vm_machdep.c revision 84935 
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 84935 2001-10-14 21:09:04Z tegge $
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, flags)
121	register struct thread *td1;
122	register struct proc *p2;
123	int flags;
124{
125	register struct proc *p1;
126	struct thread *td2;
127	struct pcb *pcb2;
128#ifdef DEV_NPX
129	int savecrit;
130#endif
131
132	p1 = td1->td_proc;
133	td2 = &p2->p_thread;
134	if ((flags & RFPROC) == 0) {
135		if ((flags & RFMEM) == 0) {
136			/* unshare user LDT */
137			struct pcb *pcb1 = td1->td_pcb;
138			struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt;
139			if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) {
140				pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len);
141				if (pcb_ldt == NULL)
142					panic("could not copy LDT");
143				pcb1->pcb_ldt = pcb_ldt;
144				set_user_ldt(pcb1);
145				user_ldt_free(pcb1);
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 = critical_enter();
156	if (PCPU_GET(npxthread) == td1)
157		npxsave(&td1->td_pcb->pcb_save);
158	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	/*
169	 * Create a new fresh stack for the new process.
170	 * Copy the trap frame for the return to user mode as if from a
171	 * syscall.  This copies most of the user mode register values.
172	 * The -16 is so we can expand the trapframe if we go to vm86.
173	 */
174	td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1;
175	bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
176
177	td2->td_frame->tf_eax = 0;		/* Child returns zero */
178	td2->td_frame->tf_eflags &= ~PSL_C;	/* success */
179	td2->td_frame->tf_edx = 1;
180
181	/*
182	 * Set registers for trampoline to user mode.  Leave space for the
183	 * return address on stack.  These are the kernel mode register values.
184	 */
185	pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
186	pcb2->pcb_edi = 0;
187	pcb2->pcb_esi = (int)fork_return;	/* fork_trampoline argument */
188	pcb2->pcb_ebp = 0;
189	pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
190	pcb2->pcb_ebx = (int)td2;		/* fork_trampoline argument */
191	pcb2->pcb_eip = (int)fork_trampoline;
192	/*-
193	 * pcb2->pcb_dr*:	cloned above.
194	 * pcb2->pcb_ldt:	duplicated below, if necessary.
195	 * pcb2->pcb_savefpu:	cloned above.
196	 * pcb2->pcb_flags:	cloned above.
197	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
198	 * pcb2->pcb_gs:	cloned above.
199	 * pcb2->pcb_ext:	cleared below.
200	 */
201
202	/*
203	 * XXX don't copy the i/o pages.  this should probably be fixed.
204	 */
205	pcb2->pcb_ext = 0;
206
207        /* Copy the LDT, if necessary. */
208	mtx_lock_spin(&sched_lock);
209        if (pcb2->pcb_ldt != 0) {
210		if (flags & RFMEM) {
211			pcb2->pcb_ldt->ldt_refcnt++;
212		} else {
213			pcb2->pcb_ldt = user_ldt_alloc(pcb2,
214				pcb2->pcb_ldt->ldt_len);
215			if (pcb2->pcb_ldt == NULL)
216				panic("could not copy LDT");
217		}
218        }
219	mtx_unlock_spin(&sched_lock);
220
221	/*
222	 * Now, cpu_switch() can schedule the new process.
223	 * pcb_esp is loaded pointing to the cpu_switch() stack frame
224	 * containing the return address when exiting cpu_switch.
225	 * This will normally be to fork_trampoline(), which will have
226	 * %ebx loaded with the new proc's pointer.  fork_trampoline()
227	 * will set up a stack to call fork_return(p, frame); to complete
228	 * the return to user-mode.
229	 */
230}
231
232/*
233 * Intercept the return address from a freshly forked process that has NOT
234 * been scheduled yet.
235 *
236 * This is needed to make kernel threads stay in kernel mode.
237 */
238void
239cpu_set_fork_handler(td, func, arg)
240	struct thread *td;
241	void (*func) __P((void *));
242	void *arg;
243{
244	/*
245	 * Note that the trap frame follows the args, so the function
246	 * is really called like this:  func(arg, frame);
247	 */
248	td->td_pcb->pcb_esi = (int) func;	/* function */
249	td->td_pcb->pcb_ebx = (int) arg;	/* first arg */
250}
251
252void
253cpu_exit(td)
254	register struct thread *td;
255{
256	struct pcb *pcb = td->td_pcb;
257
258#ifdef DEV_NPX
259	npxexit(td);
260#endif
261	if (pcb->pcb_ext != 0) {
262	        /*
263		 * XXX do we need to move the TSS off the allocated pages
264		 * before freeing them?  (not done here)
265		 */
266		kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
267		    ctob(IOPAGES + 1));
268		pcb->pcb_ext = 0;
269	}
270	if (pcb->pcb_ldt)
271		user_ldt_free(pcb);
272        if (pcb->pcb_flags & PCB_DBREGS) {
273                /*
274                 * disable all hardware breakpoints
275                 */
276                reset_dbregs();
277                pcb->pcb_flags &= ~PCB_DBREGS;
278        }
279}
280
281void
282cpu_wait(p)
283	struct proc *p;
284{
285}
286
287/*
288 * Dump the machine specific header information at the start of a core dump.
289 */
290int
291cpu_coredump(td, vp, cred)
292	struct thread *td;
293	struct vnode *vp;
294	struct ucred *cred;
295{
296	struct proc *p = td->td_proc;
297	int error;
298	caddr_t tempuser;
299
300	tempuser = malloc(ctob(UAREA_PAGES + KSTACK_PAGES), M_TEMP, M_WAITOK | M_ZERO);
301	if (!tempuser)
302		return EINVAL;
303
304	bcopy(p->p_uarea, tempuser, sizeof(struct user));
305#if 0		/* XXXKSE - broken, fixme!!!!! td_frame is in kstack! */
306	bcopy(td->td_frame,
307	      tempuser + ((caddr_t) td->td_frame - (caddr_t) p->p_uarea),
308	      sizeof(struct trapframe));
309#endif
310
311	error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser,
312			ctob(UAREA_PAGES + KSTACK_PAGES),
313			(off_t)0, UIO_SYSSPACE, IO_UNIT, cred, (int *)NULL, td);
314
315	free(tempuser, M_TEMP);
316
317	return error;
318}
319
320/*
321 * Convert kernel VA to physical address
322 */
323u_long
324kvtop(void *addr)
325{
326	vm_offset_t va;
327
328	va = pmap_kextract((vm_offset_t)addr);
329	if (va == 0)
330		panic("kvtop: zero page frame");
331	return((int)va);
332}
333
334/*
335 * Map an IO request into kernel virtual address space.
336 *
337 * All requests are (re)mapped into kernel VA space.
338 * Notice that we use b_bufsize for the size of the buffer
339 * to be mapped.  b_bcount might be modified by the driver.
340 */
341void
342vmapbuf(bp)
343	register struct buf *bp;
344{
345	register caddr_t addr, v, kva;
346	vm_offset_t pa;
347	int pidx;
348	struct vm_page *m;
349
350	GIANT_REQUIRED;
351
352	if ((bp->b_flags & B_PHYS) == 0)
353		panic("vmapbuf");
354
355	for (v = bp->b_saveaddr,
356		     addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data),
357		     pidx = 0;
358	     addr < bp->b_data + bp->b_bufsize;
359	     addr += PAGE_SIZE, v += PAGE_SIZE, pidx++) {
360		/*
361		 * Do the vm_fault if needed; do the copy-on-write thing
362		 * when reading stuff off device into memory.
363		 */
364		vm_fault_quick((addr >= bp->b_data) ? addr : bp->b_data,
365			(bp->b_iocmd == BIO_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ);
366		pa = trunc_page(pmap_kextract((vm_offset_t) addr));
367		if (pa == 0)
368			panic("vmapbuf: page not present");
369		m = PHYS_TO_VM_PAGE(pa);
370		vm_page_hold(m);
371		bp->b_pages[pidx] = m;
372	}
373	if (pidx > btoc(MAXPHYS))
374		panic("vmapbuf: mapped more than MAXPHYS");
375	pmap_qenter((vm_offset_t)bp->b_saveaddr, bp->b_pages, pidx);
376
377	kva = bp->b_saveaddr;
378	bp->b_npages = pidx;
379	bp->b_saveaddr = bp->b_data;
380	bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK);
381}
382
383/*
384 * Free the io map PTEs associated with this IO operation.
385 * We also invalidate the TLB entries and restore the original b_addr.
386 */
387void
388vunmapbuf(bp)
389	register struct buf *bp;
390{
391	int pidx;
392	int npages;
393	vm_page_t *m;
394
395	GIANT_REQUIRED;
396
397	if ((bp->b_flags & B_PHYS) == 0)
398		panic("vunmapbuf");
399
400	npages = bp->b_npages;
401	pmap_qremove(trunc_page((vm_offset_t)bp->b_data),
402		     npages);
403	m = bp->b_pages;
404	for (pidx = 0; pidx < npages; pidx++)
405		vm_page_unhold(*m++);
406
407	bp->b_data = bp->b_saveaddr;
408}
409
410/*
411 * Force reset the processor by invalidating the entire address space!
412 */
413
414#ifdef SMP
415static void
416cpu_reset_proxy()
417{
418
419	cpu_reset_proxy_active = 1;
420	while (cpu_reset_proxy_active == 1)
421		;	 /* Wait for other cpu to see that we've started */
422	stop_cpus((1<<cpu_reset_proxyid));
423	printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
424	DELAY(1000000);
425	cpu_reset_real();
426}
427#endif
428
429void
430cpu_reset()
431{
432#ifdef SMP
433	if (smp_active == 0) {
434		cpu_reset_real();
435		/* NOTREACHED */
436	} else {
437
438		u_int map;
439		int cnt;
440		printf("cpu_reset called on cpu#%d\n", PCPU_GET(cpuid));
441
442		map = PCPU_GET(other_cpus) & ~ stopped_cpus;
443
444		if (map != 0) {
445			printf("cpu_reset: Stopping other CPUs\n");
446			stop_cpus(map);		/* Stop all other CPUs */
447		}
448
449		if (PCPU_GET(cpuid) == 0) {
450			DELAY(1000000);
451			cpu_reset_real();
452			/* NOTREACHED */
453		} else {
454			/* We are not BSP (CPU #0) */
455
456			cpu_reset_proxyid = PCPU_GET(cpuid);
457			cpustop_restartfunc = cpu_reset_proxy;
458			cpu_reset_proxy_active = 0;
459			printf("cpu_reset: Restarting BSP\n");
460			started_cpus = (1<<0);		/* Restart CPU #0 */
461
462			cnt = 0;
463			while (cpu_reset_proxy_active == 0 && cnt < 10000000)
464				cnt++;	/* Wait for BSP to announce restart */
465			if (cpu_reset_proxy_active == 0)
466				printf("cpu_reset: Failed to restart BSP\n");
467			enable_intr();
468			cpu_reset_proxy_active = 2;
469
470			while (1);
471			/* NOTREACHED */
472		}
473	}
474#else
475	cpu_reset_real();
476#endif
477}
478
479static void
480cpu_reset_real()
481{
482
483#ifdef PC98
484	/*
485	 * Attempt to do a CPU reset via CPU reset port.
486	 */
487	disable_intr();
488	if ((inb(0x35) & 0xa0) != 0xa0) {
489		outb(0x37, 0x0f);		/* SHUT0 = 0. */
490		outb(0x37, 0x0b);		/* SHUT1 = 0. */
491	}
492	outb(0xf0, 0x00);		/* Reset. */
493#else
494	/*
495	 * Attempt to do a CPU reset via the keyboard controller,
496	 * do not turn of the GateA20, as any machine that fails
497	 * to do the reset here would then end up in no man's land.
498	 */
499
500#if !defined(BROKEN_KEYBOARD_RESET)
501	outb(IO_KBD + 4, 0xFE);
502	DELAY(500000);	/* wait 0.5 sec to see if that did it */
503	printf("Keyboard reset did not work, attempting CPU shutdown\n");
504	DELAY(1000000);	/* wait 1 sec for printf to complete */
505#endif
506#endif /* PC98 */
507	/* force a shutdown by unmapping entire address space ! */
508	bzero((caddr_t) PTD, PAGE_SIZE);
509
510	/* "good night, sweet prince .... <THUNK!>" */
511	invltlb();
512	/* NOTREACHED */
513	while(1);
514}
515
516int
517grow_stack(p, sp)
518	struct proc *p;
519	u_int sp;
520{
521	int rv;
522
523	rv = vm_map_growstack (p, sp);
524	if (rv != KERN_SUCCESS)
525		return (0);
526
527	return (1);
528}
529
530/*
531 * Software interrupt handler for queued VM system processing.
532 */
533void
534swi_vm(void *dummy)
535{
536	if (busdma_swi_pending != 0)
537		busdma_swi();
538}
539
540/*
541 * Tell whether this address is in some physical memory region.
542 * Currently used by the kernel coredump code in order to avoid
543 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
544 * or other unpredictable behaviour.
545 */
546
547int
548is_physical_memory(addr)
549	vm_offset_t addr;
550{
551
552#ifdef DEV_ISA
553	/* The ISA ``memory hole''. */
554	if (addr >= 0xa0000 && addr < 0x100000)
555		return 0;
556#endif
557
558	/*
559	 * stuff other tests for known memory-mapped devices (PCI?)
560	 * here
561	 */
562
563	return 1;
564}
565