xref: /freebsd/sys/i386/i386/vm_machdep.c (revision a3e8fd0b7f663db7eafff527d5c3ca3bcfa8a537)
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  * quick version of vm_fault
100  */
101 int
102 vm_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  */
120 void
121 cpu_fork(td1, p2, td2, flags)
122 	register struct thread *td1;
123 	register struct proc *p2;
124 	struct thread *td2;
125 	int flags;
126 {
127 	register struct proc *p1;
128 	struct pcb *pcb2;
129 	struct mdproc *mdp2;
130 #ifdef DEV_NPX
131 	register_t savecrit;
132 #endif
133 
134 	p1 = td1->td_proc;
135 	if ((flags & RFPROC) == 0) {
136 		if ((flags & RFMEM) == 0) {
137 			/* unshare user LDT */
138 			struct mdproc *mdp1 = &p1->p_md;
139 			struct proc_ldt *pldt = mdp1->md_ldt;
140 			if (pldt && pldt->ldt_refcnt > 1) {
141 				pldt = user_ldt_alloc(mdp1, pldt->ldt_len);
142 				if (pldt == NULL)
143 					panic("could not copy LDT");
144 				mdp1->md_ldt = pldt;
145 				set_user_ldt(mdp1);
146 				user_ldt_free(td1);
147 			}
148 		}
149 		return;
150 	}
151 
152 	/* Ensure that p1's pcb is up to date. */
153 #ifdef DEV_NPX
154 	if (td1 == curthread)
155 		td1->td_pcb->pcb_gs = rgs();
156 	savecrit = intr_disable();
157 	if (PCPU_GET(fpcurthread) == td1)
158 		npxsave(&td1->td_pcb->pcb_save);
159 	intr_restore(savecrit);
160 #endif
161 
162 	/* Point the pcb to the top of the stack */
163 	pcb2 = (struct pcb *)(td2->td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;
164 	td2->td_pcb = pcb2;
165 
166 	/* Copy p1's pcb */
167 	bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
168 
169 	/* Point mdproc and then copy over td1's contents */
170 	mdp2 = &p2->p_md;
171 	bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
172 
173 	/*
174 	 * Create a new fresh stack for the new process.
175 	 * Copy the trap frame for the return to user mode as if from a
176 	 * syscall.  This copies most of the user mode register values.
177 	 * The -16 is so we can expand the trapframe if we go to vm86.
178 	 */
179 	td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1;
180 	bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
181 
182 	td2->td_frame->tf_eax = 0;		/* Child returns zero */
183 	td2->td_frame->tf_eflags &= ~PSL_C;	/* success */
184 	td2->td_frame->tf_edx = 1;
185 
186 	/*
187 	 * Set registers for trampoline to user mode.  Leave space for the
188 	 * return address on stack.  These are the kernel mode register values.
189 	 */
190 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
191 	pcb2->pcb_edi = 0;
192 	pcb2->pcb_esi = (int)fork_return;	/* fork_trampoline argument */
193 	pcb2->pcb_ebp = 0;
194 	pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
195 	pcb2->pcb_ebx = (int)td2;		/* fork_trampoline argument */
196 	pcb2->pcb_eip = (int)fork_trampoline;
197 	pcb2->pcb_psl = td2->td_frame->tf_eflags & ~PSL_I; /* ints disabled */
198 	/*-
199 	 * pcb2->pcb_dr*:	cloned above.
200 	 * pcb2->pcb_savefpu:	cloned above.
201 	 * pcb2->pcb_flags:	cloned above.
202 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
203 	 * pcb2->pcb_gs:	cloned above.
204 	 * pcb2->pcb_ext:	cleared below.
205 	 */
206 
207 	/*
208 	 * XXX don't copy the i/o pages.  this should probably be fixed.
209 	 */
210 	pcb2->pcb_ext = 0;
211 
212         /* Copy the LDT, if necessary. */
213 	mtx_lock_spin(&sched_lock);
214         if (mdp2->md_ldt != 0) {
215 		if (flags & RFMEM) {
216 			mdp2->md_ldt->ldt_refcnt++;
217 		} else {
218 			mdp2->md_ldt = user_ldt_alloc(mdp2,
219 			    mdp2->md_ldt->ldt_len);
220 			if (mdp2->md_ldt == NULL)
221 				panic("could not copy LDT");
222 		}
223         }
224 	mtx_unlock_spin(&sched_lock);
225 
226 	/*
227 	 * Now, cpu_switch() can schedule the new process.
228 	 * pcb_esp is loaded pointing to the cpu_switch() stack frame
229 	 * containing the return address when exiting cpu_switch.
230 	 * This will normally be to fork_trampoline(), which will have
231 	 * %ebx loaded with the new proc's pointer.  fork_trampoline()
232 	 * will set up a stack to call fork_return(p, frame); to complete
233 	 * the return to user-mode.
234 	 */
235 }
236 
237 /*
238  * Intercept the return address from a freshly forked process that has NOT
239  * been scheduled yet.
240  *
241  * This is needed to make kernel threads stay in kernel mode.
242  */
243 void
244 cpu_set_fork_handler(td, func, arg)
245 	struct thread *td;
246 	void (*func)(void *);
247 	void *arg;
248 {
249 	/*
250 	 * Note that the trap frame follows the args, so the function
251 	 * is really called like this:  func(arg, frame);
252 	 */
253 	td->td_pcb->pcb_esi = (int) func;	/* function */
254 	td->td_pcb->pcb_ebx = (int) arg;	/* first arg */
255 }
256 
257 void
258 cpu_exit(struct thread *td)
259 {
260 	struct mdproc *mdp;
261 
262 	mdp = &td->td_proc->p_md;
263 	if (mdp->md_ldt)
264 		user_ldt_free(td);
265 	reset_dbregs();
266 }
267 
268 void
269 cpu_thread_exit(struct thread *td)
270 {
271 	struct pcb *pcb = td->td_pcb;
272 #ifdef DEV_NPX
273 	npxexit(td);
274 #endif
275 	if (pcb->pcb_ext != 0) {
276 		/* XXXKSE  XXXSMP  not SMP SAFE.. what locks do we have? */
277 		/* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */
278 	        /*
279 		 * XXX do we need to move the TSS off the allocated pages
280 		 * before freeing them?  (not done here)
281 		 */
282 		kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
283 		    ctob(IOPAGES + 1));
284 		pcb->pcb_ext = 0;
285 	}
286         if (pcb->pcb_flags & PCB_DBREGS) {
287                 /*
288                  * disable all hardware breakpoints
289                  */
290                 reset_dbregs();
291                 pcb->pcb_flags &= ~PCB_DBREGS;
292         }
293 }
294 
295 void
296 cpu_sched_exit(td)
297 	register struct thread *td;
298 {
299 }
300 
301 void
302 cpu_thread_setup(struct thread *td)
303 {
304 
305 	td->td_pcb =
306 	     (struct pcb *)(td->td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;
307 	td->td_frame = (struct trapframe *)((caddr_t)td->td_pcb - 16) - 1;
308 }
309 
310 /*
311  * Initialize machine state (pcb and trap frame) for a new thread about to
312  * upcall. Pu t enough state in the new thread's PCB to get it to go back
313  * userret(), where we can intercept it again to set the return (upcall)
314  * Address and stack, along with those from upcals that are from other sources
315  * such as those generated in thread_userret() itself.
316  */
317 void
318 cpu_set_upcall(struct thread *td, void *pcb)
319 {
320 	struct pcb *pcb2;
321 
322 	td->td_flags |= TDF_UPCALLING;
323 
324 	/* Point the pcb to the top of the stack. */
325 	pcb2 = td->td_pcb;
326 
327 	/*
328 	 * Copy the upcall pcb.  This loads kernel regs.
329 	 * Those not loaded individually below get their default
330 	 * values here.
331 	 *
332 	 * XXXKSE It might be a good idea to simply skip this as
333 	 * the values of the other registers may be unimportant.
334 	 * This would remove any requirement for knowing the KSE
335 	 * at this time (see the matching comment below for
336 	 * more analysis) (need a good safe default).
337 	 */
338 	bcopy(pcb, pcb2, sizeof(*pcb2));
339 
340 	/*
341 	 * Create a new fresh stack for the new thread.
342 	 * The -16 is so we can expand the trapframe if we go to vm86.
343 	 * Don't forget to set this stack value into whatever supplies
344 	 * the address for the fault handlers.
345 	 * The contexts are filled in at the time we actually DO the
346 	 * upcall as only then do we know which KSE we got.
347 	 */
348 	td->td_frame = (struct trapframe *)((caddr_t)pcb2 - 16) - 1;
349 
350 	/*
351 	 * Set registers for trampoline to user mode.  Leave space for the
352 	 * return address on stack.  These are the kernel mode register values.
353 	 */
354 	pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pdir);
355 	pcb2->pcb_edi = 0;
356 	pcb2->pcb_esi = (int)fork_return;		    /* trampoline arg */
357 	pcb2->pcb_ebp = 0;
358 	pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */
359 	pcb2->pcb_ebx = (int)td;			    /* trampoline arg */
360 	pcb2->pcb_eip = (int)fork_trampoline;
361 	pcb2->pcb_psl &= ~(PSL_I);	/* interrupts must be disabled */
362 	/*
363 	 * If we didn't copy the pcb, we'd need to do the following registers:
364 	 * pcb2->pcb_dr*:	cloned above.
365 	 * pcb2->pcb_savefpu:	cloned above.
366 	 * pcb2->pcb_flags:	cloned above.
367 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
368 	 * pcb2->pcb_gs:	cloned above.  XXXKSE ???
369 	 * pcb2->pcb_ext:	cleared below.
370 	 */
371 	 pcb2->pcb_ext = NULL;
372 }
373 
374 /*
375  * Set that machine state for performing an upcall that has to
376  * be done in thread_userret() so that those upcalls generated
377  * in thread_userret() itself can be done as well.
378  */
379 void
380 cpu_set_upcall_kse(struct thread *td, struct kse *ke)
381 {
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)ke->ke_stack.ss_sp + ke->ke_stack.ss_size - 16;
389 	td->td_frame->tf_eip = (int)ke->ke_upcall;
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)ke->ke_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 u_long
409 kvtop(void *addr)
410 {
411 	vm_offset_t va;
412 
413 	va = pmap_kextract((vm_offset_t)addr);
414 	if (va == 0)
415 		panic("kvtop: zero page frame");
416 	return((int)va);
417 }
418 
419 /*
420  * Map an IO request into kernel virtual address space.
421  *
422  * All requests are (re)mapped into kernel VA space.
423  * Notice that we use b_bufsize for the size of the buffer
424  * to be mapped.  b_bcount might be modified by the driver.
425  */
426 void
427 vmapbuf(bp)
428 	register struct buf *bp;
429 {
430 	register caddr_t addr, kva;
431 	vm_offset_t pa;
432 	int pidx;
433 	struct vm_page *m;
434 
435 	GIANT_REQUIRED;
436 
437 	if ((bp->b_flags & B_PHYS) == 0)
438 		panic("vmapbuf");
439 
440 	for (addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data), pidx = 0;
441 	     addr < bp->b_data + bp->b_bufsize;
442 	     addr += PAGE_SIZE, pidx++) {
443 		/*
444 		 * Do the vm_fault if needed; do the copy-on-write thing
445 		 * when reading stuff off device into memory.
446 		 */
447 		vm_fault_quick((addr >= bp->b_data) ? addr : bp->b_data,
448 			(bp->b_iocmd == BIO_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ);
449 		pa = trunc_page(pmap_kextract((vm_offset_t) addr));
450 		if (pa == 0)
451 			panic("vmapbuf: page not present");
452 		m = PHYS_TO_VM_PAGE(pa);
453 		vm_page_hold(m);
454 		bp->b_pages[pidx] = m;
455 	}
456 	if (pidx > btoc(MAXPHYS))
457 		panic("vmapbuf: mapped more than MAXPHYS");
458 	pmap_qenter((vm_offset_t)bp->b_saveaddr, bp->b_pages, pidx);
459 
460 	kva = bp->b_saveaddr;
461 	bp->b_npages = pidx;
462 	bp->b_saveaddr = bp->b_data;
463 	bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK);
464 }
465 
466 /*
467  * Free the io map PTEs associated with this IO operation.
468  * We also invalidate the TLB entries and restore the original b_addr.
469  */
470 void
471 vunmapbuf(bp)
472 	register struct buf *bp;
473 {
474 	int pidx;
475 	int npages;
476 	vm_page_t *m;
477 
478 	GIANT_REQUIRED;
479 
480 	if ((bp->b_flags & B_PHYS) == 0)
481 		panic("vunmapbuf");
482 
483 	npages = bp->b_npages;
484 	pmap_qremove(trunc_page((vm_offset_t)bp->b_data),
485 		     npages);
486 	m = bp->b_pages;
487 	for (pidx = 0; pidx < npages; pidx++)
488 		vm_page_unhold(*m++);
489 
490 	bp->b_data = bp->b_saveaddr;
491 }
492 
493 /*
494  * Force reset the processor by invalidating the entire address space!
495  */
496 
497 #ifdef SMP
498 static void
499 cpu_reset_proxy()
500 {
501 
502 	cpu_reset_proxy_active = 1;
503 	while (cpu_reset_proxy_active == 1)
504 		;	 /* Wait for other cpu to see that we've started */
505 	stop_cpus((1<<cpu_reset_proxyid));
506 	printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
507 	DELAY(1000000);
508 	cpu_reset_real();
509 }
510 #endif
511 
512 void
513 cpu_reset()
514 {
515 #ifdef SMP
516 	if (smp_active == 0) {
517 		cpu_reset_real();
518 		/* NOTREACHED */
519 	} else {
520 
521 		u_int map;
522 		int cnt;
523 		printf("cpu_reset called on cpu#%d\n", PCPU_GET(cpuid));
524 
525 		map = PCPU_GET(other_cpus) & ~ stopped_cpus;
526 
527 		if (map != 0) {
528 			printf("cpu_reset: Stopping other CPUs\n");
529 			stop_cpus(map);		/* Stop all other CPUs */
530 		}
531 
532 		if (PCPU_GET(cpuid) == 0) {
533 			DELAY(1000000);
534 			cpu_reset_real();
535 			/* NOTREACHED */
536 		} else {
537 			/* We are not BSP (CPU #0) */
538 
539 			cpu_reset_proxyid = PCPU_GET(cpuid);
540 			cpustop_restartfunc = cpu_reset_proxy;
541 			cpu_reset_proxy_active = 0;
542 			printf("cpu_reset: Restarting BSP\n");
543 			started_cpus = (1<<0);		/* Restart CPU #0 */
544 
545 			cnt = 0;
546 			while (cpu_reset_proxy_active == 0 && cnt < 10000000)
547 				cnt++;	/* Wait for BSP to announce restart */
548 			if (cpu_reset_proxy_active == 0)
549 				printf("cpu_reset: Failed to restart BSP\n");
550 			enable_intr();
551 			cpu_reset_proxy_active = 2;
552 
553 			while (1);
554 			/* NOTREACHED */
555 		}
556 	}
557 #else
558 	cpu_reset_real();
559 #endif
560 }
561 
562 static void
563 cpu_reset_real()
564 {
565 
566 #ifdef PC98
567 	/*
568 	 * Attempt to do a CPU reset via CPU reset port.
569 	 */
570 	disable_intr();
571 	if ((inb(0x35) & 0xa0) != 0xa0) {
572 		outb(0x37, 0x0f);		/* SHUT0 = 0. */
573 		outb(0x37, 0x0b);		/* SHUT1 = 0. */
574 	}
575 	outb(0xf0, 0x00);		/* Reset. */
576 #else
577 	/*
578 	 * Attempt to do a CPU reset via the keyboard controller,
579 	 * do not turn of the GateA20, as any machine that fails
580 	 * to do the reset here would then end up in no man's land.
581 	 */
582 
583 #if !defined(BROKEN_KEYBOARD_RESET)
584 	outb(IO_KBD + 4, 0xFE);
585 	DELAY(500000);	/* wait 0.5 sec to see if that did it */
586 	printf("Keyboard reset did not work, attempting CPU shutdown\n");
587 	DELAY(1000000);	/* wait 1 sec for printf to complete */
588 #endif
589 #endif /* PC98 */
590 	/* force a shutdown by unmapping entire address space ! */
591 	bzero((caddr_t) PTD, PAGE_SIZE);
592 
593 	/* "good night, sweet prince .... <THUNK!>" */
594 	invltlb();
595 	/* NOTREACHED */
596 	while(1);
597 }
598 
599 /*
600  * Software interrupt handler for queued VM system processing.
601  */
602 void
603 swi_vm(void *dummy)
604 {
605 	if (busdma_swi_pending != 0)
606 		busdma_swi();
607 }
608 
609 /*
610  * Tell whether this address is in some physical memory region.
611  * Currently used by the kernel coredump code in order to avoid
612  * dumping the ``ISA memory hole'' which could cause indefinite hangs,
613  * or other unpredictable behaviour.
614  */
615 
616 int
617 is_physical_memory(addr)
618 	vm_offset_t addr;
619 {
620 
621 #ifdef DEV_ISA
622 	/* The ISA ``memory hole''. */
623 	if (addr >= 0xa0000 && addr < 0x100000)
624 		return 0;
625 #endif
626 
627 	/*
628 	 * stuff other tests for known memory-mapped devices (PCI?)
629 	 * here
630 	 */
631 
632 	return 1;
633 }
634