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: releng/5.1/sys/i386/i386/vm_machdep.c 115732 2003-06-02 21:37:07Z tegge $
42 */
43
44 #include "opt_npx.h"
45 #ifdef PC98
46 #include "opt_pc98.h"
47 #endif
48 #include "opt_reset.h"
49 #include "opt_isa.h"
50 #include "opt_kstack_pages.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/kse.h>
57 #include <sys/bio.h>
58 #include <sys/buf.h>
59 #include <sys/vnode.h>
60 #include <sys/vmmeter.h>
61 #include <sys/kernel.h>
62 #include <sys/ktr.h>
63 #include <sys/mutex.h>
64 #include <sys/smp.h>
65 #include <sys/sysctl.h>
66 #include <sys/unistd.h>
67
68 #include <machine/cpu.h>
69 #include <machine/md_var.h>
70 #include <machine/pcb.h>
71 #include <machine/pcb_ext.h>
72 #include <machine/vm86.h>
73
74 #include <vm/vm.h>
75 #include <vm/vm_param.h>
76 #include <sys/lock.h>
77 #include <vm/vm_kern.h>
78 #include <vm/vm_page.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_extern.h>
81
82 #include <sys/user.h>
83
84 #ifdef PC98
85 #include <pc98/pc98/pc98.h>
86 #else
87 #include <i386/isa/isa.h>
88 #endif
89
90 static void cpu_reset_real(void);
91 #ifdef SMP
92 static void cpu_reset_proxy(void);
93 static u_int cpu_reset_proxyid;
94 static volatile u_int cpu_reset_proxy_active;
95 #endif
96 extern int _ucodesel, _udatasel;
97
98 /*
99 * Finish a fork operation, with process p2 nearly set up.
100 * Copy and update the pcb, set up the stack so that the child
101 * ready to run and return to user mode.
102 */
103 void
104 cpu_fork(td1, p2, td2, flags)
105 register struct thread *td1;
106 register struct proc *p2;
107 struct thread *td2;
108 int flags;
109 {
110 register struct proc *p1;
111 struct pcb *pcb2;
112 struct mdproc *mdp2;
113 #ifdef DEV_NPX
114 register_t savecrit;
115 #endif
116
117 p1 = td1->td_proc;
118 if ((flags & RFPROC) == 0) {
119 if ((flags & RFMEM) == 0) {
120 /* unshare user LDT */
121 struct mdproc *mdp1 = &p1->p_md;
122 struct proc_ldt *pldt = mdp1->md_ldt;
123 if (pldt && pldt->ldt_refcnt > 1) {
124 pldt = user_ldt_alloc(mdp1, pldt->ldt_len);
125 if (pldt == NULL)
126 panic("could not copy LDT");
127 mdp1->md_ldt = pldt;
128 set_user_ldt(mdp1);
129 user_ldt_free(td1);
130 }
131 }
132 return;
133 }
134
135 /* Ensure that p1's pcb is up to date. */
136 #ifdef DEV_NPX
137 if (td1 == curthread)
138 td1->td_pcb->pcb_gs = rgs();
139 savecrit = intr_disable();
140 if (PCPU_GET(fpcurthread) == td1)
141 npxsave(&td1->td_pcb->pcb_save);
142 intr_restore(savecrit);
143 #endif
144
145 /* Point the pcb to the top of the stack */
146 pcb2 = (struct pcb *)(td2->td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;
147 td2->td_pcb = pcb2;
148
149 /* Copy p1's pcb */
150 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
151
152 /* Point mdproc and then copy over td1's contents */
153 mdp2 = &p2->p_md;
154 bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
155
156 /*
157 * Create a new fresh stack for the new process.
158 * Copy the trap frame for the return to user mode as if from a
159 * syscall. This copies most of the user mode register values.
160 * The -16 is so we can expand the trapframe if we go to vm86.
161 */
162 td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1;
163 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
164
165 td2->td_frame->tf_eax = 0; /* Child returns zero */
166 td2->td_frame->tf_eflags &= ~PSL_C; /* success */
167 td2->td_frame->tf_edx = 1;
168
169 /*
170 * Set registers for trampoline to user mode. Leave space for the
171 * return address on stack. These are the kernel mode register values.
172 */
173 #ifdef PAE
174 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdpt);
175 #else
176 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
177 #endif
178 pcb2->pcb_edi = 0;
179 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */
180 pcb2->pcb_ebp = 0;
181 pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
182 pcb2->pcb_ebx = (int)td2; /* fork_trampoline argument */
183 pcb2->pcb_eip = (int)fork_trampoline;
184 pcb2->pcb_psl = PSL_KERNEL; /* ints disabled */
185 pcb2->pcb_gs = rgs();
186 /*-
187 * pcb2->pcb_dr*: cloned above.
188 * pcb2->pcb_savefpu: cloned above.
189 * pcb2->pcb_flags: cloned above.
190 * pcb2->pcb_onfault: cloned above (always NULL here?).
191 * pcb2->pcb_gs: cloned above.
192 * pcb2->pcb_ext: cleared below.
193 */
194
195 /*
196 * XXX don't copy the i/o pages. this should probably be fixed.
197 */
198 pcb2->pcb_ext = 0;
199
200 /* Copy the LDT, if necessary. */
201 mtx_lock_spin(&sched_lock);
202 if (mdp2->md_ldt != 0) {
203 if (flags & RFMEM) {
204 mdp2->md_ldt->ldt_refcnt++;
205 } else {
206 mdp2->md_ldt = user_ldt_alloc(mdp2,
207 mdp2->md_ldt->ldt_len);
208 if (mdp2->md_ldt == NULL)
209 panic("could not copy LDT");
210 }
211 }
212 mtx_unlock_spin(&sched_lock);
213
214 /*
215 * Now, cpu_switch() can schedule the new process.
216 * pcb_esp is loaded pointing to the cpu_switch() stack frame
217 * containing the return address when exiting cpu_switch.
218 * This will normally be to fork_trampoline(), which will have
219 * %ebx loaded with the new proc's pointer. fork_trampoline()
220 * will set up a stack to call fork_return(p, frame); to complete
221 * the return to user-mode.
222 */
223 }
224
225 /*
226 * Intercept the return address from a freshly forked process that has NOT
227 * been scheduled yet.
228 *
229 * This is needed to make kernel threads stay in kernel mode.
230 */
231 void
232 cpu_set_fork_handler(td, func, arg)
233 struct thread *td;
234 void (*func)(void *);
235 void *arg;
236 {
237 /*
238 * Note that the trap frame follows the args, so the function
239 * is really called like this: func(arg, frame);
240 */
241 td->td_pcb->pcb_esi = (int) func; /* function */
242 td->td_pcb->pcb_ebx = (int) arg; /* first arg */
243 }
244
245 void
246 cpu_exit(struct thread *td)
247 {
248 struct mdproc *mdp;
249
250 /* Reset pc->pcb_gs and %gs before possibly invalidating it. */
251 mdp = &td->td_proc->p_md;
252 if (mdp->md_ldt) {
253 td->td_pcb->pcb_gs = _udatasel;
254 load_gs(_udatasel);
255 user_ldt_free(td);
256 }
257 reset_dbregs();
258 }
259
260 void
261 cpu_thread_exit(struct thread *td)
262 {
263 struct pcb *pcb = td->td_pcb;
264 #ifdef DEV_NPX
265 npxexit(td);
266 #endif
267 if (pcb->pcb_flags & PCB_DBREGS) {
268 /*
269 * disable all hardware breakpoints
270 */
271 reset_dbregs();
272 pcb->pcb_flags &= ~PCB_DBREGS;
273 }
274 }
275
276 void
277 cpu_thread_clean(struct thread *td)
278 {
279 struct pcb *pcb;
280
281 pcb = td->td_pcb;
282 if (pcb->pcb_ext != 0) {
283 /* XXXKSE XXXSMP not SMP SAFE.. what locks do we have? */
284 /* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */
285 /*
286 * XXX do we need to move the TSS off the allocated pages
287 * before freeing them? (not done here)
288 */
289 mtx_lock(&Giant);
290 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
291 ctob(IOPAGES + 1));
292 mtx_unlock(&Giant);
293 pcb->pcb_ext = 0;
294 }
295 }
296
297 void
298 cpu_sched_exit(td)
299 register struct thread *td;
300 {
301 }
302
303 void
304 cpu_thread_setup(struct thread *td)
305 {
306
307 td->td_pcb =
308 (struct pcb *)(td->td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;
309 td->td_frame = (struct trapframe *)((caddr_t)td->td_pcb - 16) - 1;
310 td->td_pcb->pcb_ext = NULL;
311 }
312
313 /*
314 * Initialize machine state (pcb and trap frame) for a new thread about to
315 * upcall. Pu t enough state in the new thread's PCB to get it to go back
316 * userret(), where we can intercept it again to set the return (upcall)
317 * Address and stack, along with those from upcals that are from other sources
318 * such as those generated in thread_userret() itself.
319 */
320 void
321 cpu_set_upcall(struct thread *td, void *pcb)
322 {
323 struct pcb *pcb2;
324
325 /* Point the pcb to the top of the stack. */
326 pcb2 = td->td_pcb;
327
328 /*
329 * Copy the upcall pcb. This loads kernel regs.
330 * Those not loaded individually below get their default
331 * values here.
332 *
333 * XXXKSE It might be a good idea to simply skip this as
334 * the values of the other registers may be unimportant.
335 * This would remove any requirement for knowing the KSE
336 * at this time (see the matching comment below for
337 * more analysis) (need a good safe default).
338 */
339 bcopy(pcb, pcb2, sizeof(*pcb2));
340
341 /*
342 * Create a new fresh stack for the new thread.
343 * The -16 is so we can expand the trapframe if we go to vm86.
344 * Don't forget to set this stack value into whatever supplies
345 * the address for the fault handlers.
346 * The contexts are filled in at the time we actually DO the
347 * upcall as only then do we know which KSE we got.
348 */
349 td->td_frame = (struct trapframe *)((caddr_t)pcb2 - 16) - 1;
350
351 /*
352 * Set registers for trampoline to user mode. Leave space for the
353 * return address on stack. These are the kernel mode register values.
354 */
355 #ifdef PAE
356 pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pdpt);
357 #else
358 pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pdir);
359 #endif
360 pcb2->pcb_edi = 0;
361 pcb2->pcb_esi = (int)fork_return; /* trampoline arg */
362 pcb2->pcb_ebp = 0;
363 pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */
364 pcb2->pcb_ebx = (int)td; /* trampoline arg */
365 pcb2->pcb_eip = (int)fork_trampoline;
366 pcb2->pcb_psl &= ~(PSL_I); /* interrupts must be disabled */
367 pcb2->pcb_gs = rgs();
368 /*
369 * If we didn't copy the pcb, we'd need to do the following registers:
370 * pcb2->pcb_dr*: cloned above.
371 * pcb2->pcb_savefpu: cloned above.
372 * pcb2->pcb_flags: cloned above.
373 * pcb2->pcb_onfault: cloned above (always NULL here?).
374 * pcb2->pcb_gs: cloned above. XXXKSE ???
375 * pcb2->pcb_ext: cleared below.
376 */
377 pcb2->pcb_ext = NULL;
378 }
379
380 /*
381 * Set that machine state for performing an upcall that has to
382 * be done in thread_userret() so that those upcalls generated
383 * in thread_userret() itself can be done as well.
384 */
385 void
386 cpu_set_upcall_kse(struct thread *td, struct kse_upcall *ku)
387 {
388
389 /*
390 * Do any extra cleaning that needs to be done.
391 * The thread may have optional components
392 * that are not present in a fresh thread.
393 * This may be a recycled thread so make it look
394 * as though it's newly allocated.
395 */
396 cpu_thread_clean(td);
397
398 /*
399 * Set the trap frame to point at the beginning of the uts
400 * function.
401 */
402 td->td_frame->tf_esp =
403 (int)ku->ku_stack.ss_sp + ku->ku_stack.ss_size - 16;
404 td->td_frame->tf_eip = (int)ku->ku_func;
405
406 /*
407 * Pass the address of the mailbox for this kse to the uts
408 * function as a parameter on the stack.
409 */
410 suword((void *)(td->td_frame->tf_esp + sizeof(void *)),
411 (int)ku->ku_mailbox);
412 }
413
414 void
415 cpu_wait(p)
416 struct proc *p;
417 {
418 }
419
420 /*
421 * Convert kernel VA to physical address
422 */
423 vm_paddr_t
424 kvtop(void *addr)
425 {
426 vm_paddr_t pa;
427
428 pa = pmap_kextract((vm_offset_t)addr);
429 if (pa == 0)
430 panic("kvtop: zero page frame");
431 return (pa);
432 }
433
434 /*
435 * Force reset the processor by invalidating the entire address space!
436 */
437
438 #ifdef SMP
439 static void
440 cpu_reset_proxy()
441 {
442
443 cpu_reset_proxy_active = 1;
444 while (cpu_reset_proxy_active == 1)
445 ; /* Wait for other cpu to see that we've started */
446 stop_cpus((1<<cpu_reset_proxyid));
447 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
448 DELAY(1000000);
449 cpu_reset_real();
450 }
451 #endif
452
453 void
454 cpu_reset()
455 {
456 #ifdef SMP
457 if (smp_active == 0) {
458 cpu_reset_real();
459 /* NOTREACHED */
460 } else {
461
462 u_int map;
463 int cnt;
464 printf("cpu_reset called on cpu#%d\n", PCPU_GET(cpuid));
465
466 map = PCPU_GET(other_cpus) & ~ stopped_cpus;
467
468 if (map != 0) {
469 printf("cpu_reset: Stopping other CPUs\n");
470 stop_cpus(map); /* Stop all other CPUs */
471 }
472
473 if (PCPU_GET(cpuid) == 0) {
474 DELAY(1000000);
475 cpu_reset_real();
476 /* NOTREACHED */
477 } else {
478 /* We are not BSP (CPU #0) */
479
480 cpu_reset_proxyid = PCPU_GET(cpuid);
481 cpustop_restartfunc = cpu_reset_proxy;
482 cpu_reset_proxy_active = 0;
483 printf("cpu_reset: Restarting BSP\n");
484 started_cpus = (1<<0); /* Restart CPU #0 */
485
486 cnt = 0;
487 while (cpu_reset_proxy_active == 0 && cnt < 10000000)
488 cnt++; /* Wait for BSP to announce restart */
489 if (cpu_reset_proxy_active == 0)
490 printf("cpu_reset: Failed to restart BSP\n");
491 enable_intr();
492 cpu_reset_proxy_active = 2;
493
494 while (1);
495 /* NOTREACHED */
496 }
497 }
498 #else
499 cpu_reset_real();
500 #endif
501 }
502
503 static void
504 cpu_reset_real()
505 {
506
507 #ifdef PC98
508 /*
509 * Attempt to do a CPU reset via CPU reset port.
510 */
511 disable_intr();
512 if ((inb(0x35) & 0xa0) != 0xa0) {
513 outb(0x37, 0x0f); /* SHUT0 = 0. */
514 outb(0x37, 0x0b); /* SHUT1 = 0. */
515 }
516 outb(0xf0, 0x00); /* Reset. */
517 #else
518 /*
519 * Attempt to do a CPU reset via the keyboard controller,
520 * do not turn of the GateA20, as any machine that fails
521 * to do the reset here would then end up in no man's land.
522 */
523
524 #if !defined(BROKEN_KEYBOARD_RESET)
525 outb(IO_KBD + 4, 0xFE);
526 DELAY(500000); /* wait 0.5 sec to see if that did it */
527 printf("Keyboard reset did not work, attempting CPU shutdown\n");
528 DELAY(1000000); /* wait 1 sec for printf to complete */
529 #endif
530 #endif /* PC98 */
531 /* force a shutdown by unmapping entire address space ! */
532 bzero((caddr_t)PTD, NBPTD);
533
534 /* "good night, sweet prince .... <THUNK!>" */
535 invltlb();
536 /* NOTREACHED */
537 while(1);
538 }
539
540 /*
541 * Software interrupt handler for queued VM system processing.
542 */
543 void
544 swi_vm(void *dummy)
545 {
546 if (busdma_swi_pending != 0)
547 busdma_swi();
548 }
549
550 /*
551 * Tell whether this address is in some physical memory region.
552 * Currently used by the kernel coredump code in order to avoid
553 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
554 * or other unpredictable behaviour.
555 */
556
557 int
558 is_physical_memory(addr)
559 vm_offset_t addr;
560 {
561
562 #ifdef DEV_ISA
563 /* The ISA ``memory hole''. */
564 if (addr >= 0xa0000 && addr < 0x100000)
565 return 0;
566 #endif
567
568 /*
569 * stuff other tests for known memory-mapped devices (PCI?)
570 * here
571 */
572
573 return 1;
574 }
Cache object: c5acb6abcf28d47d20d78e2084b7a641
|