add gd_other_cpus
[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 $
12e4aaff 42 * $DragonFly: src/sys/i386/i386/Attic/vm_machdep.c,v 1.16 2003/07/03 17:24:01 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>
ef0fdad1 57#include <sys/interrupt.h>
984263bc
MD
58#include <sys/vnode.h>
59#include <sys/vmmeter.h>
60#include <sys/kernel.h>
61#include <sys/sysctl.h>
62#include <sys/unistd.h>
63
64#include <machine/clock.h>
65#include <machine/cpu.h>
66#include <machine/md_var.h>
67#ifdef SMP
68#include <machine/smp.h>
69#endif
70#include <machine/pcb.h>
71#include <machine/pcb_ext.h>
72#include <machine/vm86.h>
85100692 73#include <machine/globaldata.h> /* npxthread */
ef0fdad1 74#include <machine/ipl.h> /* SWI_ */
984263bc
MD
75
76#include <vm/vm.h>
77#include <vm/vm_param.h>
78#include <sys/lock.h>
79#include <vm/vm_kern.h>
80#include <vm/vm_page.h>
81#include <vm/vm_map.h>
82#include <vm/vm_extern.h>
83
84#include <sys/user.h>
ae8050a4 85#include <sys/thread2.h>
984263bc
MD
86
87#ifdef PC98
88#include <pc98/pc98/pc98.h>
89#else
90#include <i386/isa/isa.h>
91#endif
92
93static void cpu_reset_real __P((void));
94#ifdef SMP
95static void cpu_reset_proxy __P((void));
96static u_int cpu_reset_proxyid;
97static volatile u_int cpu_reset_proxy_active;
98#endif
99extern int _ucodesel, _udatasel;
100
101/*
102 * quick version of vm_fault
103 */
104int
105vm_fault_quick(v, prot)
106 caddr_t v;
107 int prot;
108{
109 int r;
110
111 if (prot & VM_PROT_WRITE)
112 r = subyte(v, fubyte(v));
113 else
114 r = fubyte(v);
115 return(r);
116}
117
118/*
119 * Finish a fork operation, with process p2 nearly set up.
120 * Copy and update the pcb, set up the stack so that the child
121 * ready to run and return to user mode.
122 */
123void
124cpu_fork(p1, p2, flags)
125 register struct proc *p1, *p2;
126 int flags;
127{
128 struct pcb *pcb2;
129
130 if ((flags & RFPROC) == 0) {
131#ifdef USER_LDT
132 if ((flags & RFMEM) == 0) {
133 /* unshare user LDT */
b7c628e4 134 struct pcb *pcb1 = p1->p_thread->td_pcb;
984263bc
MD
135 struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt;
136 if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) {
137 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len);
138 user_ldt_free(pcb1);
139 pcb1->pcb_ldt = pcb_ldt;
140 set_user_ldt(pcb1);
141 }
142 }
143#endif
144 return;
145 }
146
147#if NNPX > 0
148 /* Ensure that p1's pcb is up to date. */
263e4574 149 if (npxthread == p1->p_thread)
b7c628e4 150 npxsave(&p1->p_thread->td_pcb->pcb_save);
984263bc
MD
151#endif
152
153 /* Copy p1's pcb. */
b7c628e4 154 pcb2 = p2->p_thread->td_pcb;
7d0bac62 155 *pcb2 = *p1->p_thread->td_pcb;
984263bc
MD
156
157 /*
158 * Create a new fresh stack for the new process.
159 * Copy the trap frame for the return to user mode as if from a
b7c628e4
MD
160 * syscall. This copies the user mode register values. The
161 * 16 byte offset saves space for vm86, and must match
162 * common_tss.esp0 (kernel stack pointer on entry from user mode)
8ad65e08
MD
163 *
164 * pcb_esp must allocate an additional call-return pointer below
165 * the trap frame which will be restored by cpu_restore, and the
166 * thread's td_sp pointer must allocate an additonal call-return
167 * pointer below the pcb_esp call-return pointer to hold the LWKT
168 * restore function pointer.
169 *
170 * The LWKT restore function pointer must be set to cpu_restore,
171 * which is our standard heavy weight process switch-in function.
172 * YYY eventually we should shortcut fork_return and fork_trampoline
173 * to use the LWKT restore function directly so we can get rid of
174 * all the extra crap we are setting up.
984263bc 175 */
7d0bac62 176 p2->p_md.md_regs = (struct trapframe *)((char *)pcb2 - 16) - 1;
984263bc
MD
177 bcopy(p1->p_md.md_regs, p2->p_md.md_regs, sizeof(*p2->p_md.md_regs));
178
179 /*
180 * Set registers for trampoline to user mode. Leave space for the
181 * return address on stack. These are the kernel mode register values.
182 */
183 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
184 pcb2->pcb_edi = 0;
185 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */
186 pcb2->pcb_ebp = 0;
187 pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *);
188 pcb2->pcb_ebx = (int)p2; /* fork_trampoline argument */
189 pcb2->pcb_eip = (int)fork_trampoline;
8ad65e08
MD
190 p2->p_thread->td_sp = (char *)(pcb2->pcb_esp - sizeof(void *));
191 *(void **)p2->p_thread->td_sp = (void *)cpu_heavy_restore;
984263bc
MD
192 /*
193 * pcb2->pcb_ldt: duplicated below, if necessary.
194 * pcb2->pcb_savefpu: cloned above.
195 * pcb2->pcb_flags: cloned above (always 0 here?).
196 * pcb2->pcb_onfault: cloned above (always NULL here?).
197 */
198
199#ifdef SMP
200 pcb2->pcb_mpnest = 1;
201#endif
202 /*
203 * XXX don't copy the i/o pages. this should probably be fixed.
204 */
205 pcb2->pcb_ext = 0;
206
207#ifdef USER_LDT
208 /* Copy the LDT, if necessary. */
209 if (pcb2->pcb_ldt != 0) {
210 if (flags & RFMEM) {
211 pcb2->pcb_ldt->ldt_refcnt++;
212 } else {
213 pcb2->pcb_ldt = user_ldt_alloc(pcb2,
214 pcb2->pcb_ldt->ldt_len);
215 }
216 }
217#endif
218
219 /*
220 * Now, cpu_switch() can schedule the new process.
221 * pcb_esp is loaded pointing to the cpu_switch() stack frame
222 * containing the return address when exiting cpu_switch.
223 * This will normally be to fork_trampoline(), which will have
224 * %ebx loaded with the new proc's pointer. fork_trampoline()
225 * will set up a stack to call fork_return(p, frame); to complete
226 * the return to user-mode.
227 */
228}
229
230/*
231 * Intercept the return address from a freshly forked process that has NOT
232 * been scheduled yet.
233 *
234 * This is needed to make kernel threads stay in kernel mode.
235 */
236void
237cpu_set_fork_handler(p, func, arg)
238 struct proc *p;
239 void (*func) __P((void *));
240 void *arg;
241{
242 /*
243 * Note that the trap frame follows the args, so the function
244 * is really called like this: func(arg, frame);
245 */
b7c628e4
MD
246 p->p_thread->td_pcb->pcb_esi = (int) func; /* function */
247 p->p_thread->td_pcb->pcb_ebx = (int) arg; /* first arg */
984263bc
MD
248}
249
0cfcada1
MD
250void
251cpu_set_thread_handler(thread_t td, void (*rfunc)(void), void *func, void *arg)
252{
253 td->td_pcb->pcb_esi = (int)func;
254 td->td_pcb->pcb_ebx = (int) arg;
255 td->td_switch = cpu_lwkt_switch;
256 td->td_sp -= sizeof(void *);
257 *(void **)td->td_sp = rfunc; /* exit function on return */
258 td->td_sp -= sizeof(void *);
259 *(void **)td->td_sp = cpu_kthread_restore;
260}
261
984263bc 262void
99df837e 263cpu_proc_exit(void)
984263bc 264{
99df837e 265 struct thread *td = curthread;
ae8050a4 266 struct pcb *pcb;
8ad65e08 267
99df837e 268
984263bc 269#if NNPX > 0
99df837e
MD
270 KKASSERT(td->td_proc);
271 npxexit(td->td_proc);
984263bc 272#endif /* NNPX */
ae8050a4
MD
273
274 /*
275 * Cleanup the PCB
276 */
99df837e 277 pcb = td->td_pcb;
984263bc
MD
278 if (pcb->pcb_ext != 0) {
279 /*
280 * XXX do we need to move the TSS off the allocated pages
281 * before freeing them? (not done here)
282 */
283 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
284 ctob(IOPAGES + 1));
285 pcb->pcb_ext = 0;
286 }
287#ifdef USER_LDT
288 user_ldt_free(pcb);
289#endif
290 if (pcb->pcb_flags & PCB_DBREGS) {
291 /*
292 * disable all hardware breakpoints
293 */
294 reset_dbregs();
295 pcb->pcb_flags &= ~PCB_DBREGS;
296 }
12e4aaff 297 mycpu->gd_cnt.v_swtch++;
ae8050a4 298
ae8050a4 299 crit_enter();
ae8050a4 300 lwkt_deschedule_self();
99df837e
MD
301 cpu_thread_exit();
302}
303
304/*
305 * Terminate the current thread. The caller must have already acquired
306 * the thread's rwlock and placed it on a reap list or otherwise notified
307 * a reaper of its existance. We set a special assembly switch function which
308 * releases td_rwlock after it has cleaned up the MMU state and switched
309 * out the stack.
310 *
311 * Must be caller from a critical section and with the thread descheduled.
312 */
313void
314cpu_thread_exit(void)
315{
316 curthread->td_switch = cpu_exit_switch;
8ad65e08 317 lwkt_switch();
984263bc
MD
318 panic("cpu_exit");
319}
320
99df837e
MD
321/*
322 * Process Reaper. Called after the caller has acquired the thread's
323 * rwlock and removed it from the reap list.
324 */
984263bc 325void
99df837e 326cpu_proc_wait(struct proc *p)
984263bc 327{
7e1d4bf4
MD
328 struct thread *td;
329
984263bc 330 /* drop per-process resources */
7e1d4bf4
MD
331 td = pmap_dispose_proc(p);
332 if (td)
99df837e 333 lwkt_free_thread(td);
984263bc
MD
334}
335
336/*
337 * Dump the machine specific header information at the start of a core dump.
338 */
339int
dadab5e9 340cpu_coredump(struct thread *td, struct vnode *vp, struct ucred *cred)
984263bc 341{
dadab5e9 342 struct proc *p = td->td_proc;
984263bc
MD
343 int error;
344 caddr_t tempuser;
345
dadab5e9 346 KKASSERT(p);
984263bc
MD
347 tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK);
348 if (!tempuser)
349 return EINVAL;
350
351 bzero(tempuser, ctob(UPAGES));
352 bcopy(p->p_addr, tempuser, sizeof(struct user));
353 bcopy(p->p_md.md_regs,
354 tempuser + ((caddr_t) p->p_md.md_regs - (caddr_t) p->p_addr),
355 sizeof(struct trapframe));
b7c628e4 356 bcopy(p->p_thread->td_pcb, tempuser + ((char *)p->p_thread->td_pcb - (char *)p->p_addr), sizeof(struct pcb));
984263bc
MD
357
358 error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, ctob(UPAGES),
dadab5e9 359 (off_t)0, UIO_SYSSPACE, IO_UNIT, cred, (int *)NULL, td);
984263bc
MD
360
361 free(tempuser, M_TEMP);
362
363 return error;
364}
365
366#ifdef notyet
367static void
368setredzone(pte, vaddr)
369 u_short *pte;
370 caddr_t vaddr;
371{
372/* eventually do this by setting up an expand-down stack segment
373 for ss0: selector, allowing stack access down to top of u.
374 this means though that protection violations need to be handled
375 thru a double fault exception that must do an integral task
376 switch to a known good context, within which a dump can be
377 taken. a sensible scheme might be to save the initial context
378 used by sched (that has physical memory mapped 1:1 at bottom)
379 and take the dump while still in mapped mode */
380}
381#endif
382
383/*
384 * Convert kernel VA to physical address
385 */
386u_long
387kvtop(void *addr)
388{
389 vm_offset_t va;
390
391 va = pmap_kextract((vm_offset_t)addr);
392 if (va == 0)
393 panic("kvtop: zero page frame");
394 return((int)va);
395}
396
397/*
398 * Force reset the processor by invalidating the entire address space!
399 */
400
401#ifdef SMP
402static void
403cpu_reset_proxy()
404{
405 u_int saved_mp_lock;
406
407 cpu_reset_proxy_active = 1;
408 while (cpu_reset_proxy_active == 1)
409 ; /* Wait for other cpu to disable interupts */
410 saved_mp_lock = mp_lock;
411 mp_lock = 1;
412 printf("cpu_reset_proxy: Grabbed mp lock for BSP\n");
413 cpu_reset_proxy_active = 3;
414 while (cpu_reset_proxy_active == 3)
415 ; /* Wait for other cpu to enable interrupts */
416 stop_cpus((1<<cpu_reset_proxyid));
417 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
418 DELAY(1000000);
419 cpu_reset_real();
420}
421#endif
422
423void
424cpu_reset()
425{
426#ifdef SMP
427 if (smp_active == 0) {
428 cpu_reset_real();
429 /* NOTREACHED */
430 } else {
431
432 u_int map;
433 int cnt;
434 printf("cpu_reset called on cpu#%d\n",cpuid);
435
436 map = other_cpus & ~ stopped_cpus;
437
438 if (map != 0) {
439 printf("cpu_reset: Stopping other CPUs\n");
440 stop_cpus(map); /* Stop all other CPUs */
441 }
442
443 if (cpuid == 0) {
444 DELAY(1000000);
445 cpu_reset_real();
446 /* NOTREACHED */
447 } else {
448 /* We are not BSP (CPU #0) */
449
450 cpu_reset_proxyid = cpuid;
451 cpustop_restartfunc = cpu_reset_proxy;
452 printf("cpu_reset: Restarting BSP\n");
453 started_cpus = (1<<0); /* Restart CPU #0 */
454
455 cnt = 0;
456 while (cpu_reset_proxy_active == 0 && cnt < 10000000)
457 cnt++; /* Wait for BSP to announce restart */
458 if (cpu_reset_proxy_active == 0)
459 printf("cpu_reset: Failed to restart BSP\n");
460 __asm __volatile("cli" : : : "memory");
461 cpu_reset_proxy_active = 2;
462 cnt = 0;
463 while (cpu_reset_proxy_active == 2 && cnt < 10000000)
464 cnt++; /* Do nothing */
465 if (cpu_reset_proxy_active == 2) {
466 printf("cpu_reset: BSP did not grab mp lock\n");
467 cpu_reset_real(); /* XXX: Bogus ? */
468 }
469 cpu_reset_proxy_active = 4;
470 __asm __volatile("sti" : : : "memory");
471 while (1);
472 /* NOTREACHED */
473 }
474 }
475#else
476 cpu_reset_real();
477#endif
478}
479
480static void
481cpu_reset_real()
482{
483
484#ifdef PC98
485 /*
486 * Attempt to do a CPU reset via CPU reset port.
487 */
488 disable_intr();
489 if ((inb(0x35) & 0xa0) != 0xa0) {
490 outb(0x37, 0x0f); /* SHUT0 = 0. */
491 outb(0x37, 0x0b); /* SHUT1 = 0. */
492 }
493 outb(0xf0, 0x00); /* Reset. */
494#else
495 /*
496 * Attempt to do a CPU reset via the keyboard controller,
497 * do not turn of the GateA20, as any machine that fails
498 * to do the reset here would then end up in no man's land.
499 */
500
501#if !defined(BROKEN_KEYBOARD_RESET)
502 outb(IO_KBD + 4, 0xFE);
503 DELAY(500000); /* wait 0.5 sec to see if that did it */
504 printf("Keyboard reset did not work, attempting CPU shutdown\n");
505 DELAY(1000000); /* wait 1 sec for printf to complete */
506#endif
507#endif /* PC98 */
508 /* force a shutdown by unmapping entire address space ! */
509 bzero((caddr_t) PTD, PAGE_SIZE);
510
511 /* "good night, sweet prince .... <THUNK!>" */
512 invltlb();
513 /* NOTREACHED */
514 while(1);
515}
516
517int
518grow_stack(p, sp)
519 struct proc *p;
520 u_int sp;
521{
522 int rv;
523
524 rv = vm_map_growstack (p, sp);
525 if (rv != KERN_SUCCESS)
526 return (0);
527
528 return (1);
529}
530
531SYSCTL_DECL(_vm_stats_misc);
532
533static int cnt_prezero;
534
535SYSCTL_INT(_vm_stats_misc, OID_AUTO,
536 cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, "");
537
538/*
539 * Implement the pre-zeroed page mechanism.
540 * This routine is called from the idle loop.
541 */
542
543#define ZIDLE_LO(v) ((v) * 2 / 3)
544#define ZIDLE_HI(v) ((v) * 4 / 5)
545
546int
547vm_page_zero_idle()
548{
549 static int free_rover;
550 static int zero_state;
551 vm_page_t m;
552 int s;
553
554 /*
555 * Attempt to maintain approximately 1/2 of our free pages in a
556 * PG_ZERO'd state. Add some hysteresis to (attempt to) avoid
557 * generally zeroing a page when the system is near steady-state.
558 * Otherwise we might get 'flutter' during disk I/O / IPC or
559 * fast sleeps. We also do not want to be continuously zeroing
560 * pages because doing so may flush our L1 and L2 caches too much.
561 */
562
12e4aaff 563 if (zero_state && vm_page_zero_count >= ZIDLE_LO(vmstats.v_free_count))
984263bc 564 return(0);
12e4aaff 565 if (vm_page_zero_count >= ZIDLE_HI(vmstats.v_free_count))
984263bc
MD
566 return(0);
567
568#ifdef SMP
569 if (try_mplock()) {
570#endif
571 s = splvm();
572 __asm __volatile("sti" : : : "memory");
573 zero_state = 0;
574 m = vm_page_list_find(PQ_FREE, free_rover, FALSE);
575 if (m != NULL && (m->flags & PG_ZERO) == 0) {
576 vm_page_queues[m->queue].lcnt--;
577 TAILQ_REMOVE(&vm_page_queues[m->queue].pl, m, pageq);
578 m->queue = PQ_NONE;
579 splx(s);
580 pmap_zero_page(VM_PAGE_TO_PHYS(m));
581 (void)splvm();
582 vm_page_flag_set(m, PG_ZERO);
583 m->queue = PQ_FREE + m->pc;
584 vm_page_queues[m->queue].lcnt++;
585 TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m,
586 pageq);
587 ++vm_page_zero_count;
588 ++cnt_prezero;
12e4aaff 589 if (vm_page_zero_count >= ZIDLE_HI(vmstats.v_free_count))
984263bc
MD
590 zero_state = 1;
591 }
592 free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK;
593 splx(s);
594 __asm __volatile("cli" : : : "memory");
595#ifdef SMP
596 rel_mplock();
597#endif
598 return (1);
599#ifdef SMP
600 }
601#endif
602 /*
603 * We have to enable interrupts for a moment if the try_mplock fails
604 * in order to potentially take an IPI. XXX this should be in
605 * swtch.s
606 */
607 __asm __volatile("sti; nop; cli" : : : "memory");
608 return (0);
609}
610
26a0694b 611static void
ef0fdad1
MD
612swi_vm(void *arg)
613{
984263bc
MD
614 if (busdma_swi_pending != 0)
615 busdma_swi();
616}
617
ef0fdad1
MD
618static void
619swi_vm_setup(void *arg)
620{
621 register_swi(SWI_VM, swi_vm, NULL, "swi_vm");
622}
623
624SYSINIT(vm_setup, SI_SUB_CPU, SI_ORDER_ANY, swi_vm_setup, NULL);
625
626
984263bc
MD
627/*
628 * Tell whether this address is in some physical memory region.
629 * Currently used by the kernel coredump code in order to avoid
630 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
631 * or other unpredictable behaviour.
632 */
633
634#include "isa.h"
635
636int
637is_physical_memory(addr)
638 vm_offset_t addr;
639{
640
641#if NISA > 0
642 /* The ISA ``memory hole''. */
643 if (addr >= 0xa0000 && addr < 0x100000)
644 return 0;
645#endif
646
647 /*
648 * stuff other tests for known memory-mapped devices (PCI?)
649 * here
650 */
651
652 return 1;
653}