vm_machdep.c revision 90562 
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 90562 2002-02-12 05:50:43Z alc $
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/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(td1, p2, td2, flags)
121	register struct thread *td1;
122	register struct proc *p2;
123	struct thread *td2;
124	int flags;
125{
126	register struct proc *p1;
127	struct pcb *pcb2;
128	struct mdproc *mdp2;
129#ifdef DEV_NPX
130	int savecrit;
131#endif
132
133	p1 = td1->td_proc;
134	if ((flags & RFPROC) == 0) {
135		if ((flags & RFMEM) == 0) {
136			/* unshare user LDT */
137			struct mdproc *mdp1 = &p1->p_md;
138			struct proc_ldt *pldt = mdp1->md_ldt;
139			if (pldt && pldt->ldt_refcnt > 1) {
140				pldt = user_ldt_alloc(mdp1, pldt->ldt_len);
141				if (pldt == NULL)
142					panic("could not copy LDT");
143				mdp1->md_ldt = pldt;
144				set_user_ldt(mdp1);
145				user_ldt_free(td1);
146			}
147		}
148		return;
149	}
150
151	/* Ensure that p1's pcb is up to date. */
152#ifdef DEV_NPX
153	if (td1 == curthread)
154		td1->td_pcb->pcb_gs = rgs();
155	savecrit = cpu_critical_enter();
156	if (PCPU_GET(fpcurthread) == td1)
157		npxsave(&td1->td_pcb->pcb_save);
158	cpu_critical_exit(savecrit);
159#endif
160
161	/* Point the pcb to the top of the stack */
162	pcb2 = (struct pcb *)(td2->td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;
163	td2->td_pcb = pcb2;
164
165	/* Copy p1's pcb */
166	bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
167
168	/* Point mdproc and then copy over td1's contents */
169	mdp2 = &p2->p_md;
170	bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
171
172	/*
173	 * Create a new fresh stack for the new process.
174	 * Copy the trap frame for the return to user mode as if from a
175	 * syscall.  This copies most of the user mode register values.
176	 * The -16 is so we can expand the trapframe if we go to vm86.
177	 */
178	td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1;
179	bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
180
181	td2->td_frame->tf_eax = 0;		/* Child returns zero */
182	td2->td_frame->tf_eflags &= ~PSL_C;	/* success */
183	td2->td_frame->tf_edx = 1;
184
185	/*
186	 * Set registers for trampoline to user mode.  Leave space for the
187	 * return address on stack.  These are the kernel mode register values.
188	 */
189	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
190	pcb2->pcb_edi = 0;
191	pcb2->pcb_esi = (int)fork_return;	/* fork_trampoline argument */
192	pcb2->pcb_ebp = 0;
193	pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
194	pcb2->pcb_ebx = (int)td2;		/* fork_trampoline argument */
195	pcb2->pcb_eip = (int)fork_trampoline;
196	/*-
197	 * pcb2->pcb_dr*:	cloned above.
198	 * pcb2->pcb_savefpu:	cloned above.
199	 * pcb2->pcb_flags:	cloned above.
200	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
201	 * pcb2->pcb_gs:	cloned above.
202	 * pcb2->pcb_ext:	cleared below.
203	 */
204
205	/*
206	 * XXX don't copy the i/o pages.  this should probably be fixed.
207	 */
208	pcb2->pcb_ext = 0;
209
210        /* Copy the LDT, if necessary. */
211	mtx_lock_spin(&sched_lock);
212        if (mdp2->md_ldt != 0) {
213		if (flags & RFMEM) {
214			mdp2->md_ldt->ldt_refcnt++;
215		} else {
216			mdp2->md_ldt = user_ldt_alloc(mdp2,
217			    mdp2->md_ldt->ldt_len);
218			if (mdp2->md_ldt == NULL)
219				panic("could not copy LDT");
220		}
221        }
222	mtx_unlock_spin(&sched_lock);
223
224	/*
225	 * Now, cpu_switch() can schedule the new process.
226	 * pcb_esp is loaded pointing to the cpu_switch() stack frame
227	 * containing the return address when exiting cpu_switch.
228	 * This will normally be to fork_trampoline(), which will have
229	 * %ebx loaded with the new proc's pointer.  fork_trampoline()
230	 * will set up a stack to call fork_return(p, frame); to complete
231	 * the return to user-mode.
232	 */
233}
234
235/*
236 * Intercept the return address from a freshly forked process that has NOT
237 * been scheduled yet.
238 *
239 * This is needed to make kernel threads stay in kernel mode.
240 */
241void
242cpu_set_fork_handler(td, func, arg)
243	struct thread *td;
244	void (*func) __P((void *));
245	void *arg;
246{
247	/*
248	 * Note that the trap frame follows the args, so the function
249	 * is really called like this:  func(arg, frame);
250	 */
251	td->td_pcb->pcb_esi = (int) func;	/* function */
252	td->td_pcb->pcb_ebx = (int) arg;	/* first arg */
253}
254
255void
256cpu_exit(td)
257	register struct thread *td;
258{
259	struct pcb *pcb = td->td_pcb;
260	struct mdproc *mdp = &td->td_proc->p_md;
261#ifdef DEV_NPX
262	npxexit(td);
263#endif
264	if (pcb->pcb_ext != 0) {
265	        /*
266		 * XXX do we need to move the TSS off the allocated pages
267		 * before freeing them?  (not done here)
268		 */
269		kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
270		    ctob(IOPAGES + 1));
271		pcb->pcb_ext = 0;
272	}
273	if (mdp->md_ldt)
274		user_ldt_free(td);
275        if (pcb->pcb_flags & PCB_DBREGS) {
276                /*
277                 * disable all hardware breakpoints
278                 */
279                reset_dbregs();
280                pcb->pcb_flags &= ~PCB_DBREGS;
281        }
282}
283
284void
285cpu_wait(p)
286	struct proc *p;
287{
288}
289
290/*
291 * Dump the machine specific header information at the start of a core dump.
292 */
293int
294cpu_coredump(td, vp, cred)
295	struct thread *td;
296	struct vnode *vp;
297	struct ucred *cred;
298{
299	struct proc *p = td->td_proc;
300	int error;
301	caddr_t tempuser;
302
303	tempuser = malloc(ctob(UAREA_PAGES + KSTACK_PAGES), M_TEMP, M_WAITOK | M_ZERO);
304	if (!tempuser)
305		return EINVAL;
306
307	bcopy(p->p_uarea, tempuser, sizeof(struct user));
308#if 0		/* XXXKSE - broken, fixme!!!!! td_frame is in kstack! */
309	bcopy(td->td_frame,
310	      tempuser + ((caddr_t) td->td_frame - (caddr_t) p->p_uarea),
311	      sizeof(struct trapframe));
312#endif
313
314	error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser,
315			ctob(UAREA_PAGES + KSTACK_PAGES),
316			(off_t)0, UIO_SYSSPACE, IO_UNIT, cred, (int *)NULL, td);
317
318	free(tempuser, M_TEMP);
319
320	return error;
321}
322
323/*
324 * Convert kernel VA to physical address
325 */
326u_long
327kvtop(void *addr)
328{
329	vm_offset_t va;
330
331	va = pmap_kextract((vm_offset_t)addr);
332	if (va == 0)
333		panic("kvtop: zero page frame");
334	return((int)va);
335}
336
337/*
338 * Map an IO request into kernel virtual address space.
339 *
340 * All requests are (re)mapped into kernel VA space.
341 * Notice that we use b_bufsize for the size of the buffer
342 * to be mapped.  b_bcount might be modified by the driver.
343 */
344void
345vmapbuf(bp)
346	register struct buf *bp;
347{
348	register caddr_t addr, kva;
349	vm_offset_t pa;
350	int pidx;
351	struct vm_page *m;
352
353	GIANT_REQUIRED;
354
355	if ((bp->b_flags & B_PHYS) == 0)
356		panic("vmapbuf");
357
358	for (addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data), pidx = 0;
359	     addr < bp->b_data + bp->b_bufsize;
360	     addr += PAGE_SIZE, pidx++) {
361		/*
362		 * Do the vm_fault if needed; do the copy-on-write thing
363		 * when reading stuff off device into memory.
364		 */
365		vm_fault_quick((addr >= bp->b_data) ? addr : bp->b_data,
366			(bp->b_iocmd == BIO_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ);
367		pa = trunc_page(pmap_kextract((vm_offset_t) addr));
368		if (pa == 0)
369			panic("vmapbuf: page not present");
370		m = PHYS_TO_VM_PAGE(pa);
371		vm_page_hold(m);
372		bp->b_pages[pidx] = m;
373	}
374	if (pidx > btoc(MAXPHYS))
375		panic("vmapbuf: mapped more than MAXPHYS");
376	pmap_qenter((vm_offset_t)bp->b_saveaddr, bp->b_pages, pidx);
377
378	kva = bp->b_saveaddr;
379	bp->b_npages = pidx;
380	bp->b_saveaddr = bp->b_data;
381	bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK);
382}
383
384/*
385 * Free the io map PTEs associated with this IO operation.
386 * We also invalidate the TLB entries and restore the original b_addr.
387 */
388void
389vunmapbuf(bp)
390	register struct buf *bp;
391{
392	int pidx;
393	int npages;
394	vm_page_t *m;
395
396	GIANT_REQUIRED;
397
398	if ((bp->b_flags & B_PHYS) == 0)
399		panic("vunmapbuf");
400
401	npages = bp->b_npages;
402	pmap_qremove(trunc_page((vm_offset_t)bp->b_data),
403		     npages);
404	m = bp->b_pages;
405	for (pidx = 0; pidx < npages; pidx++)
406		vm_page_unhold(*m++);
407
408	bp->b_data = bp->b_saveaddr;
409}
410
411/*
412 * Force reset the processor by invalidating the entire address space!
413 */
414
415#ifdef SMP
416static void
417cpu_reset_proxy()
418{
419
420	cpu_reset_proxy_active = 1;
421	while (cpu_reset_proxy_active == 1)
422		;	 /* Wait for other cpu to see that we've started */
423	stop_cpus((1<<cpu_reset_proxyid));
424	printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
425	DELAY(1000000);
426	cpu_reset_real();
427}
428#endif
429
430void
431cpu_reset()
432{
433#ifdef SMP
434	if (smp_active == 0) {
435		cpu_reset_real();
436		/* NOTREACHED */
437	} else {
438
439		u_int map;
440		int cnt;
441		printf("cpu_reset called on cpu#%d\n", PCPU_GET(cpuid));
442
443		map = PCPU_GET(other_cpus) & ~ stopped_cpus;
444
445		if (map != 0) {
446			printf("cpu_reset: Stopping other CPUs\n");
447			stop_cpus(map);		/* Stop all other CPUs */
448		}
449
450		if (PCPU_GET(cpuid) == 0) {
451			DELAY(1000000);
452			cpu_reset_real();
453			/* NOTREACHED */
454		} else {
455			/* We are not BSP (CPU #0) */
456
457			cpu_reset_proxyid = PCPU_GET(cpuid);
458			cpustop_restartfunc = cpu_reset_proxy;
459			cpu_reset_proxy_active = 0;
460			printf("cpu_reset: Restarting BSP\n");
461			started_cpus = (1<<0);		/* Restart CPU #0 */
462
463			cnt = 0;
464			while (cpu_reset_proxy_active == 0 && cnt < 10000000)
465				cnt++;	/* Wait for BSP to announce restart */
466			if (cpu_reset_proxy_active == 0)
467				printf("cpu_reset: Failed to restart BSP\n");
468			enable_intr();
469			cpu_reset_proxy_active = 2;
470
471			while (1);
472			/* NOTREACHED */
473		}
474	}
475#else
476	cpu_reset_real();
477#endif
478}
479
480static void
481cpu_reset_real()
482{
483
484#ifdef PC98
485	/*
486	 * Attempt to do a CPU reset via CPU reset port.
487	 */
488	disable_intr();
489	if ((inb(0x35) & 0xa0) != 0xa0) {
490		outb(0x37, 0x0f);		/* SHUT0 = 0. */
491		outb(0x37, 0x0b);		/* SHUT1 = 0. */
492	}
493	outb(0xf0, 0x00);		/* Reset. */
494#else
495	/*
496	 * Attempt to do a CPU reset via the keyboard controller,
497	 * do not turn of the GateA20, as any machine that fails
498	 * to do the reset here would then end up in no man's land.
499	 */
500
501#if !defined(BROKEN_KEYBOARD_RESET)
502	outb(IO_KBD + 4, 0xFE);
503	DELAY(500000);	/* wait 0.5 sec to see if that did it */
504	printf("Keyboard reset did not work, attempting CPU shutdown\n");
505	DELAY(1000000);	/* wait 1 sec for printf to complete */
506#endif
507#endif /* PC98 */
508	/* force a shutdown by unmapping entire address space ! */
509	bzero((caddr_t) PTD, PAGE_SIZE);
510
511	/* "good night, sweet prince .... <THUNK!>" */
512	invltlb();
513	/* NOTREACHED */
514	while(1);
515}
516
517int
518grow_stack(p, sp)
519	struct proc *p;
520	u_int sp;
521{
522	int rv;
523
524	rv = vm_map_growstack (p, sp);
525	if (rv != KERN_SUCCESS)
526		return (0);
527
528	return (1);
529}
530
531/*
532 * Software interrupt handler for queued VM system processing.
533 */
534void
535swi_vm(void *dummy)
536{
537	if (busdma_swi_pending != 0)
538		busdma_swi();
539}
540
541/*
542 * Tell whether this address is in some physical memory region.
543 * Currently used by the kernel coredump code in order to avoid
544 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
545 * or other unpredictable behaviour.
546 */
547
548int
549is_physical_memory(addr)
550	vm_offset_t addr;
551{
552
553#ifdef DEV_ISA
554	/* The ISA ``memory hole''. */
555	if (addr >= 0xa0000 && addr < 0x100000)
556		return 0;
557#endif
558
559	/*
560	 * stuff other tests for known memory-mapped devices (PCI?)
561	 * here
562	 */
563
564	return 1;
565}
566