vm_machdep.c revision 93264 
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 93264 2002-03-27 05:39:23Z dillon $
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(void);
90#ifdef SMP
91static void	cpu_reset_proxy(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	register_t 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 = intr_disable();
156	if (PCPU_GET(fpcurthread) == td1)
157		npxsave(&td1->td_pcb->pcb_save);
158	intr_restore(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	pcb2->pcb_psl = td2->td_frame->tf_eflags & ~PSL_I; /* ints disabled */
197	/*-
198	 * pcb2->pcb_dr*:	cloned above.
199	 * pcb2->pcb_savefpu:	cloned above.
200	 * pcb2->pcb_flags:	cloned above.
201	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
202	 * pcb2->pcb_gs:	cloned above.
203	 * pcb2->pcb_ext:	cleared below.
204	 */
205
206	/*
207	 * XXX don't copy the i/o pages.  this should probably be fixed.
208	 */
209	pcb2->pcb_ext = 0;
210
211        /* Copy the LDT, if necessary. */
212	mtx_lock_spin(&sched_lock);
213        if (mdp2->md_ldt != 0) {
214		if (flags & RFMEM) {
215			mdp2->md_ldt->ldt_refcnt++;
216		} else {
217			mdp2->md_ldt = user_ldt_alloc(mdp2,
218			    mdp2->md_ldt->ldt_len);
219			if (mdp2->md_ldt == NULL)
220				panic("could not copy LDT");
221		}
222        }
223	mtx_unlock_spin(&sched_lock);
224
225	/*
226	 * Now, cpu_switch() can schedule the new process.
227	 * pcb_esp is loaded pointing to the cpu_switch() stack frame
228	 * containing the return address when exiting cpu_switch.
229	 * This will normally be to fork_trampoline(), which will have
230	 * %ebx loaded with the new proc's pointer.  fork_trampoline()
231	 * will set up a stack to call fork_return(p, frame); to complete
232	 * the return to user-mode.
233	 */
234}
235
236/*
237 * Intercept the return address from a freshly forked process that has NOT
238 * been scheduled yet.
239 *
240 * This is needed to make kernel threads stay in kernel mode.
241 */
242void
243cpu_set_fork_handler(td, func, arg)
244	struct thread *td;
245	void (*func)(void *);
246	void *arg;
247{
248	/*
249	 * Note that the trap frame follows the args, so the function
250	 * is really called like this:  func(arg, frame);
251	 */
252	td->td_pcb->pcb_esi = (int) func;	/* function */
253	td->td_pcb->pcb_ebx = (int) arg;	/* first arg */
254}
255
256void
257cpu_exit(td)
258	register struct thread *td;
259{
260	struct pcb *pcb = td->td_pcb;
261	struct mdproc *mdp = &td->td_proc->p_md;
262#ifdef DEV_NPX
263	npxexit(td);
264#endif
265	if (pcb->pcb_ext != 0) {
266	        /*
267		 * XXX do we need to move the TSS off the allocated pages
268		 * before freeing them?  (not done here)
269		 */
270		kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
271		    ctob(IOPAGES + 1));
272		pcb->pcb_ext = 0;
273	}
274	if (mdp->md_ldt)
275		user_ldt_free(td);
276        if (pcb->pcb_flags & PCB_DBREGS) {
277                /*
278                 * disable all hardware breakpoints
279                 */
280                reset_dbregs();
281                pcb->pcb_flags &= ~PCB_DBREGS;
282        }
283}
284
285void
286cpu_wait(p)
287	struct proc *p;
288{
289}
290
291/*
292 * Dump the machine specific header information at the start of a core dump.
293 */
294int
295cpu_coredump(td, vp, cred)
296	struct thread *td;
297	struct vnode *vp;
298	struct ucred *cred;
299{
300	struct proc *p = td->td_proc;
301	int error;
302	caddr_t tempuser;
303
304	tempuser = malloc(ctob(UAREA_PAGES + KSTACK_PAGES), M_TEMP, M_WAITOK | M_ZERO);
305	if (!tempuser)
306		return EINVAL;
307
308	bcopy(p->p_uarea, tempuser, sizeof(struct user));
309#if 0		/* XXXKSE - broken, fixme!!!!! td_frame is in kstack! */
310	bcopy(td->td_frame,
311	      tempuser + ((caddr_t) td->td_frame - (caddr_t) p->p_uarea),
312	      sizeof(struct trapframe));
313#endif
314
315	error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser,
316			ctob(UAREA_PAGES + KSTACK_PAGES),
317			(off_t)0, UIO_SYSSPACE, IO_UNIT, cred, (int *)NULL, td);
318
319	free(tempuser, M_TEMP);
320
321	return error;
322}
323
324/*
325 * Convert kernel VA to physical address
326 */
327u_long
328kvtop(void *addr)
329{
330	vm_offset_t va;
331
332	va = pmap_kextract((vm_offset_t)addr);
333	if (va == 0)
334		panic("kvtop: zero page frame");
335	return((int)va);
336}
337
338/*
339 * Map an IO request into kernel virtual address space.
340 *
341 * All requests are (re)mapped into kernel VA space.
342 * Notice that we use b_bufsize for the size of the buffer
343 * to be mapped.  b_bcount might be modified by the driver.
344 */
345void
346vmapbuf(bp)
347	register struct buf *bp;
348{
349	register caddr_t addr, kva;
350	vm_offset_t pa;
351	int pidx;
352	struct vm_page *m;
353
354	GIANT_REQUIRED;
355
356	if ((bp->b_flags & B_PHYS) == 0)
357		panic("vmapbuf");
358
359	for (addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data), pidx = 0;
360	     addr < bp->b_data + bp->b_bufsize;
361	     addr += PAGE_SIZE, pidx++) {
362		/*
363		 * Do the vm_fault if needed; do the copy-on-write thing
364		 * when reading stuff off device into memory.
365		 */
366		vm_fault_quick((addr >= bp->b_data) ? addr : bp->b_data,
367			(bp->b_iocmd == BIO_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ);
368		pa = trunc_page(pmap_kextract((vm_offset_t) addr));
369		if (pa == 0)
370			panic("vmapbuf: page not present");
371		m = PHYS_TO_VM_PAGE(pa);
372		vm_page_hold(m);
373		bp->b_pages[pidx] = m;
374	}
375	if (pidx > btoc(MAXPHYS))
376		panic("vmapbuf: mapped more than MAXPHYS");
377	pmap_qenter((vm_offset_t)bp->b_saveaddr, bp->b_pages, pidx);
378
379	kva = bp->b_saveaddr;
380	bp->b_npages = pidx;
381	bp->b_saveaddr = bp->b_data;
382	bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK);
383}
384
385/*
386 * Free the io map PTEs associated with this IO operation.
387 * We also invalidate the TLB entries and restore the original b_addr.
388 */
389void
390vunmapbuf(bp)
391	register struct buf *bp;
392{
393	int pidx;
394	int npages;
395	vm_page_t *m;
396
397	GIANT_REQUIRED;
398
399	if ((bp->b_flags & B_PHYS) == 0)
400		panic("vunmapbuf");
401
402	npages = bp->b_npages;
403	pmap_qremove(trunc_page((vm_offset_t)bp->b_data),
404		     npages);
405	m = bp->b_pages;
406	for (pidx = 0; pidx < npages; pidx++)
407		vm_page_unhold(*m++);
408
409	bp->b_data = bp->b_saveaddr;
410}
411
412/*
413 * Force reset the processor by invalidating the entire address space!
414 */
415
416#ifdef SMP
417static void
418cpu_reset_proxy()
419{
420
421	cpu_reset_proxy_active = 1;
422	while (cpu_reset_proxy_active == 1)
423		;	 /* Wait for other cpu to see that we've started */
424	stop_cpus((1<<cpu_reset_proxyid));
425	printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
426	DELAY(1000000);
427	cpu_reset_real();
428}
429#endif
430
431void
432cpu_reset()
433{
434#ifdef SMP
435	if (smp_active == 0) {
436		cpu_reset_real();
437		/* NOTREACHED */
438	} else {
439
440		u_int map;
441		int cnt;
442		printf("cpu_reset called on cpu#%d\n", PCPU_GET(cpuid));
443
444		map = PCPU_GET(other_cpus) & ~ stopped_cpus;
445
446		if (map != 0) {
447			printf("cpu_reset: Stopping other CPUs\n");
448			stop_cpus(map);		/* Stop all other CPUs */
449		}
450
451		if (PCPU_GET(cpuid) == 0) {
452			DELAY(1000000);
453			cpu_reset_real();
454			/* NOTREACHED */
455		} else {
456			/* We are not BSP (CPU #0) */
457
458			cpu_reset_proxyid = PCPU_GET(cpuid);
459			cpustop_restartfunc = cpu_reset_proxy;
460			cpu_reset_proxy_active = 0;
461			printf("cpu_reset: Restarting BSP\n");
462			started_cpus = (1<<0);		/* Restart CPU #0 */
463
464			cnt = 0;
465			while (cpu_reset_proxy_active == 0 && cnt < 10000000)
466				cnt++;	/* Wait for BSP to announce restart */
467			if (cpu_reset_proxy_active == 0)
468				printf("cpu_reset: Failed to restart BSP\n");
469			enable_intr();
470			cpu_reset_proxy_active = 2;
471
472			while (1);
473			/* NOTREACHED */
474		}
475	}
476#else
477	cpu_reset_real();
478#endif
479}
480
481static void
482cpu_reset_real()
483{
484
485#ifdef PC98
486	/*
487	 * Attempt to do a CPU reset via CPU reset port.
488	 */
489	disable_intr();
490	if ((inb(0x35) & 0xa0) != 0xa0) {
491		outb(0x37, 0x0f);		/* SHUT0 = 0. */
492		outb(0x37, 0x0b);		/* SHUT1 = 0. */
493	}
494	outb(0xf0, 0x00);		/* Reset. */
495#else
496	/*
497	 * Attempt to do a CPU reset via the keyboard controller,
498	 * do not turn of the GateA20, as any machine that fails
499	 * to do the reset here would then end up in no man's land.
500	 */
501
502#if !defined(BROKEN_KEYBOARD_RESET)
503	outb(IO_KBD + 4, 0xFE);
504	DELAY(500000);	/* wait 0.5 sec to see if that did it */
505	printf("Keyboard reset did not work, attempting CPU shutdown\n");
506	DELAY(1000000);	/* wait 1 sec for printf to complete */
507#endif
508#endif /* PC98 */
509	/* force a shutdown by unmapping entire address space ! */
510	bzero((caddr_t) PTD, PAGE_SIZE);
511
512	/* "good night, sweet prince .... <THUNK!>" */
513	invltlb();
514	/* NOTREACHED */
515	while(1);
516}
517
518/*
519 * Software interrupt handler for queued VM system processing.
520 */
521void
522swi_vm(void *dummy)
523{
524	if (busdma_swi_pending != 0)
525		busdma_swi();
526}
527
528/*
529 * Tell whether this address is in some physical memory region.
530 * Currently used by the kernel coredump code in order to avoid
531 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
532 * or other unpredictable behaviour.
533 */
534
535int
536is_physical_memory(addr)
537	vm_offset_t addr;
538{
539
540#ifdef DEV_ISA
541	/* The ISA ``memory hole''. */
542	if (addr >= 0xa0000 && addr < 0x100000)
543		return 0;
544#endif
545
546	/*
547	 * stuff other tests for known memory-mapped devices (PCI?)
548	 * here
549	 */
550
551	return 1;
552}
553