vm_machdep.c revision 79609 
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 79609 2001-07-12 06:32:51Z 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
51#include <sys/param.h>
52#include <sys/systm.h>
53#include <sys/malloc.h>
54#include <sys/proc.h>
55#include <sys/bio.h>
56#include <sys/buf.h>
57#include <sys/vnode.h>
58#include <sys/vmmeter.h>
59#include <sys/kernel.h>
60#include <sys/ktr.h>
61#include <sys/mutex.h>
62#include <sys/smp.h>
63#include <sys/sysctl.h>
64#include <sys/unistd.h>
65
66#include <machine/cpu.h>
67#include <machine/md_var.h>
68#include <machine/pcb.h>
69#include <machine/pcb_ext.h>
70#include <machine/vm86.h>
71
72#include <vm/vm.h>
73#include <vm/vm_param.h>
74#include <sys/lock.h>
75#include <vm/vm_kern.h>
76#include <vm/vm_page.h>
77#include <vm/vm_map.h>
78#include <vm/vm_extern.h>
79
80#include <sys/user.h>
81
82#ifdef PC98
83#include <pc98/pc98/pc98.h>
84#else
85#include <i386/isa/isa.h>
86#endif
87
88static void	cpu_reset_real __P((void));
89#ifdef SMP
90static void	cpu_reset_proxy __P((void));
91static u_int	cpu_reset_proxyid;
92static volatile u_int	cpu_reset_proxy_active;
93#endif
94extern int	_ucodesel, _udatasel;
95
96/*
97 * quick version of vm_fault
98 */
99int
100vm_fault_quick(v, prot)
101	caddr_t v;
102	int prot;
103{
104	int r;
105
106	if (prot & VM_PROT_WRITE)
107		r = subyte(v, fubyte(v));
108	else
109		r = fubyte(v);
110	return(r);
111}
112
113/*
114 * Finish a fork operation, with process p2 nearly set up.
115 * Copy and update the pcb, set up the stack so that the child
116 * ready to run and return to user mode.
117 */
118void
119cpu_fork(p1, p2, flags)
120	register struct proc *p1, *p2;
121	int flags;
122{
123	struct pcb *pcb2;
124#ifdef DEV_NPX
125	int savecrit;
126#endif
127
128	if ((flags & RFPROC) == 0) {
129		if ((flags & RFMEM) == 0) {
130			/* unshare user LDT */
131			struct pcb *pcb1 = &p1->p_addr->u_pcb;
132			struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt;
133			if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) {
134				pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len);
135				if (pcb_ldt == NULL)
136					panic("could not copy LDT");
137				pcb1->pcb_ldt = pcb_ldt;
138				set_user_ldt(pcb1);
139				user_ldt_free(pcb1);
140			}
141		}
142		return;
143	}
144
145	/* Ensure that p1's pcb is up to date. */
146#ifdef DEV_NPX
147	if (p1 == curproc)
148		p1->p_addr->u_pcb.pcb_gs = rgs();
149	savecrit = critical_enter();
150	if (PCPU_GET(npxproc) == p1)
151		npxsave(&p1->p_addr->u_pcb.pcb_save);
152	critical_exit(savecrit);
153#endif
154
155	/* Copy p1's pcb. */
156	p2->p_addr->u_pcb = p1->p_addr->u_pcb;
157	pcb2 = &p2->p_addr->u_pcb;
158
159	/*
160	 * Create a new fresh stack for the new process.
161	 * Copy the trap frame for the return to user mode as if from a
162	 * syscall.  This copies most of the user mode register values.
163	 */
164	p2->p_frame = (struct trapframe *)
165			   ((int)p2->p_addr + UPAGES * PAGE_SIZE - 16) - 1;
166	bcopy(p1->p_frame, p2->p_frame, sizeof(struct trapframe));
167
168	p2->p_frame->tf_eax = 0;		/* Child returns zero */
169	p2->p_frame->tf_eflags &= ~PSL_C;	/* success */
170	p2->p_frame->tf_edx = 1;
171
172	/*
173	 * Set registers for trampoline to user mode.  Leave space for the
174	 * return address on stack.  These are the kernel mode register values.
175	 */
176	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
177	pcb2->pcb_edi = 0;
178	pcb2->pcb_esi = (int)fork_return;	/* fork_trampoline argument */
179	pcb2->pcb_ebp = 0;
180	pcb2->pcb_esp = (int)p2->p_frame - sizeof(void *);
181	pcb2->pcb_ebx = (int)p2;		/* fork_trampoline argument */
182	pcb2->pcb_eip = (int)fork_trampoline;
183	/*-
184	 * pcb2->pcb_dr*:	cloned above.
185	 * pcb2->pcb_ldt:	duplicated below, if necessary.
186	 * pcb2->pcb_savefpu:	cloned above.
187	 * pcb2->pcb_flags:	cloned above.
188	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
189	 * pcb2->pcb_gs:	cloned above.
190	 * pcb2->pcb_ext:	cleared below.
191	 */
192
193	/*
194	 * XXX don't copy the i/o pages.  this should probably be fixed.
195	 */
196	pcb2->pcb_ext = 0;
197
198        /* Copy the LDT, if necessary. */
199	mtx_lock_spin(&sched_lock);
200        if (pcb2->pcb_ldt != 0) {
201		if (flags & RFMEM) {
202			pcb2->pcb_ldt->ldt_refcnt++;
203		} else {
204			pcb2->pcb_ldt = user_ldt_alloc(pcb2,
205				pcb2->pcb_ldt->ldt_len);
206			if (pcb2->pcb_ldt == NULL)
207				panic("could not copy LDT");
208		}
209        }
210	mtx_unlock_spin(&sched_lock);
211
212	/*
213	 * Now, cpu_switch() can schedule the new process.
214	 * pcb_esp is loaded pointing to the cpu_switch() stack frame
215	 * containing the return address when exiting cpu_switch.
216	 * This will normally be to fork_trampoline(), which will have
217	 * %ebx loaded with the new proc's pointer.  fork_trampoline()
218	 * will set up a stack to call fork_return(p, frame); to complete
219	 * the return to user-mode.
220	 */
221}
222
223/*
224 * Intercept the return address from a freshly forked process that has NOT
225 * been scheduled yet.
226 *
227 * This is needed to make kernel threads stay in kernel mode.
228 */
229void
230cpu_set_fork_handler(p, func, arg)
231	struct proc *p;
232	void (*func) __P((void *));
233	void *arg;
234{
235	/*
236	 * Note that the trap frame follows the args, so the function
237	 * is really called like this:  func(arg, frame);
238	 */
239	p->p_addr->u_pcb.pcb_esi = (int) func;	/* function */
240	p->p_addr->u_pcb.pcb_ebx = (int) arg;	/* first arg */
241}
242
243void
244cpu_exit(p)
245	register struct proc *p;
246{
247	struct pcb *pcb = &p->p_addr->u_pcb;
248
249#ifdef DEV_NPX
250	npxexit(p);
251#endif
252	if (pcb->pcb_ext != 0) {
253	        /*
254		 * XXX do we need to move the TSS off the allocated pages
255		 * before freeing them?  (not done here)
256		 */
257		kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
258		    ctob(IOPAGES + 1));
259		pcb->pcb_ext = 0;
260	}
261	if (pcb->pcb_ldt)
262		user_ldt_free(pcb);
263        if (pcb->pcb_flags & PCB_DBREGS) {
264                /*
265                 * disable all hardware breakpoints
266                 */
267                reset_dbregs();
268                pcb->pcb_flags &= ~PCB_DBREGS;
269        }
270	PROC_LOCK(p);
271	mtx_lock_spin(&sched_lock);
272	while (mtx_owned(&Giant))
273		mtx_unlock_flags(&Giant, MTX_NOSWITCH);
274
275	/*
276	 * We have to wait until after releasing all locks before
277	 * changing p_stat.  If we block on a mutex then we will be
278	 * back at SRUN when we resume and our parent will never
279	 * harvest us.
280	 */
281	p->p_stat = SZOMB;
282
283	wakeup(p->p_pptr);
284	PROC_UNLOCK_NOSWITCH(p);
285
286	cnt.v_swtch++;
287	cpu_throw();
288	panic("cpu_exit");
289}
290
291void
292cpu_wait(p)
293	struct proc *p;
294{
295	GIANT_REQUIRED;
296
297	/* drop per-process resources */
298	pmap_dispose_proc(p);
299
300	/* and clean-out the vmspace */
301	vmspace_free(p->p_vmspace);
302}
303
304/*
305 * Dump the machine specific header information at the start of a core dump.
306 */
307int
308cpu_coredump(p, vp, cred)
309	struct proc *p;
310	struct vnode *vp;
311	struct ucred *cred;
312{
313	int error;
314	caddr_t tempuser;
315
316	tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK | M_ZERO);
317	if (!tempuser)
318		return EINVAL;
319
320	bcopy(p->p_addr, tempuser, sizeof(struct user));
321	bcopy(p->p_frame,
322	      tempuser + ((caddr_t) p->p_frame - (caddr_t) p->p_addr),
323	      sizeof(struct trapframe));
324
325	error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser,
326			ctob(UPAGES),
327			(off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT,
328			cred, (int *)NULL, p);
329
330	free(tempuser, M_TEMP);
331
332	return error;
333}
334
335#ifdef notyet
336static void
337setredzone(pte, vaddr)
338	u_short *pte;
339	caddr_t vaddr;
340{
341/* eventually do this by setting up an expand-down stack segment
342   for ss0: selector, allowing stack access down to top of u.
343   this means though that protection violations need to be handled
344   thru a double fault exception that must do an integral task
345   switch to a known good context, within which a dump can be
346   taken. a sensible scheme might be to save the initial context
347   used by sched (that has physical memory mapped 1:1 at bottom)
348   and take the dump while still in mapped mode */
349}
350#endif
351
352/*
353 * Convert kernel VA to physical address
354 */
355u_long
356kvtop(void *addr)
357{
358	vm_offset_t va;
359
360	va = pmap_kextract((vm_offset_t)addr);
361	if (va == 0)
362		panic("kvtop: zero page frame");
363	return((int)va);
364}
365
366/*
367 * Map an IO request into kernel virtual address space.
368 *
369 * All requests are (re)mapped into kernel VA space.
370 * Notice that we use b_bufsize for the size of the buffer
371 * to be mapped.  b_bcount might be modified by the driver.
372 */
373void
374vmapbuf(bp)
375	register struct buf *bp;
376{
377	register caddr_t addr, v, kva;
378	vm_offset_t pa;
379
380	GIANT_REQUIRED;
381
382	if ((bp->b_flags & B_PHYS) == 0)
383		panic("vmapbuf");
384
385	for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data);
386	    addr < bp->b_data + bp->b_bufsize;
387	    addr += PAGE_SIZE, v += PAGE_SIZE) {
388		/*
389		 * Do the vm_fault if needed; do the copy-on-write thing
390		 * when reading stuff off device into memory.
391		 */
392		vm_fault_quick(addr,
393			(bp->b_iocmd == BIO_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ);
394		pa = trunc_page(pmap_kextract((vm_offset_t) addr));
395		if (pa == 0)
396			panic("vmapbuf: page not present");
397		vm_page_hold(PHYS_TO_VM_PAGE(pa));
398		pmap_kenter((vm_offset_t) v, pa);
399	}
400
401	kva = bp->b_saveaddr;
402	bp->b_saveaddr = bp->b_data;
403	bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK);
404}
405
406/*
407 * Free the io map PTEs associated with this IO operation.
408 * We also invalidate the TLB entries and restore the original b_addr.
409 */
410void
411vunmapbuf(bp)
412	register struct buf *bp;
413{
414	register caddr_t addr;
415	vm_offset_t pa;
416
417	GIANT_REQUIRED;
418
419	if ((bp->b_flags & B_PHYS) == 0)
420		panic("vunmapbuf");
421
422	for (addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data);
423	    addr < bp->b_data + bp->b_bufsize;
424	    addr += PAGE_SIZE) {
425		pa = trunc_page(pmap_kextract((vm_offset_t) addr));
426		pmap_kremove((vm_offset_t) addr);
427		vm_page_unhold(PHYS_TO_VM_PAGE(pa));
428	}
429
430	bp->b_data = bp->b_saveaddr;
431}
432
433/*
434 * Force reset the processor by invalidating the entire address space!
435 */
436
437#ifdef SMP
438static void
439cpu_reset_proxy()
440{
441
442	cpu_reset_proxy_active = 1;
443	while (cpu_reset_proxy_active == 1)
444		;	 /* Wait for other cpu to see that we've started */
445	stop_cpus((1<<cpu_reset_proxyid));
446	printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
447	DELAY(1000000);
448	cpu_reset_real();
449}
450#endif
451
452void
453cpu_reset()
454{
455#ifdef SMP
456	if (smp_active == 0) {
457		cpu_reset_real();
458		/* NOTREACHED */
459	} else {
460
461		u_int map;
462		int cnt;
463		printf("cpu_reset called on cpu#%d\n", PCPU_GET(cpuid));
464
465		map = PCPU_GET(other_cpus) & ~ stopped_cpus;
466
467		if (map != 0) {
468			printf("cpu_reset: Stopping other CPUs\n");
469			stop_cpus(map);		/* Stop all other CPUs */
470		}
471
472		if (PCPU_GET(cpuid) == 0) {
473			DELAY(1000000);
474			cpu_reset_real();
475			/* NOTREACHED */
476		} else {
477			/* We are not BSP (CPU #0) */
478
479			cpu_reset_proxyid = PCPU_GET(cpuid);
480			cpustop_restartfunc = cpu_reset_proxy;
481			cpu_reset_proxy_active = 0;
482			printf("cpu_reset: Restarting BSP\n");
483			started_cpus = (1<<0);		/* Restart CPU #0 */
484
485			cnt = 0;
486			while (cpu_reset_proxy_active == 0 && cnt < 10000000)
487				cnt++;	/* Wait for BSP to announce restart */
488			if (cpu_reset_proxy_active == 0)
489				printf("cpu_reset: Failed to restart BSP\n");
490			enable_intr();
491			cpu_reset_proxy_active = 2;
492
493			while (1);
494			/* NOTREACHED */
495		}
496	}
497#else
498	cpu_reset_real();
499#endif
500}
501
502static void
503cpu_reset_real()
504{
505
506#ifdef PC98
507	/*
508	 * Attempt to do a CPU reset via CPU reset port.
509	 */
510	disable_intr();
511	if ((inb(0x35) & 0xa0) != 0xa0) {
512		outb(0x37, 0x0f);		/* SHUT0 = 0. */
513		outb(0x37, 0x0b);		/* SHUT1 = 0. */
514	}
515	outb(0xf0, 0x00);		/* Reset. */
516#else
517	/*
518	 * Attempt to do a CPU reset via the keyboard controller,
519	 * do not turn of the GateA20, as any machine that fails
520	 * to do the reset here would then end up in no man's land.
521	 */
522
523#if !defined(BROKEN_KEYBOARD_RESET)
524	outb(IO_KBD + 4, 0xFE);
525	DELAY(500000);	/* wait 0.5 sec to see if that did it */
526	printf("Keyboard reset did not work, attempting CPU shutdown\n");
527	DELAY(1000000);	/* wait 1 sec for printf to complete */
528#endif
529#endif /* PC98 */
530	/* force a shutdown by unmapping entire address space ! */
531	bzero((caddr_t) PTD, PAGE_SIZE);
532
533	/* "good night, sweet prince .... <THUNK!>" */
534	invltlb();
535	/* NOTREACHED */
536	while(1);
537}
538
539int
540grow_stack(p, sp)
541	struct proc *p;
542	u_int sp;
543{
544	int rv;
545
546	rv = vm_map_growstack (p, sp);
547	if (rv != KERN_SUCCESS)
548		return (0);
549
550	return (1);
551}
552
553/*
554 * Software interrupt handler for queued VM system processing.
555 */
556void
557swi_vm(void *dummy)
558{
559	if (busdma_swi_pending != 0)
560		busdma_swi();
561}
562
563/*
564 * Tell whether this address is in some physical memory region.
565 * Currently used by the kernel coredump code in order to avoid
566 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
567 * or other unpredictable behaviour.
568 */
569
570int
571is_physical_memory(addr)
572	vm_offset_t addr;
573{
574
575#ifdef DEV_ISA
576	/* The ISA ``memory hole''. */
577	if (addr >= 0xa0000 && addr < 0x100000)
578		return 0;
579#endif
580
581	/*
582	 * stuff other tests for known memory-mapped devices (PCI?)
583	 * here
584	 */
585
586	return 1;
587}
588