vm_machdep.c revision 83276 
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 83276 2001-09-10 04:28:58Z 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}
272
273void
274cpu_wait(p)
275	struct proc *p;
276{
277}
278
279/*
280 * Dump the machine specific header information at the start of a core dump.
281 */
282int
283cpu_coredump(p, vp, cred)
284	struct proc *p;
285	struct vnode *vp;
286	struct ucred *cred;
287{
288	int error;
289	caddr_t tempuser;
290
291	tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK | M_ZERO);
292	if (!tempuser)
293		return EINVAL;
294
295	bcopy(p->p_addr, tempuser, sizeof(struct user));
296	bcopy(p->p_frame,
297	      tempuser + ((caddr_t) p->p_frame - (caddr_t) p->p_addr),
298	      sizeof(struct trapframe));
299
300	error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, ctob(UPAGES),
301			(off_t)0, UIO_SYSSPACE, IO_UNIT, cred, (int *)NULL, p);
302
303	free(tempuser, M_TEMP);
304
305	return error;
306}
307
308/*
309 * Convert kernel VA to physical address
310 */
311u_long
312kvtop(void *addr)
313{
314	vm_offset_t va;
315
316	va = pmap_kextract((vm_offset_t)addr);
317	if (va == 0)
318		panic("kvtop: zero page frame");
319	return((int)va);
320}
321
322/*
323 * Map an IO request into kernel virtual address space.
324 *
325 * All requests are (re)mapped into kernel VA space.
326 * Notice that we use b_bufsize for the size of the buffer
327 * to be mapped.  b_bcount might be modified by the driver.
328 */
329void
330vmapbuf(bp)
331	register struct buf *bp;
332{
333	register caddr_t addr, v, kva;
334	vm_offset_t pa;
335
336	GIANT_REQUIRED;
337
338	if ((bp->b_flags & B_PHYS) == 0)
339		panic("vmapbuf");
340
341	for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data);
342	    addr < bp->b_data + bp->b_bufsize;
343	    addr += PAGE_SIZE, v += PAGE_SIZE) {
344		/*
345		 * Do the vm_fault if needed; do the copy-on-write thing
346		 * when reading stuff off device into memory.
347		 */
348		vm_fault_quick(addr,
349			(bp->b_iocmd == BIO_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ);
350		pa = trunc_page(pmap_kextract((vm_offset_t) addr));
351		if (pa == 0)
352			panic("vmapbuf: page not present");
353		vm_page_hold(PHYS_TO_VM_PAGE(pa));
354		pmap_kenter((vm_offset_t) v, pa);
355	}
356
357	kva = bp->b_saveaddr;
358	bp->b_saveaddr = bp->b_data;
359	bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK);
360}
361
362/*
363 * Free the io map PTEs associated with this IO operation.
364 * We also invalidate the TLB entries and restore the original b_addr.
365 */
366void
367vunmapbuf(bp)
368	register struct buf *bp;
369{
370	register caddr_t addr;
371	vm_offset_t pa;
372
373	GIANT_REQUIRED;
374
375	if ((bp->b_flags & B_PHYS) == 0)
376		panic("vunmapbuf");
377
378	for (addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data);
379	    addr < bp->b_data + bp->b_bufsize;
380	    addr += PAGE_SIZE) {
381		pa = trunc_page(pmap_kextract((vm_offset_t) addr));
382		pmap_kremove((vm_offset_t) addr);
383		vm_page_unhold(PHYS_TO_VM_PAGE(pa));
384	}
385
386	bp->b_data = bp->b_saveaddr;
387}
388
389/*
390 * Force reset the processor by invalidating the entire address space!
391 */
392
393#ifdef SMP
394static void
395cpu_reset_proxy()
396{
397
398	cpu_reset_proxy_active = 1;
399	while (cpu_reset_proxy_active == 1)
400		;	 /* Wait for other cpu to see that we've started */
401	stop_cpus((1<<cpu_reset_proxyid));
402	printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
403	DELAY(1000000);
404	cpu_reset_real();
405}
406#endif
407
408void
409cpu_reset()
410{
411#ifdef SMP
412	if (smp_active == 0) {
413		cpu_reset_real();
414		/* NOTREACHED */
415	} else {
416
417		u_int map;
418		int cnt;
419		printf("cpu_reset called on cpu#%d\n", PCPU_GET(cpuid));
420
421		map = PCPU_GET(other_cpus) & ~ stopped_cpus;
422
423		if (map != 0) {
424			printf("cpu_reset: Stopping other CPUs\n");
425			stop_cpus(map);		/* Stop all other CPUs */
426		}
427
428		if (PCPU_GET(cpuid) == 0) {
429			DELAY(1000000);
430			cpu_reset_real();
431			/* NOTREACHED */
432		} else {
433			/* We are not BSP (CPU #0) */
434
435			cpu_reset_proxyid = PCPU_GET(cpuid);
436			cpustop_restartfunc = cpu_reset_proxy;
437			cpu_reset_proxy_active = 0;
438			printf("cpu_reset: Restarting BSP\n");
439			started_cpus = (1<<0);		/* Restart CPU #0 */
440
441			cnt = 0;
442			while (cpu_reset_proxy_active == 0 && cnt < 10000000)
443				cnt++;	/* Wait for BSP to announce restart */
444			if (cpu_reset_proxy_active == 0)
445				printf("cpu_reset: Failed to restart BSP\n");
446			enable_intr();
447			cpu_reset_proxy_active = 2;
448
449			while (1);
450			/* NOTREACHED */
451		}
452	}
453#else
454	cpu_reset_real();
455#endif
456}
457
458static void
459cpu_reset_real()
460{
461
462#ifdef PC98
463	/*
464	 * Attempt to do a CPU reset via CPU reset port.
465	 */
466	disable_intr();
467	if ((inb(0x35) & 0xa0) != 0xa0) {
468		outb(0x37, 0x0f);		/* SHUT0 = 0. */
469		outb(0x37, 0x0b);		/* SHUT1 = 0. */
470	}
471	outb(0xf0, 0x00);		/* Reset. */
472#else
473	/*
474	 * Attempt to do a CPU reset via the keyboard controller,
475	 * do not turn of the GateA20, as any machine that fails
476	 * to do the reset here would then end up in no man's land.
477	 */
478
479#if !defined(BROKEN_KEYBOARD_RESET)
480	outb(IO_KBD + 4, 0xFE);
481	DELAY(500000);	/* wait 0.5 sec to see if that did it */
482	printf("Keyboard reset did not work, attempting CPU shutdown\n");
483	DELAY(1000000);	/* wait 1 sec for printf to complete */
484#endif
485#endif /* PC98 */
486	/* force a shutdown by unmapping entire address space ! */
487	bzero((caddr_t) PTD, PAGE_SIZE);
488
489	/* "good night, sweet prince .... <THUNK!>" */
490	invltlb();
491	/* NOTREACHED */
492	while(1);
493}
494
495int
496grow_stack(p, sp)
497	struct proc *p;
498	u_int sp;
499{
500	int rv;
501
502	rv = vm_map_growstack (p, sp);
503	if (rv != KERN_SUCCESS)
504		return (0);
505
506	return (1);
507}
508
509/*
510 * Software interrupt handler for queued VM system processing.
511 */
512void
513swi_vm(void *dummy)
514{
515	if (busdma_swi_pending != 0)
516		busdma_swi();
517}
518
519/*
520 * Tell whether this address is in some physical memory region.
521 * Currently used by the kernel coredump code in order to avoid
522 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
523 * or other unpredictable behaviour.
524 */
525
526int
527is_physical_memory(addr)
528	vm_offset_t addr;
529{
530
531#ifdef DEV_ISA
532	/* The ISA ``memory hole''. */
533	if (addr >= 0xa0000 && addr < 0x100000)
534		return 0;
535#endif
536
537	/*
538	 * stuff other tests for known memory-mapped devices (PCI?)
539	 * here
540	 */
541
542	return 1;
543}
544