vm_machdep.c revision 83223 
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: head/sys/amd64/amd64/vm_machdep.c 83223 2001-09-08 22:18:58Z peter $
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_upages.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/bio.h>
57#include <sys/buf.h>
58#include <sys/vnode.h>
59#include <sys/vmmeter.h>
60#include <sys/kernel.h>
61#include <sys/ktr.h>
62#include <sys/mutex.h>
63#include <sys/smp.h>
64#include <sys/sysctl.h>
65#include <sys/unistd.h>
66
67#include <machine/cpu.h>
68#include <machine/md_var.h>
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>
82
83#ifdef PC98
84#include <pc98/pc98/pc98.h>
85#else
86#include <i386/isa/isa.h>
87#endif
88
89static void	cpu_reset_real __P((void));
90#ifdef SMP
91static void	cpu_reset_proxy __P((void));
92static u_int	cpu_reset_proxyid;
93static volatile u_int	cpu_reset_proxy_active;
94#endif
95extern int	_ucodesel, _udatasel;
96
97/*
98 * quick version of vm_fault
99 */
100int
101vm_fault_quick(v, prot)
102	caddr_t v;
103	int prot;
104{
105	int r;
106
107	if (prot & VM_PROT_WRITE)
108		r = subyte(v, fubyte(v));
109	else
110		r = fubyte(v);
111	return(r);
112}
113
114/*
115 * Finish a fork operation, with process p2 nearly set up.
116 * Copy and update the pcb, set up the stack so that the child
117 * ready to run and return to user mode.
118 */
119void
120cpu_fork(p1, p2, flags)
121	register struct proc *p1, *p2;
122	int flags;
123{
124	struct pcb *pcb2;
125#ifdef DEV_NPX
126	int savecrit;
127#endif
128
129	if ((flags & RFPROC) == 0) {
130		if ((flags & RFMEM) == 0) {
131			/* unshare user LDT */
132			struct pcb *pcb1 = &p1->p_addr->u_pcb;
133			struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt;
134			if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) {
135				pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len);
136				if (pcb_ldt == NULL)
137					panic("could not copy LDT");
138				pcb1->pcb_ldt = pcb_ldt;
139				set_user_ldt(pcb1);
140				user_ldt_free(pcb1);
141			}
142		}
143		return;
144	}
145
146	/* Ensure that p1's pcb is up to date. */
147#ifdef DEV_NPX
148	if (p1 == curproc)
149		p1->p_addr->u_pcb.pcb_gs = rgs();
150	savecrit = critical_enter();
151	if (PCPU_GET(npxproc) == p1)
152		npxsave(&p1->p_addr->u_pcb.pcb_save);
153	critical_exit(savecrit);
154#endif
155
156	/* Copy p1's pcb. */
157	p2->p_addr->u_pcb = p1->p_addr->u_pcb;
158	pcb2 = &p2->p_addr->u_pcb;
159
160	/*
161	 * Create a new fresh stack for the new process.
162	 * Copy the trap frame for the return to user mode as if from a
163	 * syscall.  This copies most of the user mode register values.
164	 */
165	p2->p_frame = (struct trapframe *)
166			   ((int)p2->p_addr + UPAGES * PAGE_SIZE - 16) - 1;
167	bcopy(p1->p_frame, p2->p_frame, sizeof(struct trapframe));
168
169	p2->p_frame->tf_eax = 0;		/* Child returns zero */
170	p2->p_frame->tf_eflags &= ~PSL_C;	/* success */
171	p2->p_frame->tf_edx = 1;
172
173	/*
174	 * Set registers for trampoline to user mode.  Leave space for the
175	 * return address on stack.  These are the kernel mode register values.
176	 */
177	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
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)p2->p_frame - sizeof(void *);
182	pcb2->pcb_ebx = (int)p2;		/* fork_trampoline argument */
183	pcb2->pcb_eip = (int)fork_trampoline;
184	/*-
185	 * pcb2->pcb_dr*:	cloned above.
186	 * pcb2->pcb_ldt:	duplicated below, if necessary.
187	 * pcb2->pcb_savefpu:	cloned above.
188	 * pcb2->pcb_flags:	cloned above.
189	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
190	 * pcb2->pcb_gs:	cloned above.
191	 * pcb2->pcb_ext:	cleared below.
192	 */
193
194	/*
195	 * XXX don't copy the i/o pages.  this should probably be fixed.
196	 */
197	pcb2->pcb_ext = 0;
198
199        /* Copy the LDT, if necessary. */
200	mtx_lock_spin(&sched_lock);
201        if (pcb2->pcb_ldt != 0) {
202		if (flags & RFMEM) {
203			pcb2->pcb_ldt->ldt_refcnt++;
204		} else {
205			pcb2->pcb_ldt = user_ldt_alloc(pcb2,
206				pcb2->pcb_ldt->ldt_len);
207			if (pcb2->pcb_ldt == NULL)
208				panic("could not copy LDT");
209		}
210        }
211	mtx_unlock_spin(&sched_lock);
212
213	/*
214	 * Now, cpu_switch() can schedule the new process.
215	 * pcb_esp is loaded pointing to the cpu_switch() stack frame
216	 * containing the return address when exiting cpu_switch.
217	 * This will normally be to fork_trampoline(), which will have
218	 * %ebx loaded with the new proc's pointer.  fork_trampoline()
219	 * will set up a stack to call fork_return(p, frame); to complete
220	 * the return to user-mode.
221	 */
222}
223
224/*
225 * Intercept the return address from a freshly forked process that has NOT
226 * been scheduled yet.
227 *
228 * This is needed to make kernel threads stay in kernel mode.
229 */
230void
231cpu_set_fork_handler(p, func, arg)
232	struct proc *p;
233	void (*func) __P((void *));
234	void *arg;
235{
236	/*
237	 * Note that the trap frame follows the args, so the function
238	 * is really called like this:  func(arg, frame);
239	 */
240	p->p_addr->u_pcb.pcb_esi = (int) func;	/* function */
241	p->p_addr->u_pcb.pcb_ebx = (int) arg;	/* first arg */
242}
243
244void
245cpu_exit(p)
246	register struct proc *p;
247{
248	struct pcb *pcb = &p->p_addr->u_pcb;
249
250#ifdef DEV_NPX
251	npxexit(p);
252#endif
253	if (pcb->pcb_ext != 0) {
254	        /*
255		 * XXX do we need to move the TSS off the allocated pages
256		 * before freeing them?  (not done here)
257		 */
258		kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
259		    ctob(IOPAGES + 1));
260		pcb->pcb_ext = 0;
261	}
262	if (pcb->pcb_ldt)
263		user_ldt_free(pcb);
264        if (pcb->pcb_flags & PCB_DBREGS) {
265                /*
266                 * disable all hardware breakpoints
267                 */
268                reset_dbregs();
269                pcb->pcb_flags &= ~PCB_DBREGS;
270        }
271	PROC_LOCK(p);
272	mtx_lock_spin(&sched_lock);
273	while (mtx_owned(&Giant))
274		mtx_unlock_flags(&Giant, MTX_NOSWITCH);
275
276	/*
277	 * We have to wait until after releasing all locks before
278	 * changing p_stat.  If we block on a mutex then we will be
279	 * back at SRUN when we resume and our parent will never
280	 * harvest us.
281	 */
282	p->p_stat = SZOMB;
283
284	wakeup(p->p_pptr);
285	PROC_UNLOCK_NOSWITCH(p);
286
287	cnt.v_swtch++;
288	cpu_throw();
289	panic("cpu_exit");
290}
291
292void
293cpu_wait(p)
294	struct proc *p;
295{
296	GIANT_REQUIRED;
297
298	/* drop per-process resources */
299	pmap_dispose_proc(p);
300
301	/* and clean-out the vmspace */
302	vmspace_free(p->p_vmspace);
303}
304
305/*
306 * Dump the machine specific header information at the start of a core dump.
307 */
308int
309cpu_coredump(p, vp, cred)
310	struct proc *p;
311	struct vnode *vp;
312	struct ucred *cred;
313{
314	int error;
315	caddr_t tempuser;
316
317	tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK | M_ZERO);
318	if (!tempuser)
319		return EINVAL;
320
321	bcopy(p->p_addr, tempuser, sizeof(struct user));
322	bcopy(p->p_frame,
323	      tempuser + ((caddr_t) p->p_frame - (caddr_t) p->p_addr),
324	      sizeof(struct trapframe));
325
326	error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, ctob(UPAGES),
327			(off_t)0, UIO_SYSSPACE, IO_UNIT, cred, (int *)NULL, p);
328
329	free(tempuser, M_TEMP);
330
331	return error;
332}
333
334/*
335 * Convert kernel VA to physical address
336 */
337u_long
338kvtop(void *addr)
339{
340	vm_offset_t va;
341
342	va = pmap_kextract((vm_offset_t)addr);
343	if (va == 0)
344		panic("kvtop: zero page frame");
345	return((int)va);
346}
347
348/*
349 * Map an IO request into kernel virtual address space.
350 *
351 * All requests are (re)mapped into kernel VA space.
352 * Notice that we use b_bufsize for the size of the buffer
353 * to be mapped.  b_bcount might be modified by the driver.
354 */
355void
356vmapbuf(bp)
357	register struct buf *bp;
358{
359	register caddr_t addr, v, kva;
360	vm_offset_t pa;
361
362	GIANT_REQUIRED;
363
364	if ((bp->b_flags & B_PHYS) == 0)
365		panic("vmapbuf");
366
367	for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data);
368	    addr < bp->b_data + bp->b_bufsize;
369	    addr += PAGE_SIZE, v += PAGE_SIZE) {
370		/*
371		 * Do the vm_fault if needed; do the copy-on-write thing
372		 * when reading stuff off device into memory.
373		 */
374		vm_fault_quick(addr,
375			(bp->b_iocmd == BIO_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ);
376		pa = trunc_page(pmap_kextract((vm_offset_t) addr));
377		if (pa == 0)
378			panic("vmapbuf: page not present");
379		vm_page_hold(PHYS_TO_VM_PAGE(pa));
380		pmap_kenter((vm_offset_t) v, pa);
381	}
382
383	kva = bp->b_saveaddr;
384	bp->b_saveaddr = bp->b_data;
385	bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK);
386}
387
388/*
389 * Free the io map PTEs associated with this IO operation.
390 * We also invalidate the TLB entries and restore the original b_addr.
391 */
392void
393vunmapbuf(bp)
394	register struct buf *bp;
395{
396	register caddr_t addr;
397	vm_offset_t pa;
398
399	GIANT_REQUIRED;
400
401	if ((bp->b_flags & B_PHYS) == 0)
402		panic("vunmapbuf");
403
404	for (addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data);
405	    addr < bp->b_data + bp->b_bufsize;
406	    addr += PAGE_SIZE) {
407		pa = trunc_page(pmap_kextract((vm_offset_t) addr));
408		pmap_kremove((vm_offset_t) addr);
409		vm_page_unhold(PHYS_TO_VM_PAGE(pa));
410	}
411
412	bp->b_data = bp->b_saveaddr;
413}
414
415/*
416 * Force reset the processor by invalidating the entire address space!
417 */
418
419#ifdef SMP
420static void
421cpu_reset_proxy()
422{
423
424	cpu_reset_proxy_active = 1;
425	while (cpu_reset_proxy_active == 1)
426		;	 /* Wait for other cpu to see that we've started */
427	stop_cpus((1<<cpu_reset_proxyid));
428	printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
429	DELAY(1000000);
430	cpu_reset_real();
431}
432#endif
433
434void
435cpu_reset()
436{
437#ifdef SMP
438	if (smp_active == 0) {
439		cpu_reset_real();
440		/* NOTREACHED */
441	} else {
442
443		u_int map;
444		int cnt;
445		printf("cpu_reset called on cpu#%d\n", PCPU_GET(cpuid));
446
447		map = PCPU_GET(other_cpus) & ~ stopped_cpus;
448
449		if (map != 0) {
450			printf("cpu_reset: Stopping other CPUs\n");
451			stop_cpus(map);		/* Stop all other CPUs */
452		}
453
454		if (PCPU_GET(cpuid) == 0) {
455			DELAY(1000000);
456			cpu_reset_real();
457			/* NOTREACHED */
458		} else {
459			/* We are not BSP (CPU #0) */
460
461			cpu_reset_proxyid = PCPU_GET(cpuid);
462			cpustop_restartfunc = cpu_reset_proxy;
463			cpu_reset_proxy_active = 0;
464			printf("cpu_reset: Restarting BSP\n");
465			started_cpus = (1<<0);		/* Restart CPU #0 */
466
467			cnt = 0;
468			while (cpu_reset_proxy_active == 0 && cnt < 10000000)
469				cnt++;	/* Wait for BSP to announce restart */
470			if (cpu_reset_proxy_active == 0)
471				printf("cpu_reset: Failed to restart BSP\n");
472			enable_intr();
473			cpu_reset_proxy_active = 2;
474
475			while (1);
476			/* NOTREACHED */
477		}
478	}
479#else
480	cpu_reset_real();
481#endif
482}
483
484static void
485cpu_reset_real()
486{
487
488#ifdef PC98
489	/*
490	 * Attempt to do a CPU reset via CPU reset port.
491	 */
492	disable_intr();
493	if ((inb(0x35) & 0xa0) != 0xa0) {
494		outb(0x37, 0x0f);		/* SHUT0 = 0. */
495		outb(0x37, 0x0b);		/* SHUT1 = 0. */
496	}
497	outb(0xf0, 0x00);		/* Reset. */
498#else
499	/*
500	 * Attempt to do a CPU reset via the keyboard controller,
501	 * do not turn of the GateA20, as any machine that fails
502	 * to do the reset here would then end up in no man's land.
503	 */
504
505#if !defined(BROKEN_KEYBOARD_RESET)
506	outb(IO_KBD + 4, 0xFE);
507	DELAY(500000);	/* wait 0.5 sec to see if that did it */
508	printf("Keyboard reset did not work, attempting CPU shutdown\n");
509	DELAY(1000000);	/* wait 1 sec for printf to complete */
510#endif
511#endif /* PC98 */
512	/* force a shutdown by unmapping entire address space ! */
513	bzero((caddr_t) PTD, PAGE_SIZE);
514
515	/* "good night, sweet prince .... <THUNK!>" */
516	invltlb();
517	/* NOTREACHED */
518	while(1);
519}
520
521int
522grow_stack(p, sp)
523	struct proc *p;
524	u_int sp;
525{
526	int rv;
527
528	rv = vm_map_growstack (p, sp);
529	if (rv != KERN_SUCCESS)
530		return (0);
531
532	return (1);
533}
534
535/*
536 * Software interrupt handler for queued VM system processing.
537 */
538void
539swi_vm(void *dummy)
540{
541	if (busdma_swi_pending != 0)
542		busdma_swi();
543}
544
545/*
546 * Tell whether this address is in some physical memory region.
547 * Currently used by the kernel coredump code in order to avoid
548 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
549 * or other unpredictable behaviour.
550 */
551
552int
553is_physical_memory(addr)
554	vm_offset_t addr;
555{
556
557#ifdef DEV_ISA
558	/* The ISA ``memory hole''. */
559	if (addr >= 0xa0000 && addr < 0x100000)
560		return 0;
561#endif
562
563	/*
564	 * stuff other tests for known memory-mapped devices (PCI?)
565	 * here
566	 */
567
568	return 1;
569}
570