thread stage 3: create independant thread structure, unembed from proc.
[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 $
263e4574 42 * $DragonFly: src/sys/i386/i386/Attic/vm_machdep.c,v 1.4 2003/06/18 16:30:09 dillon Exp $
984263bc
MD
43 */
44
45#include "npx.h"
46#include "opt_user_ldt.h"
47#ifdef PC98
48#include "opt_pc98.h"
49#endif
50#include "opt_reset.h"
51
52#include <sys/param.h>
53#include <sys/systm.h>
54#include <sys/malloc.h>
55#include <sys/proc.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
89static void cpu_reset_real __P((void));
90#ifdef SMP
91static void cpu_reset_proxy __P((void));
92static u_int cpu_reset_proxyid;
93static volatile u_int cpu_reset_proxy_active;
94#endif
95extern int _ucodesel, _udatasel;
96
97/*
98 * quick version of vm_fault
99 */
100int
101vm_fault_quick(v, prot)
102 caddr_t v;
103 int prot;
104{
105 int r;
106
107 if (prot & VM_PROT_WRITE)
108 r = subyte(v, fubyte(v));
109 else
110 r = fubyte(v);
111 return(r);
112}
113
114/*
115 * Finish a fork operation, with process p2 nearly set up.
116 * Copy and update the pcb, set up the stack so that the child
117 * ready to run and return to user mode.
118 */
119void
120cpu_fork(p1, p2, flags)
121 register struct proc *p1, *p2;
122 int flags;
123{
124 struct pcb *pcb2;
125
126 if ((flags & RFPROC) == 0) {
127#ifdef USER_LDT
128 if ((flags & RFMEM) == 0) {
129 /* unshare user LDT */
130 struct pcb *pcb1 = &p1->p_addr->u_pcb;
131 struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt;
132 if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) {
133 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len);
134 user_ldt_free(pcb1);
135 pcb1->pcb_ldt = pcb_ldt;
136 set_user_ldt(pcb1);
137 }
138 }
139#endif
140 return;
141 }
142
143#if NNPX > 0
144 /* Ensure that p1's pcb is up to date. */
263e4574 145 if (npxthread == p1->p_thread)
984263bc
MD
146 npxsave(&p1->p_addr->u_pcb.pcb_save);
147#endif
148
149 /* Copy p1's pcb. */
150 p2->p_addr->u_pcb = p1->p_addr->u_pcb;
151 pcb2 = &p2->p_addr->u_pcb;
152
153 /*
154 * Create a new fresh stack for the new process.
155 * Copy the trap frame for the return to user mode as if from a
156 * syscall. This copies the user mode register values.
157 */
158 p2->p_md.md_regs = (struct trapframe *)
159 ((int)p2->p_addr + UPAGES * PAGE_SIZE - 16) - 1;
160 bcopy(p1->p_md.md_regs, p2->p_md.md_regs, sizeof(*p2->p_md.md_regs));
161
162 /*
163 * Set registers for trampoline to user mode. Leave space for the
164 * return address on stack. These are the kernel mode register values.
165 */
166 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
167 pcb2->pcb_edi = 0;
168 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */
169 pcb2->pcb_ebp = 0;
170 pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *);
171 pcb2->pcb_ebx = (int)p2; /* fork_trampoline argument */
172 pcb2->pcb_eip = (int)fork_trampoline;
173 /*
174 * pcb2->pcb_ldt: duplicated below, if necessary.
175 * pcb2->pcb_savefpu: cloned above.
176 * pcb2->pcb_flags: cloned above (always 0 here?).
177 * pcb2->pcb_onfault: cloned above (always NULL here?).
178 */
179
180#ifdef SMP
181 pcb2->pcb_mpnest = 1;
182#endif
183 /*
184 * XXX don't copy the i/o pages. this should probably be fixed.
185 */
186 pcb2->pcb_ext = 0;
187
188#ifdef USER_LDT
189 /* Copy the LDT, if necessary. */
190 if (pcb2->pcb_ldt != 0) {
191 if (flags & RFMEM) {
192 pcb2->pcb_ldt->ldt_refcnt++;
193 } else {
194 pcb2->pcb_ldt = user_ldt_alloc(pcb2,
195 pcb2->pcb_ldt->ldt_len);
196 }
197 }
198#endif
199
200 /*
201 * Now, cpu_switch() can schedule the new process.
202 * pcb_esp is loaded pointing to the cpu_switch() stack frame
203 * containing the return address when exiting cpu_switch.
204 * This will normally be to fork_trampoline(), which will have
205 * %ebx loaded with the new proc's pointer. fork_trampoline()
206 * will set up a stack to call fork_return(p, frame); to complete
207 * the return to user-mode.
208 */
209}
210
211/*
212 * Intercept the return address from a freshly forked process that has NOT
213 * been scheduled yet.
214 *
215 * This is needed to make kernel threads stay in kernel mode.
216 */
217void
218cpu_set_fork_handler(p, func, arg)
219 struct proc *p;
220 void (*func) __P((void *));
221 void *arg;
222{
223 /*
224 * Note that the trap frame follows the args, so the function
225 * is really called like this: func(arg, frame);
226 */
227 p->p_addr->u_pcb.pcb_esi = (int) func; /* function */
228 p->p_addr->u_pcb.pcb_ebx = (int) arg; /* first arg */
229}
230
231void
232cpu_exit(p)
233 register struct proc *p;
234{
235 struct pcb *pcb = &p->p_addr->u_pcb;
236
237#if NNPX > 0
238 npxexit(p);
239#endif /* NNPX */
240 if (pcb->pcb_ext != 0) {
241 /*
242 * XXX do we need to move the TSS off the allocated pages
243 * before freeing them? (not done here)
244 */
245 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
246 ctob(IOPAGES + 1));
247 pcb->pcb_ext = 0;
248 }
249#ifdef USER_LDT
250 user_ldt_free(pcb);
251#endif
252 if (pcb->pcb_flags & PCB_DBREGS) {
253 /*
254 * disable all hardware breakpoints
255 */
256 reset_dbregs();
257 pcb->pcb_flags &= ~PCB_DBREGS;
258 }
259 cnt.v_swtch++;
260 cpu_switch(p);
261 panic("cpu_exit");
262}
263
264void
265cpu_wait(p)
266 struct proc *p;
267{
268 /* drop per-process resources */
269 pmap_dispose_proc(p);
263e4574 270 pmap_dispose_thread(p->p_thread);
984263bc
MD
271}
272
273/*
274 * Dump the machine specific header information at the start of a core dump.
275 */
276int
277cpu_coredump(p, vp, cred)
278 struct proc *p;
279 struct vnode *vp;
280 struct ucred *cred;
281{
282 int error;
283 caddr_t tempuser;
284
285 tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK);
286 if (!tempuser)
287 return EINVAL;
288
289 bzero(tempuser, ctob(UPAGES));
290 bcopy(p->p_addr, tempuser, sizeof(struct user));
291 bcopy(p->p_md.md_regs,
292 tempuser + ((caddr_t) p->p_md.md_regs - (caddr_t) p->p_addr),
293 sizeof(struct trapframe));
294
295 error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, ctob(UPAGES),
296 (off_t)0, UIO_SYSSPACE, IO_UNIT, cred, (int *)NULL, p);
297
298 free(tempuser, M_TEMP);
299
300 return error;
301}
302
303#ifdef notyet
304static void
305setredzone(pte, vaddr)
306 u_short *pte;
307 caddr_t vaddr;
308{
309/* eventually do this by setting up an expand-down stack segment
310 for ss0: selector, allowing stack access down to top of u.
311 this means though that protection violations need to be handled
312 thru a double fault exception that must do an integral task
313 switch to a known good context, within which a dump can be
314 taken. a sensible scheme might be to save the initial context
315 used by sched (that has physical memory mapped 1:1 at bottom)
316 and take the dump while still in mapped mode */
317}
318#endif
319
320/*
321 * Convert kernel VA to physical address
322 */
323u_long
324kvtop(void *addr)
325{
326 vm_offset_t va;
327
328 va = pmap_kextract((vm_offset_t)addr);
329 if (va == 0)
330 panic("kvtop: zero page frame");
331 return((int)va);
332}
333
334/*
335 * Force reset the processor by invalidating the entire address space!
336 */
337
338#ifdef SMP
339static void
340cpu_reset_proxy()
341{
342 u_int saved_mp_lock;
343
344 cpu_reset_proxy_active = 1;
345 while (cpu_reset_proxy_active == 1)
346 ; /* Wait for other cpu to disable interupts */
347 saved_mp_lock = mp_lock;
348 mp_lock = 1;
349 printf("cpu_reset_proxy: Grabbed mp lock for BSP\n");
350 cpu_reset_proxy_active = 3;
351 while (cpu_reset_proxy_active == 3)
352 ; /* Wait for other cpu to enable interrupts */
353 stop_cpus((1<<cpu_reset_proxyid));
354 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
355 DELAY(1000000);
356 cpu_reset_real();
357}
358#endif
359
360void
361cpu_reset()
362{
363#ifdef SMP
364 if (smp_active == 0) {
365 cpu_reset_real();
366 /* NOTREACHED */
367 } else {
368
369 u_int map;
370 int cnt;
371 printf("cpu_reset called on cpu#%d\n",cpuid);
372
373 map = other_cpus & ~ stopped_cpus;
374
375 if (map != 0) {
376 printf("cpu_reset: Stopping other CPUs\n");
377 stop_cpus(map); /* Stop all other CPUs */
378 }
379
380 if (cpuid == 0) {
381 DELAY(1000000);
382 cpu_reset_real();
383 /* NOTREACHED */
384 } else {
385 /* We are not BSP (CPU #0) */
386
387 cpu_reset_proxyid = cpuid;
388 cpustop_restartfunc = cpu_reset_proxy;
389 printf("cpu_reset: Restarting BSP\n");
390 started_cpus = (1<<0); /* Restart CPU #0 */
391
392 cnt = 0;
393 while (cpu_reset_proxy_active == 0 && cnt < 10000000)
394 cnt++; /* Wait for BSP to announce restart */
395 if (cpu_reset_proxy_active == 0)
396 printf("cpu_reset: Failed to restart BSP\n");
397 __asm __volatile("cli" : : : "memory");
398 cpu_reset_proxy_active = 2;
399 cnt = 0;
400 while (cpu_reset_proxy_active == 2 && cnt < 10000000)
401 cnt++; /* Do nothing */
402 if (cpu_reset_proxy_active == 2) {
403 printf("cpu_reset: BSP did not grab mp lock\n");
404 cpu_reset_real(); /* XXX: Bogus ? */
405 }
406 cpu_reset_proxy_active = 4;
407 __asm __volatile("sti" : : : "memory");
408 while (1);
409 /* NOTREACHED */
410 }
411 }
412#else
413 cpu_reset_real();
414#endif
415}
416
417static void
418cpu_reset_real()
419{
420
421#ifdef PC98
422 /*
423 * Attempt to do a CPU reset via CPU reset port.
424 */
425 disable_intr();
426 if ((inb(0x35) & 0xa0) != 0xa0) {
427 outb(0x37, 0x0f); /* SHUT0 = 0. */
428 outb(0x37, 0x0b); /* SHUT1 = 0. */
429 }
430 outb(0xf0, 0x00); /* Reset. */
431#else
432 /*
433 * Attempt to do a CPU reset via the keyboard controller,
434 * do not turn of the GateA20, as any machine that fails
435 * to do the reset here would then end up in no man's land.
436 */
437
438#if !defined(BROKEN_KEYBOARD_RESET)
439 outb(IO_KBD + 4, 0xFE);
440 DELAY(500000); /* wait 0.5 sec to see if that did it */
441 printf("Keyboard reset did not work, attempting CPU shutdown\n");
442 DELAY(1000000); /* wait 1 sec for printf to complete */
443#endif
444#endif /* PC98 */
445 /* force a shutdown by unmapping entire address space ! */
446 bzero((caddr_t) PTD, PAGE_SIZE);
447
448 /* "good night, sweet prince .... <THUNK!>" */
449 invltlb();
450 /* NOTREACHED */
451 while(1);
452}
453
454int
455grow_stack(p, sp)
456 struct proc *p;
457 u_int sp;
458{
459 int rv;
460
461 rv = vm_map_growstack (p, sp);
462 if (rv != KERN_SUCCESS)
463 return (0);
464
465 return (1);
466}
467
468SYSCTL_DECL(_vm_stats_misc);
469
470static int cnt_prezero;
471
472SYSCTL_INT(_vm_stats_misc, OID_AUTO,
473 cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, "");
474
475/*
476 * Implement the pre-zeroed page mechanism.
477 * This routine is called from the idle loop.
478 */
479
480#define ZIDLE_LO(v) ((v) * 2 / 3)
481#define ZIDLE_HI(v) ((v) * 4 / 5)
482
483int
484vm_page_zero_idle()
485{
486 static int free_rover;
487 static int zero_state;
488 vm_page_t m;
489 int s;
490
491 /*
492 * Attempt to maintain approximately 1/2 of our free pages in a
493 * PG_ZERO'd state. Add some hysteresis to (attempt to) avoid
494 * generally zeroing a page when the system is near steady-state.
495 * Otherwise we might get 'flutter' during disk I/O / IPC or
496 * fast sleeps. We also do not want to be continuously zeroing
497 * pages because doing so may flush our L1 and L2 caches too much.
498 */
499
500 if (zero_state && vm_page_zero_count >= ZIDLE_LO(cnt.v_free_count))
501 return(0);
502 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count))
503 return(0);
504
505#ifdef SMP
506 if (try_mplock()) {
507#endif
508 s = splvm();
509 __asm __volatile("sti" : : : "memory");
510 zero_state = 0;
511 m = vm_page_list_find(PQ_FREE, free_rover, FALSE);
512 if (m != NULL && (m->flags & PG_ZERO) == 0) {
513 vm_page_queues[m->queue].lcnt--;
514 TAILQ_REMOVE(&vm_page_queues[m->queue].pl, m, pageq);
515 m->queue = PQ_NONE;
516 splx(s);
517 pmap_zero_page(VM_PAGE_TO_PHYS(m));
518 (void)splvm();
519 vm_page_flag_set(m, PG_ZERO);
520 m->queue = PQ_FREE + m->pc;
521 vm_page_queues[m->queue].lcnt++;
522 TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m,
523 pageq);
524 ++vm_page_zero_count;
525 ++cnt_prezero;
526 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count))
527 zero_state = 1;
528 }
529 free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK;
530 splx(s);
531 __asm __volatile("cli" : : : "memory");
532#ifdef SMP
533 rel_mplock();
534#endif
535 return (1);
536#ifdef SMP
537 }
538#endif
539 /*
540 * We have to enable interrupts for a moment if the try_mplock fails
541 * in order to potentially take an IPI. XXX this should be in
542 * swtch.s
543 */
544 __asm __volatile("sti; nop; cli" : : : "memory");
545 return (0);
546}
547
548/*
549 * Software interrupt handler for queued VM system processing.
550 */
551void
552swi_vm()
553{
554 if (busdma_swi_pending != 0)
555 busdma_swi();
556}
557
558/*
559 * Tell whether this address is in some physical memory region.
560 * Currently used by the kernel coredump code in order to avoid
561 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
562 * or other unpredictable behaviour.
563 */
564
565#include "isa.h"
566
567int
568is_physical_memory(addr)
569 vm_offset_t addr;
570{
571
572#if NISA > 0
573 /* The ISA ``memory hole''. */
574 if (addr >= 0xa0000 && addr < 0x100000)
575 return 0;
576#endif
577
578 /*
579 * stuff other tests for known memory-mapped devices (PCI?)
580 * here
581 */
582
583 return 1;
584}