proc->thread stage 6: kernel threads now create processless LWKT threads.
[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 $
0cfcada1 42 * $DragonFly: src/sys/i386/i386/Attic/vm_machdep.c,v 1.11 2003/06/27 01:53:24 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
0cfcada1
MD
247void
248cpu_set_thread_handler(thread_t td, void (*rfunc)(void), void *func, void *arg)
249{
250 td->td_pcb->pcb_esi = (int)func;
251 td->td_pcb->pcb_ebx = (int) arg;
252 td->td_switch = cpu_lwkt_switch;
253 td->td_sp -= sizeof(void *);
254 *(void **)td->td_sp = rfunc; /* exit function on return */
255 td->td_sp -= sizeof(void *);
256 *(void **)td->td_sp = cpu_kthread_restore;
257}
258
984263bc
MD
259void
260cpu_exit(p)
261 register struct proc *p;
262{
ae8050a4 263 struct pcb *pcb;
8ad65e08 264
984263bc
MD
265#if NNPX > 0
266 npxexit(p);
267#endif /* NNPX */
ae8050a4
MD
268
269 /*
270 * Cleanup the PCB
271 */
272 pcb = curthread->td_pcb;
984263bc
MD
273 if (pcb->pcb_ext != 0) {
274 /*
275 * XXX do we need to move the TSS off the allocated pages
276 * before freeing them? (not done here)
277 */
278 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
279 ctob(IOPAGES + 1));
280 pcb->pcb_ext = 0;
281 }
282#ifdef USER_LDT
283 user_ldt_free(pcb);
284#endif
285 if (pcb->pcb_flags & PCB_DBREGS) {
286 /*
287 * disable all hardware breakpoints
288 */
289 reset_dbregs();
290 pcb->pcb_flags &= ~PCB_DBREGS;
291 }
292 cnt.v_swtch++;
ae8050a4
MD
293
294 /*
295 * Set a special switch function which will release td_rwlock after
296 * the thread has been derferenced.
297 */
298 crit_enter();
299 KASSERT(curthread->td_switch == cpu_heavy_switch,
300 ("cpu_exit: unexpected switchout"));
301 curthread->td_switch = cpu_exit_switch;
302 lwkt_deschedule_self();
8ad65e08 303 lwkt_switch();
984263bc
MD
304 panic("cpu_exit");
305}
306
307void
308cpu_wait(p)
309 struct proc *p;
310{
7e1d4bf4
MD
311 struct thread *td;
312
984263bc 313 /* drop per-process resources */
7e1d4bf4
MD
314 td = pmap_dispose_proc(p);
315 if (td)
316 pmap_dispose_thread(td);
984263bc
MD
317}
318
319/*
320 * Dump the machine specific header information at the start of a core dump.
321 */
322int
dadab5e9 323cpu_coredump(struct thread *td, struct vnode *vp, struct ucred *cred)
984263bc 324{
dadab5e9 325 struct proc *p = td->td_proc;
984263bc
MD
326 int error;
327 caddr_t tempuser;
328
dadab5e9 329 KKASSERT(p);
984263bc
MD
330 tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK);
331 if (!tempuser)
332 return EINVAL;
333
334 bzero(tempuser, ctob(UPAGES));
335 bcopy(p->p_addr, tempuser, sizeof(struct user));
336 bcopy(p->p_md.md_regs,
337 tempuser + ((caddr_t) p->p_md.md_regs - (caddr_t) p->p_addr),
338 sizeof(struct trapframe));
b7c628e4 339 bcopy(p->p_thread->td_pcb, tempuser + ((char *)p->p_thread->td_pcb - (char *)p->p_addr), sizeof(struct pcb));
984263bc
MD
340
341 error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, ctob(UPAGES),
dadab5e9 342 (off_t)0, UIO_SYSSPACE, IO_UNIT, cred, (int *)NULL, td);
984263bc
MD
343
344 free(tempuser, M_TEMP);
345
346 return error;
347}
348
349#ifdef notyet
350static void
351setredzone(pte, vaddr)
352 u_short *pte;
353 caddr_t vaddr;
354{
355/* eventually do this by setting up an expand-down stack segment
356 for ss0: selector, allowing stack access down to top of u.
357 this means though that protection violations need to be handled
358 thru a double fault exception that must do an integral task
359 switch to a known good context, within which a dump can be
360 taken. a sensible scheme might be to save the initial context
361 used by sched (that has physical memory mapped 1:1 at bottom)
362 and take the dump while still in mapped mode */
363}
364#endif
365
366/*
367 * Convert kernel VA to physical address
368 */
369u_long
370kvtop(void *addr)
371{
372 vm_offset_t va;
373
374 va = pmap_kextract((vm_offset_t)addr);
375 if (va == 0)
376 panic("kvtop: zero page frame");
377 return((int)va);
378}
379
380/*
381 * Force reset the processor by invalidating the entire address space!
382 */
383
384#ifdef SMP
385static void
386cpu_reset_proxy()
387{
388 u_int saved_mp_lock;
389
390 cpu_reset_proxy_active = 1;
391 while (cpu_reset_proxy_active == 1)
392 ; /* Wait for other cpu to disable interupts */
393 saved_mp_lock = mp_lock;
394 mp_lock = 1;
395 printf("cpu_reset_proxy: Grabbed mp lock for BSP\n");
396 cpu_reset_proxy_active = 3;
397 while (cpu_reset_proxy_active == 3)
398 ; /* Wait for other cpu to enable interrupts */
399 stop_cpus((1<<cpu_reset_proxyid));
400 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
401 DELAY(1000000);
402 cpu_reset_real();
403}
404#endif
405
406void
407cpu_reset()
408{
409#ifdef SMP
410 if (smp_active == 0) {
411 cpu_reset_real();
412 /* NOTREACHED */
413 } else {
414
415 u_int map;
416 int cnt;
417 printf("cpu_reset called on cpu#%d\n",cpuid);
418
419 map = other_cpus & ~ stopped_cpus;
420
421 if (map != 0) {
422 printf("cpu_reset: Stopping other CPUs\n");
423 stop_cpus(map); /* Stop all other CPUs */
424 }
425
426 if (cpuid == 0) {
427 DELAY(1000000);
428 cpu_reset_real();
429 /* NOTREACHED */
430 } else {
431 /* We are not BSP (CPU #0) */
432
433 cpu_reset_proxyid = cpuid;
434 cpustop_restartfunc = cpu_reset_proxy;
435 printf("cpu_reset: Restarting BSP\n");
436 started_cpus = (1<<0); /* Restart CPU #0 */
437
438 cnt = 0;
439 while (cpu_reset_proxy_active == 0 && cnt < 10000000)
440 cnt++; /* Wait for BSP to announce restart */
441 if (cpu_reset_proxy_active == 0)
442 printf("cpu_reset: Failed to restart BSP\n");
443 __asm __volatile("cli" : : : "memory");
444 cpu_reset_proxy_active = 2;
445 cnt = 0;
446 while (cpu_reset_proxy_active == 2 && cnt < 10000000)
447 cnt++; /* Do nothing */
448 if (cpu_reset_proxy_active == 2) {
449 printf("cpu_reset: BSP did not grab mp lock\n");
450 cpu_reset_real(); /* XXX: Bogus ? */
451 }
452 cpu_reset_proxy_active = 4;
453 __asm __volatile("sti" : : : "memory");
454 while (1);
455 /* NOTREACHED */
456 }
457 }
458#else
459 cpu_reset_real();
460#endif
461}
462
463static void
464cpu_reset_real()
465{
466
467#ifdef PC98
468 /*
469 * Attempt to do a CPU reset via CPU reset port.
470 */
471 disable_intr();
472 if ((inb(0x35) & 0xa0) != 0xa0) {
473 outb(0x37, 0x0f); /* SHUT0 = 0. */
474 outb(0x37, 0x0b); /* SHUT1 = 0. */
475 }
476 outb(0xf0, 0x00); /* Reset. */
477#else
478 /*
479 * Attempt to do a CPU reset via the keyboard controller,
480 * do not turn of the GateA20, as any machine that fails
481 * to do the reset here would then end up in no man's land.
482 */
483
484#if !defined(BROKEN_KEYBOARD_RESET)
485 outb(IO_KBD + 4, 0xFE);
486 DELAY(500000); /* wait 0.5 sec to see if that did it */
487 printf("Keyboard reset did not work, attempting CPU shutdown\n");
488 DELAY(1000000); /* wait 1 sec for printf to complete */
489#endif
490#endif /* PC98 */
491 /* force a shutdown by unmapping entire address space ! */
492 bzero((caddr_t) PTD, PAGE_SIZE);
493
494 /* "good night, sweet prince .... <THUNK!>" */
495 invltlb();
496 /* NOTREACHED */
497 while(1);
498}
499
500int
501grow_stack(p, sp)
502 struct proc *p;
503 u_int sp;
504{
505 int rv;
506
507 rv = vm_map_growstack (p, sp);
508 if (rv != KERN_SUCCESS)
509 return (0);
510
511 return (1);
512}
513
514SYSCTL_DECL(_vm_stats_misc);
515
516static int cnt_prezero;
517
518SYSCTL_INT(_vm_stats_misc, OID_AUTO,
519 cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, "");
520
521/*
522 * Implement the pre-zeroed page mechanism.
523 * This routine is called from the idle loop.
524 */
525
526#define ZIDLE_LO(v) ((v) * 2 / 3)
527#define ZIDLE_HI(v) ((v) * 4 / 5)
528
529int
530vm_page_zero_idle()
531{
532 static int free_rover;
533 static int zero_state;
534 vm_page_t m;
535 int s;
536
537 /*
538 * Attempt to maintain approximately 1/2 of our free pages in a
539 * PG_ZERO'd state. Add some hysteresis to (attempt to) avoid
540 * generally zeroing a page when the system is near steady-state.
541 * Otherwise we might get 'flutter' during disk I/O / IPC or
542 * fast sleeps. We also do not want to be continuously zeroing
543 * pages because doing so may flush our L1 and L2 caches too much.
544 */
545
546 if (zero_state && vm_page_zero_count >= ZIDLE_LO(cnt.v_free_count))
547 return(0);
548 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count))
549 return(0);
550
551#ifdef SMP
552 if (try_mplock()) {
553#endif
554 s = splvm();
555 __asm __volatile("sti" : : : "memory");
556 zero_state = 0;
557 m = vm_page_list_find(PQ_FREE, free_rover, FALSE);
558 if (m != NULL && (m->flags & PG_ZERO) == 0) {
559 vm_page_queues[m->queue].lcnt--;
560 TAILQ_REMOVE(&vm_page_queues[m->queue].pl, m, pageq);
561 m->queue = PQ_NONE;
562 splx(s);
563 pmap_zero_page(VM_PAGE_TO_PHYS(m));
564 (void)splvm();
565 vm_page_flag_set(m, PG_ZERO);
566 m->queue = PQ_FREE + m->pc;
567 vm_page_queues[m->queue].lcnt++;
568 TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m,
569 pageq);
570 ++vm_page_zero_count;
571 ++cnt_prezero;
572 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count))
573 zero_state = 1;
574 }
575 free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK;
576 splx(s);
577 __asm __volatile("cli" : : : "memory");
578#ifdef SMP
579 rel_mplock();
580#endif
581 return (1);
582#ifdef SMP
583 }
584#endif
585 /*
586 * We have to enable interrupts for a moment if the try_mplock fails
587 * in order to potentially take an IPI. XXX this should be in
588 * swtch.s
589 */
590 __asm __volatile("sti; nop; cli" : : : "memory");
591 return (0);
592}
593
594/*
595 * Software interrupt handler for queued VM system processing.
596 */
597void
598swi_vm()
599{
600 if (busdma_swi_pending != 0)
601 busdma_swi();
602}
603
604/*
605 * Tell whether this address is in some physical memory region.
606 * Currently used by the kernel coredump code in order to avoid
607 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
608 * or other unpredictable behaviour.
609 */
610
611#include "isa.h"
612
613int
614is_physical_memory(addr)
615 vm_offset_t addr;
616{
617
618#if NISA > 0
619 /* The ISA ``memory hole''. */
620 if (addr >= 0xa0000 && addr < 0x100000)
621 return 0;
622#endif
623
624 /*
625 * stuff other tests for known memory-mapped devices (PCI?)
626 * here
627 */
628
629 return 1;
630}