proc->thread stage 3.5: Add an IO_CORE flag so coda doesn't have to dig
[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 $
7d0bac62 42 * $DragonFly: src/sys/i386/i386/Attic/vm_machdep.c,v 1.9 2003/06/22 04:30:39 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>
ae8050a4 82#include <sys/thread2.h>
984263bc
MD
83
84#ifdef PC98
85#include <pc98/pc98/pc98.h>
86#else
87#include <i386/isa/isa.h>
88#endif
89
90static void cpu_reset_real __P((void));
91#ifdef SMP
92static void cpu_reset_proxy __P((void));
93static u_int cpu_reset_proxyid;
94static volatile u_int cpu_reset_proxy_active;
95#endif
96extern int _ucodesel, _udatasel;
97
98/*
99 * quick version of vm_fault
100 */
101int
102vm_fault_quick(v, prot)
103 caddr_t v;
104 int prot;
105{
106 int r;
107
108 if (prot & VM_PROT_WRITE)
109 r = subyte(v, fubyte(v));
110 else
111 r = fubyte(v);
112 return(r);
113}
114
115/*
116 * Finish a fork operation, with process p2 nearly set up.
117 * Copy and update the pcb, set up the stack so that the child
118 * ready to run and return to user mode.
119 */
120void
121cpu_fork(p1, p2, flags)
122 register struct proc *p1, *p2;
123 int flags;
124{
125 struct pcb *pcb2;
126
127 if ((flags & RFPROC) == 0) {
128#ifdef USER_LDT
129 if ((flags & RFMEM) == 0) {
130 /* unshare user LDT */
b7c628e4 131 struct pcb *pcb1 = p1->p_thread->td_pcb;
984263bc
MD
132 struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt;
133 if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) {
134 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len);
135 user_ldt_free(pcb1);
136 pcb1->pcb_ldt = pcb_ldt;
137 set_user_ldt(pcb1);
138 }
139 }
140#endif
141 return;
142 }
143
144#if NNPX > 0
145 /* Ensure that p1's pcb is up to date. */
263e4574 146 if (npxthread == p1->p_thread)
b7c628e4 147 npxsave(&p1->p_thread->td_pcb->pcb_save);
984263bc
MD
148#endif
149
150 /* Copy p1's pcb. */
b7c628e4 151 pcb2 = p2->p_thread->td_pcb;
7d0bac62 152 *pcb2 = *p1->p_thread->td_pcb;
984263bc
MD
153
154 /*
155 * Create a new fresh stack for the new process.
156 * Copy the trap frame for the return to user mode as if from a
b7c628e4
MD
157 * syscall. This copies the user mode register values. The
158 * 16 byte offset saves space for vm86, and must match
159 * common_tss.esp0 (kernel stack pointer on entry from user mode)
8ad65e08
MD
160 *
161 * pcb_esp must allocate an additional call-return pointer below
162 * the trap frame which will be restored by cpu_restore, and the
163 * thread's td_sp pointer must allocate an additonal call-return
164 * pointer below the pcb_esp call-return pointer to hold the LWKT
165 * restore function pointer.
166 *
167 * The LWKT restore function pointer must be set to cpu_restore,
168 * which is our standard heavy weight process switch-in function.
169 * YYY eventually we should shortcut fork_return and fork_trampoline
170 * to use the LWKT restore function directly so we can get rid of
171 * all the extra crap we are setting up.
984263bc 172 */
7d0bac62 173 p2->p_md.md_regs = (struct trapframe *)((char *)pcb2 - 16) - 1;
984263bc
MD
174 bcopy(p1->p_md.md_regs, p2->p_md.md_regs, sizeof(*p2->p_md.md_regs));
175
176 /*
177 * Set registers for trampoline to user mode. Leave space for the
178 * return address on stack. These are the kernel mode register values.
179 */
180 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
181 pcb2->pcb_edi = 0;
182 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */
183 pcb2->pcb_ebp = 0;
184 pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *);
185 pcb2->pcb_ebx = (int)p2; /* fork_trampoline argument */
186 pcb2->pcb_eip = (int)fork_trampoline;
8ad65e08
MD
187 p2->p_thread->td_sp = (char *)(pcb2->pcb_esp - sizeof(void *));
188 *(void **)p2->p_thread->td_sp = (void *)cpu_heavy_restore;
984263bc
MD
189 /*
190 * pcb2->pcb_ldt: duplicated below, if necessary.
191 * pcb2->pcb_savefpu: cloned above.
192 * pcb2->pcb_flags: cloned above (always 0 here?).
193 * pcb2->pcb_onfault: cloned above (always NULL here?).
194 */
195
196#ifdef SMP
197 pcb2->pcb_mpnest = 1;
198#endif
199 /*
200 * XXX don't copy the i/o pages. this should probably be fixed.
201 */
202 pcb2->pcb_ext = 0;
203
204#ifdef USER_LDT
205 /* Copy the LDT, if necessary. */
206 if (pcb2->pcb_ldt != 0) {
207 if (flags & RFMEM) {
208 pcb2->pcb_ldt->ldt_refcnt++;
209 } else {
210 pcb2->pcb_ldt = user_ldt_alloc(pcb2,
211 pcb2->pcb_ldt->ldt_len);
212 }
213 }
214#endif
215
216 /*
217 * Now, cpu_switch() can schedule the new process.
218 * pcb_esp is loaded pointing to the cpu_switch() stack frame
219 * containing the return address when exiting cpu_switch.
220 * This will normally be to fork_trampoline(), which will have
221 * %ebx loaded with the new proc's pointer. fork_trampoline()
222 * will set up a stack to call fork_return(p, frame); to complete
223 * the return to user-mode.
224 */
225}
226
227/*
228 * Intercept the return address from a freshly forked process that has NOT
229 * been scheduled yet.
230 *
231 * This is needed to make kernel threads stay in kernel mode.
232 */
233void
234cpu_set_fork_handler(p, func, arg)
235 struct proc *p;
236 void (*func) __P((void *));
237 void *arg;
238{
239 /*
240 * Note that the trap frame follows the args, so the function
241 * is really called like this: func(arg, frame);
242 */
b7c628e4
MD
243 p->p_thread->td_pcb->pcb_esi = (int) func; /* function */
244 p->p_thread->td_pcb->pcb_ebx = (int) arg; /* first arg */
984263bc
MD
245}
246
247void
248cpu_exit(p)
249 register struct proc *p;
250{
ae8050a4 251 struct pcb *pcb;
8ad65e08 252
984263bc
MD
253#if NNPX > 0
254 npxexit(p);
255#endif /* NNPX */
ae8050a4
MD
256
257 /*
258 * Cleanup the PCB
259 */
260 pcb = curthread->td_pcb;
984263bc
MD
261 if (pcb->pcb_ext != 0) {
262 /*
263 * XXX do we need to move the TSS off the allocated pages
264 * before freeing them? (not done here)
265 */
266 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
267 ctob(IOPAGES + 1));
268 pcb->pcb_ext = 0;
269 }
270#ifdef USER_LDT
271 user_ldt_free(pcb);
272#endif
273 if (pcb->pcb_flags & PCB_DBREGS) {
274 /*
275 * disable all hardware breakpoints
276 */
277 reset_dbregs();
278 pcb->pcb_flags &= ~PCB_DBREGS;
279 }
280 cnt.v_swtch++;
ae8050a4
MD
281
282 /*
283 * Set a special switch function which will release td_rwlock after
284 * the thread has been derferenced.
285 */
286 crit_enter();
287 KASSERT(curthread->td_switch == cpu_heavy_switch,
288 ("cpu_exit: unexpected switchout"));
289 curthread->td_switch = cpu_exit_switch;
290 lwkt_deschedule_self();
8ad65e08 291 lwkt_switch();
984263bc
MD
292 panic("cpu_exit");
293}
294
295void
296cpu_wait(p)
297 struct proc *p;
298{
7e1d4bf4
MD
299 struct thread *td;
300
984263bc 301 /* drop per-process resources */
7e1d4bf4
MD
302 td = pmap_dispose_proc(p);
303 if (td)
304 pmap_dispose_thread(td);
984263bc
MD
305}
306
307/*
308 * Dump the machine specific header information at the start of a core dump.
309 */
310int
311cpu_coredump(p, vp, cred)
312 struct proc *p;
313 struct vnode *vp;
314 struct ucred *cred;
315{
316 int error;
317 caddr_t tempuser;
318
319 tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK);
320 if (!tempuser)
321 return EINVAL;
322
323 bzero(tempuser, ctob(UPAGES));
324 bcopy(p->p_addr, tempuser, sizeof(struct user));
325 bcopy(p->p_md.md_regs,
326 tempuser + ((caddr_t) p->p_md.md_regs - (caddr_t) p->p_addr),
327 sizeof(struct trapframe));
b7c628e4 328 bcopy(p->p_thread->td_pcb, tempuser + ((char *)p->p_thread->td_pcb - (char *)p->p_addr), sizeof(struct pcb));
984263bc
MD
329
330 error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, ctob(UPAGES),
331 (off_t)0, UIO_SYSSPACE, IO_UNIT, cred, (int *)NULL, p);
332
333 free(tempuser, M_TEMP);
334
335 return error;
336}
337
338#ifdef notyet
339static void
340setredzone(pte, vaddr)
341 u_short *pte;
342 caddr_t vaddr;
343{
344/* eventually do this by setting up an expand-down stack segment
345 for ss0: selector, allowing stack access down to top of u.
346 this means though that protection violations need to be handled
347 thru a double fault exception that must do an integral task
348 switch to a known good context, within which a dump can be
349 taken. a sensible scheme might be to save the initial context
350 used by sched (that has physical memory mapped 1:1 at bottom)
351 and take the dump while still in mapped mode */
352}
353#endif
354
355/*
356 * Convert kernel VA to physical address
357 */
358u_long
359kvtop(void *addr)
360{
361 vm_offset_t va;
362
363 va = pmap_kextract((vm_offset_t)addr);
364 if (va == 0)
365 panic("kvtop: zero page frame");
366 return((int)va);
367}
368
369/*
370 * Force reset the processor by invalidating the entire address space!
371 */
372
373#ifdef SMP
374static void
375cpu_reset_proxy()
376{
377 u_int saved_mp_lock;
378
379 cpu_reset_proxy_active = 1;
380 while (cpu_reset_proxy_active == 1)
381 ; /* Wait for other cpu to disable interupts */
382 saved_mp_lock = mp_lock;
383 mp_lock = 1;
384 printf("cpu_reset_proxy: Grabbed mp lock for BSP\n");
385 cpu_reset_proxy_active = 3;
386 while (cpu_reset_proxy_active == 3)
387 ; /* Wait for other cpu to enable interrupts */
388 stop_cpus((1<<cpu_reset_proxyid));
389 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
390 DELAY(1000000);
391 cpu_reset_real();
392}
393#endif
394
395void
396cpu_reset()
397{
398#ifdef SMP
399 if (smp_active == 0) {
400 cpu_reset_real();
401 /* NOTREACHED */
402 } else {
403
404 u_int map;
405 int cnt;
406 printf("cpu_reset called on cpu#%d\n",cpuid);
407
408 map = other_cpus & ~ stopped_cpus;
409
410 if (map != 0) {
411 printf("cpu_reset: Stopping other CPUs\n");
412 stop_cpus(map); /* Stop all other CPUs */
413 }
414
415 if (cpuid == 0) {
416 DELAY(1000000);
417 cpu_reset_real();
418 /* NOTREACHED */
419 } else {
420 /* We are not BSP (CPU #0) */
421
422 cpu_reset_proxyid = cpuid;
423 cpustop_restartfunc = cpu_reset_proxy;
424 printf("cpu_reset: Restarting BSP\n");
425 started_cpus = (1<<0); /* Restart CPU #0 */
426
427 cnt = 0;
428 while (cpu_reset_proxy_active == 0 && cnt < 10000000)
429 cnt++; /* Wait for BSP to announce restart */
430 if (cpu_reset_proxy_active == 0)
431 printf("cpu_reset: Failed to restart BSP\n");
432 __asm __volatile("cli" : : : "memory");
433 cpu_reset_proxy_active = 2;
434 cnt = 0;
435 while (cpu_reset_proxy_active == 2 && cnt < 10000000)
436 cnt++; /* Do nothing */
437 if (cpu_reset_proxy_active == 2) {
438 printf("cpu_reset: BSP did not grab mp lock\n");
439 cpu_reset_real(); /* XXX: Bogus ? */
440 }
441 cpu_reset_proxy_active = 4;
442 __asm __volatile("sti" : : : "memory");
443 while (1);
444 /* NOTREACHED */
445 }
446 }
447#else
448 cpu_reset_real();
449#endif
450}
451
452static void
453cpu_reset_real()
454{
455
456#ifdef PC98
457 /*
458 * Attempt to do a CPU reset via CPU reset port.
459 */
460 disable_intr();
461 if ((inb(0x35) & 0xa0) != 0xa0) {
462 outb(0x37, 0x0f); /* SHUT0 = 0. */
463 outb(0x37, 0x0b); /* SHUT1 = 0. */
464 }
465 outb(0xf0, 0x00); /* Reset. */
466#else
467 /*
468 * Attempt to do a CPU reset via the keyboard controller,
469 * do not turn of the GateA20, as any machine that fails
470 * to do the reset here would then end up in no man's land.
471 */
472
473#if !defined(BROKEN_KEYBOARD_RESET)
474 outb(IO_KBD + 4, 0xFE);
475 DELAY(500000); /* wait 0.5 sec to see if that did it */
476 printf("Keyboard reset did not work, attempting CPU shutdown\n");
477 DELAY(1000000); /* wait 1 sec for printf to complete */
478#endif
479#endif /* PC98 */
480 /* force a shutdown by unmapping entire address space ! */
481 bzero((caddr_t) PTD, PAGE_SIZE);
482
483 /* "good night, sweet prince .... <THUNK!>" */
484 invltlb();
485 /* NOTREACHED */
486 while(1);
487}
488
489int
490grow_stack(p, sp)
491 struct proc *p;
492 u_int sp;
493{
494 int rv;
495
496 rv = vm_map_growstack (p, sp);
497 if (rv != KERN_SUCCESS)
498 return (0);
499
500 return (1);
501}
502
503SYSCTL_DECL(_vm_stats_misc);
504
505static int cnt_prezero;
506
507SYSCTL_INT(_vm_stats_misc, OID_AUTO,
508 cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, "");
509
510/*
511 * Implement the pre-zeroed page mechanism.
512 * This routine is called from the idle loop.
513 */
514
515#define ZIDLE_LO(v) ((v) * 2 / 3)
516#define ZIDLE_HI(v) ((v) * 4 / 5)
517
518int
519vm_page_zero_idle()
520{
521 static int free_rover;
522 static int zero_state;
523 vm_page_t m;
524 int s;
525
526 /*
527 * Attempt to maintain approximately 1/2 of our free pages in a
528 * PG_ZERO'd state. Add some hysteresis to (attempt to) avoid
529 * generally zeroing a page when the system is near steady-state.
530 * Otherwise we might get 'flutter' during disk I/O / IPC or
531 * fast sleeps. We also do not want to be continuously zeroing
532 * pages because doing so may flush our L1 and L2 caches too much.
533 */
534
535 if (zero_state && vm_page_zero_count >= ZIDLE_LO(cnt.v_free_count))
536 return(0);
537 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count))
538 return(0);
539
540#ifdef SMP
541 if (try_mplock()) {
542#endif
543 s = splvm();
544 __asm __volatile("sti" : : : "memory");
545 zero_state = 0;
546 m = vm_page_list_find(PQ_FREE, free_rover, FALSE);
547 if (m != NULL && (m->flags & PG_ZERO) == 0) {
548 vm_page_queues[m->queue].lcnt--;
549 TAILQ_REMOVE(&vm_page_queues[m->queue].pl, m, pageq);
550 m->queue = PQ_NONE;
551 splx(s);
552 pmap_zero_page(VM_PAGE_TO_PHYS(m));
553 (void)splvm();
554 vm_page_flag_set(m, PG_ZERO);
555 m->queue = PQ_FREE + m->pc;
556 vm_page_queues[m->queue].lcnt++;
557 TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m,
558 pageq);
559 ++vm_page_zero_count;
560 ++cnt_prezero;
561 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count))
562 zero_state = 1;
563 }
564 free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK;
565 splx(s);
566 __asm __volatile("cli" : : : "memory");
567#ifdef SMP
568 rel_mplock();
569#endif
570 return (1);
571#ifdef SMP
572 }
573#endif
574 /*
575 * We have to enable interrupts for a moment if the try_mplock fails
576 * in order to potentially take an IPI. XXX this should be in
577 * swtch.s
578 */
579 __asm __volatile("sti; nop; cli" : : : "memory");
580 return (0);
581}
582
583/*
584 * Software interrupt handler for queued VM system processing.
585 */
586void
587swi_vm()
588{
589 if (busdma_swi_pending != 0)
590 busdma_swi();
591}
592
593/*
594 * Tell whether this address is in some physical memory region.
595 * Currently used by the kernel coredump code in order to avoid
596 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
597 * or other unpredictable behaviour.
598 */
599
600#include "isa.h"
601
602int
603is_physical_memory(addr)
604 vm_offset_t addr;
605{
606
607#if NISA > 0
608 /* The ISA ``memory hole''. */
609 if (addr >= 0xa0000 && addr < 0x100000)
610 return 0;
611#endif
612
613 /*
614 * stuff other tests for known memory-mapped devices (PCI?)
615 * here
616 */
617
618 return 1;
619}