Initial import from FreeBSD RELENG_4:
[dragonfly.git] / sys / i386 / i386 / vm_machdep.c
CommitLineData
984263bc
MD
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: src/sys/i386/i386/vm_machdep.c,v 1.132.2.9 2003/01/25 19:02:23 dillon Exp $
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/buf.h>
56#include <sys/vnode.h>
57#include <sys/vmmeter.h>
58#include <sys/kernel.h>
59#include <sys/sysctl.h>
60#include <sys/unistd.h>
61
62#include <machine/clock.h>
63#include <machine/cpu.h>
64#include <machine/md_var.h>
65#ifdef SMP
66#include <machine/smp.h>
67#endif
68#include <machine/pcb.h>
69#include <machine/pcb_ext.h>
70#include <machine/vm86.h>
71
72#include <vm/vm.h>
73#include <vm/vm_param.h>
74#include <sys/lock.h>
75#include <vm/vm_kern.h>
76#include <vm/vm_page.h>
77#include <vm/vm_map.h>
78#include <vm/vm_extern.h>
79
80#include <sys/user.h>
81
82#ifdef PC98
83#include <pc98/pc98/pc98.h>
84#else
85#include <i386/isa/isa.h>
86#endif
87
88static void cpu_reset_real __P((void));
89#ifdef SMP
90static void cpu_reset_proxy __P((void));
91static u_int cpu_reset_proxyid;
92static volatile u_int cpu_reset_proxy_active;
93#endif
94extern int _ucodesel, _udatasel;
95
96/*
97 * quick version of vm_fault
98 */
99int
100vm_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 */
118void
119cpu_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_save);
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 bcopy(p1->p_md.md_regs, p2->p_md.md_regs, sizeof(*p2->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 = 0;
167 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */
168 pcb2->pcb_ebp = 0;
169 pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *);
170 pcb2->pcb_ebx = (int)p2; /* fork_trampoline argument */
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 fork_trampoline(), which will have
204 * %ebx loaded with the new proc's pointer. fork_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 */
216void
217cpu_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
230void
231cpu_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
263void
264cpu_wait(p)
265 struct proc *p;
266{
267 /* drop per-process resources */
268 pmap_dispose_proc(p);
269}
270
271/*
272 * Dump the machine specific header information at the start of a core dump.
273 */
274int
275cpu_coredump(p, vp, cred)
276 struct proc *p;
277 struct vnode *vp;
278 struct ucred *cred;
279{
280 int error;
281 caddr_t tempuser;
282
283 tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK);
284 if (!tempuser)
285 return EINVAL;
286
287 bzero(tempuser, ctob(UPAGES));
288 bcopy(p->p_addr, tempuser, sizeof(struct user));
289 bcopy(p->p_md.md_regs,
290 tempuser + ((caddr_t) p->p_md.md_regs - (caddr_t) p->p_addr),
291 sizeof(struct trapframe));
292
293 error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, ctob(UPAGES),
294 (off_t)0, UIO_SYSSPACE, IO_UNIT, cred, (int *)NULL, p);
295
296 free(tempuser, M_TEMP);
297
298 return error;
299}
300
301#ifdef notyet
302static void
303setredzone(pte, vaddr)
304 u_short *pte;
305 caddr_t vaddr;
306{
307/* eventually do this by setting up an expand-down stack segment
308 for ss0: selector, allowing stack access down to top of u.
309 this means though that protection violations need to be handled
310 thru a double fault exception that must do an integral task
311 switch to a known good context, within which a dump can be
312 taken. a sensible scheme might be to save the initial context
313 used by sched (that has physical memory mapped 1:1 at bottom)
314 and take the dump while still in mapped mode */
315}
316#endif
317
318/*
319 * Convert kernel VA to physical address
320 */
321u_long
322kvtop(void *addr)
323{
324 vm_offset_t va;
325
326 va = pmap_kextract((vm_offset_t)addr);
327 if (va == 0)
328 panic("kvtop: zero page frame");
329 return((int)va);
330}
331
332/*
333 * Force reset the processor by invalidating the entire address space!
334 */
335
336#ifdef SMP
337static void
338cpu_reset_proxy()
339{
340 u_int saved_mp_lock;
341
342 cpu_reset_proxy_active = 1;
343 while (cpu_reset_proxy_active == 1)
344 ; /* Wait for other cpu to disable interupts */
345 saved_mp_lock = mp_lock;
346 mp_lock = 1;
347 printf("cpu_reset_proxy: Grabbed mp lock for BSP\n");
348 cpu_reset_proxy_active = 3;
349 while (cpu_reset_proxy_active == 3)
350 ; /* Wait for other cpu to enable interrupts */
351 stop_cpus((1<<cpu_reset_proxyid));
352 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
353 DELAY(1000000);
354 cpu_reset_real();
355}
356#endif
357
358void
359cpu_reset()
360{
361#ifdef SMP
362 if (smp_active == 0) {
363 cpu_reset_real();
364 /* NOTREACHED */
365 } else {
366
367 u_int map;
368 int cnt;
369 printf("cpu_reset called on cpu#%d\n",cpuid);
370
371 map = other_cpus & ~ stopped_cpus;
372
373 if (map != 0) {
374 printf("cpu_reset: Stopping other CPUs\n");
375 stop_cpus(map); /* Stop all other CPUs */
376 }
377
378 if (cpuid == 0) {
379 DELAY(1000000);
380 cpu_reset_real();
381 /* NOTREACHED */
382 } else {
383 /* We are not BSP (CPU #0) */
384
385 cpu_reset_proxyid = cpuid;
386 cpustop_restartfunc = cpu_reset_proxy;
387 printf("cpu_reset: Restarting BSP\n");
388 started_cpus = (1<<0); /* Restart CPU #0 */
389
390 cnt = 0;
391 while (cpu_reset_proxy_active == 0 && cnt < 10000000)
392 cnt++; /* Wait for BSP to announce restart */
393 if (cpu_reset_proxy_active == 0)
394 printf("cpu_reset: Failed to restart BSP\n");
395 __asm __volatile("cli" : : : "memory");
396 cpu_reset_proxy_active = 2;
397 cnt = 0;
398 while (cpu_reset_proxy_active == 2 && cnt < 10000000)
399 cnt++; /* Do nothing */
400 if (cpu_reset_proxy_active == 2) {
401 printf("cpu_reset: BSP did not grab mp lock\n");
402 cpu_reset_real(); /* XXX: Bogus ? */
403 }
404 cpu_reset_proxy_active = 4;
405 __asm __volatile("sti" : : : "memory");
406 while (1);
407 /* NOTREACHED */
408 }
409 }
410#else
411 cpu_reset_real();
412#endif
413}
414
415static void
416cpu_reset_real()
417{
418
419#ifdef PC98
420 /*
421 * Attempt to do a CPU reset via CPU reset port.
422 */
423 disable_intr();
424 if ((inb(0x35) & 0xa0) != 0xa0) {
425 outb(0x37, 0x0f); /* SHUT0 = 0. */
426 outb(0x37, 0x0b); /* SHUT1 = 0. */
427 }
428 outb(0xf0, 0x00); /* Reset. */
429#else
430 /*
431 * Attempt to do a CPU reset via the keyboard controller,
432 * do not turn of the GateA20, as any machine that fails
433 * to do the reset here would then end up in no man's land.
434 */
435
436#if !defined(BROKEN_KEYBOARD_RESET)
437 outb(IO_KBD + 4, 0xFE);
438 DELAY(500000); /* wait 0.5 sec to see if that did it */
439 printf("Keyboard reset did not work, attempting CPU shutdown\n");
440 DELAY(1000000); /* wait 1 sec for printf to complete */
441#endif
442#endif /* PC98 */
443 /* force a shutdown by unmapping entire address space ! */
444 bzero((caddr_t) PTD, PAGE_SIZE);
445
446 /* "good night, sweet prince .... <THUNK!>" */
447 invltlb();
448 /* NOTREACHED */
449 while(1);
450}
451
452int
453grow_stack(p, sp)
454 struct proc *p;
455 u_int sp;
456{
457 int rv;
458
459 rv = vm_map_growstack (p, sp);
460 if (rv != KERN_SUCCESS)
461 return (0);
462
463 return (1);
464}
465
466SYSCTL_DECL(_vm_stats_misc);
467
468static int cnt_prezero;
469
470SYSCTL_INT(_vm_stats_misc, OID_AUTO,
471 cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, "");
472
473/*
474 * Implement the pre-zeroed page mechanism.
475 * This routine is called from the idle loop.
476 */
477
478#define ZIDLE_LO(v) ((v) * 2 / 3)
479#define ZIDLE_HI(v) ((v) * 4 / 5)
480
481int
482vm_page_zero_idle()
483{
484 static int free_rover;
485 static int zero_state;
486 vm_page_t m;
487 int s;
488
489 /*
490 * Attempt to maintain approximately 1/2 of our free pages in a
491 * PG_ZERO'd state. Add some hysteresis to (attempt to) avoid
492 * generally zeroing a page when the system is near steady-state.
493 * Otherwise we might get 'flutter' during disk I/O / IPC or
494 * fast sleeps. We also do not want to be continuously zeroing
495 * pages because doing so may flush our L1 and L2 caches too much.
496 */
497
498 if (zero_state && vm_page_zero_count >= ZIDLE_LO(cnt.v_free_count))
499 return(0);
500 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count))
501 return(0);
502
503#ifdef SMP
504 if (try_mplock()) {
505#endif
506 s = splvm();
507 __asm __volatile("sti" : : : "memory");
508 zero_state = 0;
509 m = vm_page_list_find(PQ_FREE, free_rover, FALSE);
510 if (m != NULL && (m->flags & PG_ZERO) == 0) {
511 vm_page_queues[m->queue].lcnt--;
512 TAILQ_REMOVE(&vm_page_queues[m->queue].pl, m, pageq);
513 m->queue = PQ_NONE;
514 splx(s);
515 pmap_zero_page(VM_PAGE_TO_PHYS(m));
516 (void)splvm();
517 vm_page_flag_set(m, PG_ZERO);
518 m->queue = PQ_FREE + m->pc;
519 vm_page_queues[m->queue].lcnt++;
520 TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m,
521 pageq);
522 ++vm_page_zero_count;
523 ++cnt_prezero;
524 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count))
525 zero_state = 1;
526 }
527 free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK;
528 splx(s);
529 __asm __volatile("cli" : : : "memory");
530#ifdef SMP
531 rel_mplock();
532#endif
533 return (1);
534#ifdef SMP
535 }
536#endif
537 /*
538 * We have to enable interrupts for a moment if the try_mplock fails
539 * in order to potentially take an IPI. XXX this should be in
540 * swtch.s
541 */
542 __asm __volatile("sti; nop; cli" : : : "memory");
543 return (0);
544}
545
546/*
547 * Software interrupt handler for queued VM system processing.
548 */
549void
550swi_vm()
551{
552 if (busdma_swi_pending != 0)
553 busdma_swi();
554}
555
556/*
557 * Tell whether this address is in some physical memory region.
558 * Currently used by the kernel coredump code in order to avoid
559 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
560 * or other unpredictable behaviour.
561 */
562
563#include "isa.h"
564
565int
566is_physical_memory(addr)
567 vm_offset_t addr;
568{
569
570#if NISA > 0
571 /* The ISA ``memory hole''. */
572 if (addr >= 0xa0000 && addr < 0x100000)
573 return 0;
574#endif
575
576 /*
577 * stuff other tests for known memory-mapped devices (PCI?)
578 * here
579 */
580
581 return 1;
582}