xref: /freebsd/sys/i386/i386/vm_machdep.c (revision 729362425c09cf6b362366aabc6fb547eee8035a)
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  * $FreeBSD$
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 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
174 	pcb2->pcb_edi = 0;
175 	pcb2->pcb_esi = (int)fork_return;	/* fork_trampoline argument */
176 	pcb2->pcb_ebp = 0;
177 	pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
178 	pcb2->pcb_ebx = (int)td2;		/* fork_trampoline argument */
179 	pcb2->pcb_eip = (int)fork_trampoline;
180 	pcb2->pcb_psl = td2->td_frame->tf_eflags & ~PSL_I; /* ints disabled */
181 	/*-
182 	 * pcb2->pcb_dr*:	cloned above.
183 	 * pcb2->pcb_savefpu:	cloned above.
184 	 * pcb2->pcb_flags:	cloned above.
185 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
186 	 * pcb2->pcb_gs:	cloned above.
187 	 * pcb2->pcb_ext:	cleared below.
188 	 */
189 
190 	/*
191 	 * XXX don't copy the i/o pages.  this should probably be fixed.
192 	 */
193 	pcb2->pcb_ext = 0;
194 
195         /* Copy the LDT, if necessary. */
196 	mtx_lock_spin(&sched_lock);
197         if (mdp2->md_ldt != 0) {
198 		if (flags & RFMEM) {
199 			mdp2->md_ldt->ldt_refcnt++;
200 		} else {
201 			mdp2->md_ldt = user_ldt_alloc(mdp2,
202 			    mdp2->md_ldt->ldt_len);
203 			if (mdp2->md_ldt == NULL)
204 				panic("could not copy LDT");
205 		}
206         }
207 	mtx_unlock_spin(&sched_lock);
208 
209 	/*
210 	 * Now, cpu_switch() can schedule the new process.
211 	 * pcb_esp is loaded pointing to the cpu_switch() stack frame
212 	 * containing the return address when exiting cpu_switch.
213 	 * This will normally be to fork_trampoline(), which will have
214 	 * %ebx loaded with the new proc's pointer.  fork_trampoline()
215 	 * will set up a stack to call fork_return(p, frame); to complete
216 	 * the return to user-mode.
217 	 */
218 }
219 
220 /*
221  * Intercept the return address from a freshly forked process that has NOT
222  * been scheduled yet.
223  *
224  * This is needed to make kernel threads stay in kernel mode.
225  */
226 void
227 cpu_set_fork_handler(td, func, arg)
228 	struct thread *td;
229 	void (*func)(void *);
230 	void *arg;
231 {
232 	/*
233 	 * Note that the trap frame follows the args, so the function
234 	 * is really called like this:  func(arg, frame);
235 	 */
236 	td->td_pcb->pcb_esi = (int) func;	/* function */
237 	td->td_pcb->pcb_ebx = (int) arg;	/* first arg */
238 }
239 
240 void
241 cpu_exit(struct thread *td)
242 {
243 	struct mdproc *mdp;
244 
245 	mdp = &td->td_proc->p_md;
246 	if (mdp->md_ldt)
247 		user_ldt_free(td);
248 	reset_dbregs();
249 }
250 
251 void
252 cpu_thread_exit(struct thread *td)
253 {
254 	struct pcb *pcb = td->td_pcb;
255 #ifdef DEV_NPX
256 	npxexit(td);
257 #endif
258         if (pcb->pcb_flags & PCB_DBREGS) {
259                 /*
260                  * disable all hardware breakpoints
261                  */
262                 reset_dbregs();
263                 pcb->pcb_flags &= ~PCB_DBREGS;
264         }
265 }
266 
267 void
268 cpu_thread_clean(struct thread *td)
269 {
270 	struct pcb *pcb;
271 
272 	pcb = td->td_pcb;
273 	if (pcb->pcb_ext != 0) {
274 		/* XXXKSE  XXXSMP  not SMP SAFE.. what locks do we have? */
275 		/* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */
276 		/*
277 		 * XXX do we need to move the TSS off the allocated pages
278 		 * before freeing them?  (not done here)
279 		 */
280 		mtx_lock(&Giant);
281 		kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
282 		    ctob(IOPAGES + 1));
283 		mtx_unlock(&Giant);
284 		pcb->pcb_ext = 0;
285 	}
286 }
287 
288 void
289 cpu_sched_exit(td)
290 	register struct thread *td;
291 {
292 }
293 
294 void
295 cpu_thread_setup(struct thread *td)
296 {
297 
298 	td->td_pcb =
299 	     (struct pcb *)(td->td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;
300 	td->td_frame = (struct trapframe *)((caddr_t)td->td_pcb - 16) - 1;
301 }
302 
303 /*
304  * Initialize machine state (pcb and trap frame) for a new thread about to
305  * upcall. Pu t enough state in the new thread's PCB to get it to go back
306  * userret(), where we can intercept it again to set the return (upcall)
307  * Address and stack, along with those from upcals that are from other sources
308  * such as those generated in thread_userret() itself.
309  */
310 void
311 cpu_set_upcall(struct thread *td, void *pcb)
312 {
313 	struct pcb *pcb2;
314 
315 	/* Point the pcb to the top of the stack. */
316 	pcb2 = td->td_pcb;
317 
318 	/*
319 	 * Copy the upcall pcb.  This loads kernel regs.
320 	 * Those not loaded individually below get their default
321 	 * values here.
322 	 *
323 	 * XXXKSE It might be a good idea to simply skip this as
324 	 * the values of the other registers may be unimportant.
325 	 * This would remove any requirement for knowing the KSE
326 	 * at this time (see the matching comment below for
327 	 * more analysis) (need a good safe default).
328 	 */
329 	bcopy(pcb, pcb2, sizeof(*pcb2));
330 
331 	/*
332 	 * Create a new fresh stack for the new thread.
333 	 * The -16 is so we can expand the trapframe if we go to vm86.
334 	 * Don't forget to set this stack value into whatever supplies
335 	 * the address for the fault handlers.
336 	 * The contexts are filled in at the time we actually DO the
337 	 * upcall as only then do we know which KSE we got.
338 	 */
339 	td->td_frame = (struct trapframe *)((caddr_t)pcb2 - 16) - 1;
340 
341 	/*
342 	 * Set registers for trampoline to user mode.  Leave space for the
343 	 * return address on stack.  These are the kernel mode register values.
344 	 */
345 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pdir);
346 	pcb2->pcb_edi = 0;
347 	pcb2->pcb_esi = (int)fork_return;		    /* trampoline arg */
348 	pcb2->pcb_ebp = 0;
349 	pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */
350 	pcb2->pcb_ebx = (int)td;			    /* trampoline arg */
351 	pcb2->pcb_eip = (int)fork_trampoline;
352 	pcb2->pcb_psl &= ~(PSL_I);	/* interrupts must be disabled */
353 	/*
354 	 * If we didn't copy the pcb, we'd need to do the following registers:
355 	 * pcb2->pcb_dr*:	cloned above.
356 	 * pcb2->pcb_savefpu:	cloned above.
357 	 * pcb2->pcb_flags:	cloned above.
358 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
359 	 * pcb2->pcb_gs:	cloned above.  XXXKSE ???
360 	 * pcb2->pcb_ext:	cleared below.
361 	 */
362 	 pcb2->pcb_ext = NULL;
363 }
364 
365 /*
366  * Set that machine state for performing an upcall that has to
367  * be done in thread_userret() so that those upcalls generated
368  * in thread_userret() itself can be done as well.
369  */
370 void
371 cpu_set_upcall_kse(struct thread *td, struct kse_upcall *ku)
372 {
373 
374 	/*
375 	 * Do any extra cleaning that needs to be done.
376 	 * The thread may have optional components
377 	 * that are not present in a fresh thread.
378 	 * This may be a recycled thread so make it look
379 	 * as though it's newly allocated.
380 	 */
381 	cpu_thread_clean(td);
382 
383 	/*
384 	 * Set the trap frame to point at the beginning of the uts
385 	 * function.
386 	 */
387 	td->td_frame->tf_esp =
388 	    (int)ku->ku_stack.ss_sp + ku->ku_stack.ss_size - 16;
389 	td->td_frame->tf_eip = (int)ku->ku_func;
390 
391 	/*
392 	 * Pass the address of the mailbox for this kse to the uts
393 	 * function as a parameter on the stack.
394 	 */
395 	suword((void *)(td->td_frame->tf_esp + sizeof(void *)),
396 	    (int)ku->ku_mailbox);
397 }
398 
399 void
400 cpu_wait(p)
401 	struct proc *p;
402 {
403 }
404 
405 /*
406  * Convert kernel VA to physical address
407  */
408 vm_paddr_t
409 kvtop(void *addr)
410 {
411 	vm_paddr_t pa;
412 
413 	pa = pmap_kextract((vm_offset_t)addr);
414 	if (pa == 0)
415 		panic("kvtop: zero page frame");
416 	return (pa);
417 }
418 
419 /*
420  * Force reset the processor by invalidating the entire address space!
421  */
422 
423 #ifdef SMP
424 static void
425 cpu_reset_proxy()
426 {
427 
428 	cpu_reset_proxy_active = 1;
429 	while (cpu_reset_proxy_active == 1)
430 		;	 /* Wait for other cpu to see that we've started */
431 	stop_cpus((1<<cpu_reset_proxyid));
432 	printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
433 	DELAY(1000000);
434 	cpu_reset_real();
435 }
436 #endif
437 
438 void
439 cpu_reset()
440 {
441 #ifdef SMP
442 	if (smp_active == 0) {
443 		cpu_reset_real();
444 		/* NOTREACHED */
445 	} else {
446 
447 		u_int map;
448 		int cnt;
449 		printf("cpu_reset called on cpu#%d\n", PCPU_GET(cpuid));
450 
451 		map = PCPU_GET(other_cpus) & ~ stopped_cpus;
452 
453 		if (map != 0) {
454 			printf("cpu_reset: Stopping other CPUs\n");
455 			stop_cpus(map);		/* Stop all other CPUs */
456 		}
457 
458 		if (PCPU_GET(cpuid) == 0) {
459 			DELAY(1000000);
460 			cpu_reset_real();
461 			/* NOTREACHED */
462 		} else {
463 			/* We are not BSP (CPU #0) */
464 
465 			cpu_reset_proxyid = PCPU_GET(cpuid);
466 			cpustop_restartfunc = cpu_reset_proxy;
467 			cpu_reset_proxy_active = 0;
468 			printf("cpu_reset: Restarting BSP\n");
469 			started_cpus = (1<<0);		/* Restart CPU #0 */
470 
471 			cnt = 0;
472 			while (cpu_reset_proxy_active == 0 && cnt < 10000000)
473 				cnt++;	/* Wait for BSP to announce restart */
474 			if (cpu_reset_proxy_active == 0)
475 				printf("cpu_reset: Failed to restart BSP\n");
476 			enable_intr();
477 			cpu_reset_proxy_active = 2;
478 
479 			while (1);
480 			/* NOTREACHED */
481 		}
482 	}
483 #else
484 	cpu_reset_real();
485 #endif
486 }
487 
488 static void
489 cpu_reset_real()
490 {
491 
492 #ifdef PC98
493 	/*
494 	 * Attempt to do a CPU reset via CPU reset port.
495 	 */
496 	disable_intr();
497 	if ((inb(0x35) & 0xa0) != 0xa0) {
498 		outb(0x37, 0x0f);		/* SHUT0 = 0. */
499 		outb(0x37, 0x0b);		/* SHUT1 = 0. */
500 	}
501 	outb(0xf0, 0x00);		/* Reset. */
502 #else
503 	/*
504 	 * Attempt to do a CPU reset via the keyboard controller,
505 	 * do not turn of the GateA20, as any machine that fails
506 	 * to do the reset here would then end up in no man's land.
507 	 */
508 
509 #if !defined(BROKEN_KEYBOARD_RESET)
510 	outb(IO_KBD + 4, 0xFE);
511 	DELAY(500000);	/* wait 0.5 sec to see if that did it */
512 	printf("Keyboard reset did not work, attempting CPU shutdown\n");
513 	DELAY(1000000);	/* wait 1 sec for printf to complete */
514 #endif
515 #endif /* PC98 */
516 	/* force a shutdown by unmapping entire address space ! */
517 	bzero((caddr_t)PTD, NBPTD);
518 
519 	/* "good night, sweet prince .... <THUNK!>" */
520 	invltlb();
521 	/* NOTREACHED */
522 	while(1);
523 }
524 
525 /*
526  * Software interrupt handler for queued VM system processing.
527  */
528 void
529 swi_vm(void *dummy)
530 {
531 	if (busdma_swi_pending != 0)
532 		busdma_swi();
533 }
534 
535 /*
536  * Tell whether this address is in some physical memory region.
537  * Currently used by the kernel coredump code in order to avoid
538  * dumping the ``ISA memory hole'' which could cause indefinite hangs,
539  * or other unpredictable behaviour.
540  */
541 
542 int
543 is_physical_memory(addr)
544 	vm_offset_t addr;
545 {
546 
547 #ifdef DEV_ISA
548 	/* The ISA ``memory hole''. */
549 	if (addr >= 0xa0000 && addr < 0x100000)
550 		return 0;
551 #endif
552 
553 	/*
554 	 * stuff other tests for known memory-mapped devices (PCI?)
555 	 * here
556 	 */
557 
558 	return 1;
559 }
560