xref: /freebsd/sys/i386/i386/vm_machdep.c (revision b601c69bdbe8755d26570261d7fd4c02ee4eff74)
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 #include "opt_reset.h"
50 
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/malloc.h>
54 #include <sys/proc.h>
55 #include <sys/bio.h>
56 #include <sys/buf.h>
57 #include <sys/vnode.h>
58 #include <sys/vmmeter.h>
59 #include <sys/kernel.h>
60 #include <sys/sysctl.h>
61 #include <sys/unistd.h>
62 
63 #include <machine/clock.h>
64 #include <machine/cpu.h>
65 #include <machine/md_var.h>
66 #ifdef SMP
67 #include <machine/smp.h>
68 #endif
69 #include <machine/pcb.h>
70 #include <machine/pcb_ext.h>
71 #include <machine/vm86.h>
72 
73 #include <vm/vm.h>
74 #include <vm/vm_param.h>
75 #include <sys/lock.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_page.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_extern.h>
80 
81 #include <sys/user.h>
82 
83 #ifdef PC98
84 #include <pc98/pc98/pc98.h>
85 #else
86 #include <i386/isa/isa.h>
87 #endif
88 
89 static void	cpu_reset_real __P((void));
90 #ifdef SMP
91 static void	cpu_reset_proxy __P((void));
92 static u_int	cpu_reset_proxyid;
93 static volatile u_int	cpu_reset_proxy_active;
94 #endif
95 
96 /*
97  * quick version of vm_fault
98  */
99 int
100 vm_fault_quick(v, prot)
101 	caddr_t v;
102 	int prot;
103 {
104 	int r;
105 
106 	if (prot & VM_PROT_WRITE)
107 		r = subyte(v, fubyte(v));
108 	else
109 		r = fubyte(v);
110 	return(r);
111 }
112 
113 /*
114  * Finish a fork operation, with process p2 nearly set up.
115  * Copy and update the pcb, set up the stack so that the child
116  * ready to run and return to user mode.
117  */
118 void
119 cpu_fork(p1, p2, flags)
120 	register struct proc *p1, *p2;
121 	int flags;
122 {
123 	struct pcb *pcb2;
124 
125 	if ((flags & RFPROC) == 0) {
126 #ifdef USER_LDT
127 		if ((flags & RFMEM) == 0) {
128 			/* unshare user LDT */
129 			struct pcb *pcb1 = &p1->p_addr->u_pcb;
130 			struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt;
131 			if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) {
132 				pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len);
133 				user_ldt_free(pcb1);
134 				pcb1->pcb_ldt = pcb_ldt;
135 				set_user_ldt(pcb1);
136 			}
137 		}
138 #endif
139 		return;
140 	}
141 
142 #if NNPX > 0
143 	/* Ensure that p1's pcb is up to date. */
144 	if (npxproc == p1)
145 		npxsave(&p1->p_addr->u_pcb.pcb_savefpu);
146 #endif
147 
148 	/* Copy p1's pcb. */
149 	p2->p_addr->u_pcb = p1->p_addr->u_pcb;
150 	pcb2 = &p2->p_addr->u_pcb;
151 
152 	/*
153 	 * Create a new fresh stack for the new process.
154 	 * Copy the trap frame for the return to user mode as if from a
155 	 * syscall.  This copies the user mode register values.
156 	 */
157 	p2->p_md.md_regs = (struct trapframe *)
158 			   ((int)p2->p_addr + UPAGES * PAGE_SIZE - 16) - 1;
159 	*p2->p_md.md_regs = *p1->p_md.md_regs;
160 
161 	/*
162 	 * Set registers for trampoline to user mode.  Leave space for the
163 	 * return address on stack.  These are the kernel mode register values.
164 	 */
165 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
166 	pcb2->pcb_edi = p2->p_md.md_regs->tf_edi;
167 	pcb2->pcb_esi = (int)fork_return;
168 	pcb2->pcb_ebp = p2->p_md.md_regs->tf_ebp;
169 	pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *);
170 	pcb2->pcb_ebx = (int)p2;
171 	pcb2->pcb_eip = (int)fork_trampoline;
172 	/*
173 	 * pcb2->pcb_ldt:	duplicated below, if necessary.
174 	 * pcb2->pcb_savefpu:	cloned above.
175 	 * pcb2->pcb_flags:	cloned above (always 0 here?).
176 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
177 	 */
178 
179 #ifdef SMP
180 	pcb2->pcb_mpnest = 1;
181 #endif
182 	/*
183 	 * XXX don't copy the i/o pages.  this should probably be fixed.
184 	 */
185 	pcb2->pcb_ext = 0;
186 
187 #ifdef USER_LDT
188         /* Copy the LDT, if necessary. */
189         if (pcb2->pcb_ldt != 0) {
190 		if (flags & RFMEM) {
191 			pcb2->pcb_ldt->ldt_refcnt++;
192 		} else {
193 			pcb2->pcb_ldt = user_ldt_alloc(pcb2,
194 				pcb2->pcb_ldt->ldt_len);
195 		}
196         }
197 #endif
198 
199 	/*
200 	 * Now, cpu_switch() can schedule the new process.
201 	 * pcb_esp is loaded pointing to the cpu_switch() stack frame
202 	 * containing the return address when exiting cpu_switch.
203 	 * This will normally be to proc_trampoline(), which will have
204 	 * %ebx loaded with the new proc's pointer.  proc_trampoline()
205 	 * will set up a stack to call fork_return(p, frame); to complete
206 	 * the return to user-mode.
207 	 */
208 }
209 
210 /*
211  * Intercept the return address from a freshly forked process that has NOT
212  * been scheduled yet.
213  *
214  * This is needed to make kernel threads stay in kernel mode.
215  */
216 void
217 cpu_set_fork_handler(p, func, arg)
218 	struct proc *p;
219 	void (*func) __P((void *));
220 	void *arg;
221 {
222 	/*
223 	 * Note that the trap frame follows the args, so the function
224 	 * is really called like this:  func(arg, frame);
225 	 */
226 	p->p_addr->u_pcb.pcb_esi = (int) func;	/* function */
227 	p->p_addr->u_pcb.pcb_ebx = (int) arg;	/* first arg */
228 }
229 
230 void
231 cpu_exit(p)
232 	register struct proc *p;
233 {
234 	struct pcb *pcb = &p->p_addr->u_pcb;
235 
236 #if NNPX > 0
237 	npxexit(p);
238 #endif	/* NNPX */
239 	if (pcb->pcb_ext != 0) {
240 	        /*
241 		 * XXX do we need to move the TSS off the allocated pages
242 		 * before freeing them?  (not done here)
243 		 */
244 		kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
245 		    ctob(IOPAGES + 1));
246 		pcb->pcb_ext = 0;
247 	}
248 #ifdef USER_LDT
249 	user_ldt_free(pcb);
250 #endif
251         if (pcb->pcb_flags & PCB_DBREGS) {
252                 /*
253                  * disable all hardware breakpoints
254                  */
255                 reset_dbregs();
256                 pcb->pcb_flags &= ~PCB_DBREGS;
257         }
258 	cnt.v_swtch++;
259 	cpu_switch(p);
260 	panic("cpu_exit");
261 }
262 
263 void
264 cpu_wait(p)
265 	struct proc *p;
266 {
267 	/* drop per-process resources */
268 	pmap_dispose_proc(p);
269 
270 	/* and clean-out the vmspace */
271 	vmspace_free(p->p_vmspace);
272 }
273 
274 /*
275  * Dump the machine specific header information at the start of a core dump.
276  */
277 int
278 cpu_coredump(p, vp, cred)
279 	struct proc *p;
280 	struct vnode *vp;
281 	struct ucred *cred;
282 {
283 	int error;
284 	caddr_t tempuser;
285 
286 	tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK);
287 	if (!tempuser)
288 		return EINVAL;
289 
290 	bzero(tempuser, ctob(UPAGES));
291 	bcopy(p->p_addr, tempuser, sizeof(struct user));
292 	bcopy(p->p_md.md_regs,
293 	      tempuser + ((caddr_t) p->p_md.md_regs - (caddr_t) p->p_addr),
294 	      sizeof(struct trapframe));
295 
296 	error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser,
297 			ctob(UPAGES),
298 			(off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT,
299 			cred, (int *)NULL, p);
300 
301 	free(tempuser, M_TEMP);
302 
303 	return error;
304 }
305 
306 #ifdef notyet
307 static void
308 setredzone(pte, vaddr)
309 	u_short *pte;
310 	caddr_t vaddr;
311 {
312 /* eventually do this by setting up an expand-down stack segment
313    for ss0: selector, allowing stack access down to top of u.
314    this means though that protection violations need to be handled
315    thru a double fault exception that must do an integral task
316    switch to a known good context, within which a dump can be
317    taken. a sensible scheme might be to save the initial context
318    used by sched (that has physical memory mapped 1:1 at bottom)
319    and take the dump while still in mapped mode */
320 }
321 #endif
322 
323 /*
324  * Convert kernel VA to physical address
325  */
326 u_long
327 kvtop(void *addr)
328 {
329 	vm_offset_t va;
330 
331 	va = pmap_kextract((vm_offset_t)addr);
332 	if (va == 0)
333 		panic("kvtop: zero page frame");
334 	return((int)va);
335 }
336 
337 /*
338  * Map an IO request into kernel virtual address space.
339  *
340  * All requests are (re)mapped into kernel VA space.
341  * Notice that we use b_bufsize for the size of the buffer
342  * to be mapped.  b_bcount might be modified by the driver.
343  */
344 void
345 vmapbuf(bp)
346 	register struct buf *bp;
347 {
348 	register caddr_t addr, v, kva;
349 	vm_offset_t pa;
350 
351 	if ((bp->b_flags & B_PHYS) == 0)
352 		panic("vmapbuf");
353 
354 	for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data);
355 	    addr < bp->b_data + bp->b_bufsize;
356 	    addr += PAGE_SIZE, v += PAGE_SIZE) {
357 		/*
358 		 * Do the vm_fault if needed; do the copy-on-write thing
359 		 * when reading stuff off device into memory.
360 		 */
361 		vm_fault_quick(addr,
362 			(bp->b_iocmd == BIO_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ);
363 		pa = trunc_page(pmap_kextract((vm_offset_t) addr));
364 		if (pa == 0)
365 			panic("vmapbuf: page not present");
366 		vm_page_hold(PHYS_TO_VM_PAGE(pa));
367 		pmap_kenter((vm_offset_t) v, pa);
368 	}
369 
370 	kva = bp->b_saveaddr;
371 	bp->b_saveaddr = bp->b_data;
372 	bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK);
373 }
374 
375 /*
376  * Free the io map PTEs associated with this IO operation.
377  * We also invalidate the TLB entries and restore the original b_addr.
378  */
379 void
380 vunmapbuf(bp)
381 	register struct buf *bp;
382 {
383 	register caddr_t addr;
384 	vm_offset_t pa;
385 
386 	if ((bp->b_flags & B_PHYS) == 0)
387 		panic("vunmapbuf");
388 
389 	for (addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data);
390 	    addr < bp->b_data + bp->b_bufsize;
391 	    addr += PAGE_SIZE) {
392 		pa = trunc_page(pmap_kextract((vm_offset_t) addr));
393 		pmap_kremove((vm_offset_t) addr);
394 		vm_page_unhold(PHYS_TO_VM_PAGE(pa));
395 	}
396 
397 	bp->b_data = bp->b_saveaddr;
398 }
399 
400 /*
401  * Force reset the processor by invalidating the entire address space!
402  */
403 
404 #ifdef SMP
405 static void
406 cpu_reset_proxy()
407 {
408 	u_int saved_mp_lock;
409 
410 	cpu_reset_proxy_active = 1;
411 	while (cpu_reset_proxy_active == 1)
412 		;	 /* Wait for other cpu to disable interupts */
413 	saved_mp_lock = mp_lock;
414 	mp_lock = 1;
415 	printf("cpu_reset_proxy: Grabbed mp lock for BSP\n");
416 	cpu_reset_proxy_active = 3;
417 	while (cpu_reset_proxy_active == 3)
418 		;	/* Wait for other cpu to enable interrupts */
419 	stop_cpus((1<<cpu_reset_proxyid));
420 	printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
421 	DELAY(1000000);
422 	cpu_reset_real();
423 }
424 #endif
425 
426 void
427 cpu_reset()
428 {
429 #ifdef SMP
430 	if (smp_active == 0) {
431 		cpu_reset_real();
432 		/* NOTREACHED */
433 	} else {
434 
435 		u_int map;
436 		int cnt;
437 		printf("cpu_reset called on cpu#%d\n",cpuid);
438 
439 		map = other_cpus & ~ stopped_cpus;
440 
441 		if (map != 0) {
442 			printf("cpu_reset: Stopping other CPUs\n");
443 			stop_cpus(map);		/* Stop all other CPUs */
444 		}
445 
446 		if (cpuid == 0) {
447 			DELAY(1000000);
448 			cpu_reset_real();
449 			/* NOTREACHED */
450 		} else {
451 			/* We are not BSP (CPU #0) */
452 
453 			cpu_reset_proxyid = cpuid;
454 			cpustop_restartfunc = cpu_reset_proxy;
455 			printf("cpu_reset: Restarting BSP\n");
456 			started_cpus = (1<<0);		/* Restart CPU #0 */
457 
458 			cnt = 0;
459 			while (cpu_reset_proxy_active == 0 && cnt < 10000000)
460 				cnt++;	/* Wait for BSP to announce restart */
461 			if (cpu_reset_proxy_active == 0)
462 				printf("cpu_reset: Failed to restart BSP\n");
463 			__asm __volatile("cli" : : : "memory");
464 			cpu_reset_proxy_active = 2;
465 			cnt = 0;
466 			while (cpu_reset_proxy_active == 2 && cnt < 10000000)
467 				cnt++;	/* Do nothing */
468 			if (cpu_reset_proxy_active == 2) {
469 				printf("cpu_reset: BSP did not grab mp lock\n");
470 				cpu_reset_real();	/* XXX: Bogus ? */
471 			}
472 			cpu_reset_proxy_active = 4;
473 			__asm __volatile("sti" : : : "memory");
474 			while (1);
475 			/* NOTREACHED */
476 		}
477 	}
478 #else
479 	cpu_reset_real();
480 #endif
481 }
482 
483 static void
484 cpu_reset_real()
485 {
486 
487 #ifdef PC98
488 	/*
489 	 * Attempt to do a CPU reset via CPU reset port.
490 	 */
491 	disable_intr();
492 	if ((inb(0x35) & 0xa0) != 0xa0) {
493 		outb(0x37, 0x0f);		/* SHUT0 = 0. */
494 		outb(0x37, 0x0b);		/* SHUT1 = 0. */
495 	}
496 	outb(0xf0, 0x00);		/* Reset. */
497 #else
498 	/*
499 	 * Attempt to do a CPU reset via the keyboard controller,
500 	 * do not turn of the GateA20, as any machine that fails
501 	 * to do the reset here would then end up in no man's land.
502 	 */
503 
504 #if !defined(BROKEN_KEYBOARD_RESET)
505 	outb(IO_KBD + 4, 0xFE);
506 	DELAY(500000);	/* wait 0.5 sec to see if that did it */
507 	printf("Keyboard reset did not work, attempting CPU shutdown\n");
508 	DELAY(1000000);	/* wait 1 sec for printf to complete */
509 #endif
510 #endif /* PC98 */
511 	/* force a shutdown by unmapping entire address space ! */
512 	bzero((caddr_t) PTD, PAGE_SIZE);
513 
514 	/* "good night, sweet prince .... <THUNK!>" */
515 	invltlb();
516 	/* NOTREACHED */
517 	while(1);
518 }
519 
520 int
521 grow_stack(p, sp)
522 	struct proc *p;
523 	u_int sp;
524 {
525 	int rv;
526 
527 	rv = vm_map_growstack (p, sp);
528 	if (rv != KERN_SUCCESS)
529 		return (0);
530 
531 	return (1);
532 }
533 
534 SYSCTL_DECL(_vm_stats_misc);
535 
536 static int cnt_prezero;
537 
538 SYSCTL_INT(_vm_stats_misc, OID_AUTO,
539 	cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, "");
540 
541 /*
542  * Implement the pre-zeroed page mechanism.
543  * This routine is called from the idle loop.
544  */
545 
546 #define ZIDLE_LO(v)	((v) * 2 / 3)
547 #define ZIDLE_HI(v)	((v) * 4 / 5)
548 
549 int
550 vm_page_zero_idle()
551 {
552 	static int free_rover;
553 	static int zero_state;
554 	vm_page_t m;
555 	int s;
556 
557 	/*
558 	 * Attempt to maintain approximately 1/2 of our free pages in a
559 	 * PG_ZERO'd state.   Add some hysteresis to (attempt to) avoid
560 	 * generally zeroing a page when the system is near steady-state.
561 	 * Otherwise we might get 'flutter' during disk I/O / IPC or
562 	 * fast sleeps.  We also do not want to be continuously zeroing
563 	 * pages because doing so may flush our L1 and L2 caches too much.
564 	 */
565 
566 	if (zero_state && vm_page_zero_count >= ZIDLE_LO(cnt.v_free_count))
567 		return(0);
568 	if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count))
569 		return(0);
570 
571 #ifdef SMP
572 	if (try_mplock()) {
573 #endif
574 		s = splvm();
575 		__asm __volatile("sti" : : : "memory");
576 		zero_state = 0;
577 		m = vm_page_list_find(PQ_FREE, free_rover, FALSE);
578 		if (m != NULL && (m->flags & PG_ZERO) == 0) {
579 			vm_page_queues[m->queue].lcnt--;
580 			TAILQ_REMOVE(&vm_page_queues[m->queue].pl, m, pageq);
581 			m->queue = PQ_NONE;
582 			splx(s);
583 			pmap_zero_page(VM_PAGE_TO_PHYS(m));
584 			(void)splvm();
585 			vm_page_flag_set(m, PG_ZERO);
586 			m->queue = PQ_FREE + m->pc;
587 			vm_page_queues[m->queue].lcnt++;
588 			TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m,
589 			    pageq);
590 			++vm_page_zero_count;
591 			++cnt_prezero;
592 			if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count))
593 				zero_state = 1;
594 		}
595 		free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK;
596 		splx(s);
597 		__asm __volatile("cli" : : : "memory");
598 #ifdef SMP
599 		rel_mplock();
600 #endif
601 		return (1);
602 #ifdef SMP
603 	}
604 #endif
605 	/*
606 	 * We have to enable interrupts for a moment if the try_mplock fails
607 	 * in order to potentially take an IPI.   XXX this should be in
608 	 * swtch.s
609 	 */
610 	__asm __volatile("sti; nop; cli" : : : "memory");
611 	return (0);
612 }
613 
614 /*
615  * Software interrupt handler for queued VM system processing.
616  */
617 void
618 swi_vm()
619 {
620 	if (busdma_swi_pending != 0)
621 		busdma_swi();
622 }
623 
624 /*
625  * Tell whether this address is in some physical memory region.
626  * Currently used by the kernel coredump code in order to avoid
627  * dumping the ``ISA memory hole'' which could cause indefinite hangs,
628  * or other unpredictable behaviour.
629  */
630 
631 #include "isa.h"
632 
633 int
634 is_physical_memory(addr)
635 	vm_offset_t addr;
636 {
637 
638 #if NISA > 0
639 	/* The ISA ``memory hole''. */
640 	if (addr >= 0xa0000 && addr < 0x100000)
641 		return 0;
642 #endif
643 
644 	/*
645 	 * stuff other tests for known memory-mapped devices (PCI?)
646 	 * here
647 	 */
648 
649 	return 1;
650 }
651