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
419 	DELAY(1000000);
420 	cpu_reset_real();
421 }
422 #endif
423 
424 void
425 cpu_reset()
426 {
427 #ifdef SMP
428 	if (smp_active == 0) {
429 		cpu_reset_real();
430 		/* NOTREACHED */
431 	} else {
432 
433 		u_int map;
434 		int cnt;
435 		printf("cpu_reset called on cpu#%d\n",cpuid);
436 
437 		map = other_cpus & ~ stopped_cpus;
438 
439 		if (map != 0) {
440 			printf("cpu_reset: Stopping other CPUs\n");
441 			stop_cpus(map);		/* Stop all other CPUs */
442 		}
443 
444 		if (cpuid == 0) {
445 			DELAY(1000000);
446 			cpu_reset_real();
447 			/* NOTREACHED */
448 		} else {
449 			/* We are not BSP (CPU #0) */
450 
451 			cpu_reset_proxyid = cpuid;
452 			cpustop_restartfunc = cpu_reset_proxy;
453 			printf("cpu_reset: Restarting BSP\n");
454 			started_cpus = (1<<0);		/* Restart CPU #0 */
455 
456 			cnt = 0;
457 			while (cpu_reset_proxy_active == 0 && cnt < 10000000)
458 				cnt++;	/* Wait for BSP to announce restart */
459 			if (cpu_reset_proxy_active == 0)
460 				printf("cpu_reset: Failed to restart BSP\n");
461 			__asm __volatile("cli" : : : "memory");
462 			cpu_reset_proxy_active = 2;
463 			cnt = 0;
464 			while (cpu_reset_proxy_active == 2 && cnt < 10000000)
465 				cnt++;	/* Do nothing */
466 			if (cpu_reset_proxy_active == 2) {
467 				printf("cpu_reset: BSP did not grab mp lock\n");
468 				cpu_reset_real();	/* XXX: Bogus ? */
469 			}
470 			cpu_reset_proxy_active = 4;
471 			__asm __volatile("sti" : : : "memory");
472 			while (1);
473 			/* NOTREACHED */
474 		}
475 	}
476 #else
477 	cpu_reset_real();
478 #endif
479 }
480 
481 static void
482 cpu_reset_real()
483 {
484 
485 #ifdef PC98
486 	/*
487 	 * Attempt to do a CPU reset via CPU reset port.
488 	 */
489 	disable_intr();
490 	if ((inb(0x35) & 0xa0) != 0xa0) {
491 		outb(0x37, 0x0f);		/* SHUT0 = 0. */
492 		outb(0x37, 0x0b);		/* SHUT1 = 0. */
493 	}
494 	outb(0xf0, 0x00);		/* Reset. */
495 #else
496 	/*
497 	 * Attempt to do a CPU reset via the keyboard controller,
498 	 * do not turn of the GateA20, as any machine that fails
499 	 * to do the reset here would then end up in no man's land.
500 	 */
501 
502 #if !defined(BROKEN_KEYBOARD_RESET)
503 	outb(IO_KBD + 4, 0xFE);
504 	DELAY(500000);	/* wait 0.5 sec to see if that did it */
505 	printf("Keyboard reset did not work, attempting CPU shutdown\n");
506 	DELAY(1000000);	/* wait 1 sec for printf to complete */
507 #endif
508 #endif /* PC98 */
509 	/* force a shutdown by unmapping entire address space ! */
510 	bzero((caddr_t) PTD, PAGE_SIZE);
511 
512 	/* "good night, sweet prince .... <THUNK!>" */
513 	invltlb();
514 	/* NOTREACHED */
515 	while(1);
516 }
517 
518 int
519 grow_stack(p, sp)
520 	struct proc *p;
521 	u_int sp;
522 {
523 	int rv;
524 
525 	rv = vm_map_growstack (p, sp);
526 	if (rv != KERN_SUCCESS)
527 		return (0);
528 
529 	return (1);
530 }
531 
532 SYSCTL_DECL(_vm_stats_misc);
533 
534 static int cnt_prezero;
535 
536 SYSCTL_INT(_vm_stats_misc, OID_AUTO,
537 	cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, "");
538 
539 /*
540  * Implement the pre-zeroed page mechanism.
541  * This routine is called from the idle loop.
542  */
543 
544 #define ZIDLE_LO(v)	((v) * 2 / 3)
545 #define ZIDLE_HI(v)	((v) * 4 / 5)
546 
547 int
548 vm_page_zero_idle()
549 {
550 	static int free_rover;
551 	static int zero_state;
552 	vm_page_t m;
553 	int s;
554 
555 	/*
556 	 * Attempt to maintain approximately 1/2 of our free pages in a
557 	 * PG_ZERO'd state.   Add some hysteresis to (attempt to) avoid
558 	 * generally zeroing a page when the system is near steady-state.
559 	 * Otherwise we might get 'flutter' during disk I/O / IPC or
560 	 * fast sleeps.  We also do not want to be continuously zeroing
561 	 * pages because doing so may flush our L1 and L2 caches too much.
562 	 */
563 
564 	if (zero_state && vm_page_zero_count >= ZIDLE_LO(cnt.v_free_count))
565 		return(0);
566 	if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count))
567 		return(0);
568 
569 #ifdef SMP
570 	if (try_mplock()) {
571 #endif
572 		s = splvm();
573 		__asm __volatile("sti" : : : "memory");
574 		zero_state = 0;
575 		m = vm_page_list_find(PQ_FREE, free_rover, FALSE);
576 		if (m != NULL && (m->flags & PG_ZERO) == 0) {
577 			vm_page_queues[m->queue].lcnt--;
578 			TAILQ_REMOVE(&vm_page_queues[m->queue].pl, m, pageq);
579 			m->queue = PQ_NONE;
580 			splx(s);
581 			pmap_zero_page(VM_PAGE_TO_PHYS(m));
582 			(void)splvm();
583 			vm_page_flag_set(m, PG_ZERO);
584 			m->queue = PQ_FREE + m->pc;
585 			vm_page_queues[m->queue].lcnt++;
586 			TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m,
587 			    pageq);
588 			++vm_page_zero_count;
589 			++cnt_prezero;
590 			if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count))
591 				zero_state = 1;
592 		}
593 		free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK;
594 		splx(s);
595 		__asm __volatile("cli" : : : "memory");
596 #ifdef SMP
597 		rel_mplock();
598 #endif
599 		return (1);
600 #ifdef SMP
601 	}
602 #endif
603 	/*
604 	 * We have to enable interrupts for a moment if the try_mplock fails
605 	 * in order to potentially take an IPI.   XXX this should be in
606 	 * swtch.s
607 	 */
608 	__asm __volatile("sti; nop; cli" : : : "memory");
609 	return (0);
610 }
611 
612 /*
613  * Software interrupt handler for queued VM system processing.
614  */
615 void
616 swi_vm()
617 {
618 	if (busdma_swi_pending != 0)
619 		busdma_swi();
620 }
621 
622 /*
623  * Tell whether this address is in some physical memory region.
624  * Currently used by the kernel coredump code in order to avoid
625  * dumping the ``ISA memory hole'' which could cause indefinite hangs,
626  * or other unpredictable behaviour.
627  */
628 
629 #include "isa.h"
630 
631 int
632 is_physical_memory(addr)
633 	vm_offset_t addr;
634 {
635 
636 #if NISA > 0
637 	/* The ISA ``memory hole''. */
638 	if (addr >= 0xa0000 && addr < 0x100000)
639 		return 0;
640 #endif
641 
642 	/*
643 	 * stuff other tests for known memory-mapped devices (PCI?)
644 	 * here
645 	 */
646 
647 	return 1;
648 }
649