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
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.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
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.
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
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: src/sys/i386/i386/vm_machdep.c,v 1.132.2.9 2003/01/25 19:02:23 dillon Exp $
42 * $DragonFly: src/sys/i386/i386/Attic/vm_machdep.c,v 1.2 2003/06/17 04:28:35 dillon Exp $
46 #include "opt_user_ldt.h"
50 #include "opt_reset.h"
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/malloc.h>
57 #include <sys/vnode.h>
58 #include <sys/vmmeter.h>
59 #include <sys/kernel.h>
60 #include <sys/sysctl.h>
61 #include <sys/unistd.h>
63 #include <machine/clock.h>
64 #include <machine/cpu.h>
65 #include <machine/md_var.h>
67 #include <machine/smp.h>
69 #include <machine/pcb.h>
70 #include <machine/pcb_ext.h>
71 #include <machine/vm86.h>
74 #include <vm/vm_param.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>
84 #include <pc98/pc98/pc98.h>
86 #include <i386/isa/isa.h>
89 static void cpu_reset_real __P((void));
91 static void cpu_reset_proxy __P((void));
92 static u_int cpu_reset_proxyid;
93 static volatile u_int cpu_reset_proxy_active;
95 extern int _ucodesel, _udatasel;
98 * quick version of vm_fault
101 vm_fault_quick(v, prot)
107 if (prot & VM_PROT_WRITE)
108 r = subyte(v, fubyte(v));
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.
120 cpu_fork(p1, p2, flags)
121 register struct proc *p1, *p2;
126 if ((flags & RFPROC) == 0) {
128 if ((flags & RFMEM) == 0) {
129 /* unshare user LDT */
130 struct pcb *pcb1 = &p1->p_addr->u_pcb;
131 struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt;
132 if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) {
133 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len);
135 pcb1->pcb_ldt = pcb_ldt;
144 /* Ensure that p1's pcb is up to date. */
146 npxsave(&p1->p_addr->u_pcb.pcb_save);
150 p2->p_addr->u_pcb = p1->p_addr->u_pcb;
151 pcb2 = &p2->p_addr->u_pcb;
154 * Create a new fresh stack for the new process.
155 * Copy the trap frame for the return to user mode as if from a
156 * syscall. This copies the user mode register values.
158 p2->p_md.md_regs = (struct trapframe *)
159 ((int)p2->p_addr + UPAGES * PAGE_SIZE - 16) - 1;
160 bcopy(p1->p_md.md_regs, p2->p_md.md_regs, sizeof(*p2->p_md.md_regs));
163 * Set registers for trampoline to user mode. Leave space for the
164 * return address on stack. These are the kernel mode register values.
166 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
168 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */
170 pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *);
171 pcb2->pcb_ebx = (int)p2; /* fork_trampoline argument */
172 pcb2->pcb_eip = (int)fork_trampoline;
174 * pcb2->pcb_ldt: duplicated below, if necessary.
175 * pcb2->pcb_savefpu: cloned above.
176 * pcb2->pcb_flags: cloned above (always 0 here?).
177 * pcb2->pcb_onfault: cloned above (always NULL here?).
181 pcb2->pcb_mpnest = 1;
184 * XXX don't copy the i/o pages. this should probably be fixed.
189 /* Copy the LDT, if necessary. */
190 if (pcb2->pcb_ldt != 0) {
192 pcb2->pcb_ldt->ldt_refcnt++;
194 pcb2->pcb_ldt = user_ldt_alloc(pcb2,
195 pcb2->pcb_ldt->ldt_len);
201 * Now, cpu_switch() can schedule the new process.
202 * pcb_esp is loaded pointing to the cpu_switch() stack frame
203 * containing the return address when exiting cpu_switch.
204 * This will normally be to fork_trampoline(), which will have
205 * %ebx loaded with the new proc's pointer. fork_trampoline()
206 * will set up a stack to call fork_return(p, frame); to complete
207 * the return to user-mode.
212 * Intercept the return address from a freshly forked process that has NOT
213 * been scheduled yet.
215 * This is needed to make kernel threads stay in kernel mode.
218 cpu_set_fork_handler(p, func, arg)
220 void (*func) __P((void *));
224 * Note that the trap frame follows the args, so the function
225 * is really called like this: func(arg, frame);
227 p->p_addr->u_pcb.pcb_esi = (int) func; /* function */
228 p->p_addr->u_pcb.pcb_ebx = (int) arg; /* first arg */
233 register struct proc *p;
235 struct pcb *pcb = &p->p_addr->u_pcb;
240 if (pcb->pcb_ext != 0) {
242 * XXX do we need to move the TSS off the allocated pages
243 * before freeing them? (not done here)
245 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
252 if (pcb->pcb_flags & PCB_DBREGS) {
254 * disable all hardware breakpoints
257 pcb->pcb_flags &= ~PCB_DBREGS;
268 /* drop per-process resources */
269 pmap_dispose_proc(p);
273 * Dump the machine specific header information at the start of a core dump.
276 cpu_coredump(p, vp, cred)
284 tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK);
288 bzero(tempuser, ctob(UPAGES));
289 bcopy(p->p_addr, tempuser, sizeof(struct user));
290 bcopy(p->p_md.md_regs,
291 tempuser + ((caddr_t) p->p_md.md_regs - (caddr_t) p->p_addr),
292 sizeof(struct trapframe));
294 error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, ctob(UPAGES),
295 (off_t)0, UIO_SYSSPACE, IO_UNIT, cred, (int *)NULL, p);
297 free(tempuser, M_TEMP);
304 setredzone(pte, vaddr)
308 /* eventually do this by setting up an expand-down stack segment
309 for ss0: selector, allowing stack access down to top of u.
310 this means though that protection violations need to be handled
311 thru a double fault exception that must do an integral task
312 switch to a known good context, within which a dump can be
313 taken. a sensible scheme might be to save the initial context
314 used by sched (that has physical memory mapped 1:1 at bottom)
315 and take the dump while still in mapped mode */
320 * Convert kernel VA to physical address
327 va = pmap_kextract((vm_offset_t)addr);
329 panic("kvtop: zero page frame");
334 * Force reset the processor by invalidating the entire address space!
343 cpu_reset_proxy_active = 1;
344 while (cpu_reset_proxy_active == 1)
345 ; /* Wait for other cpu to disable interupts */
346 saved_mp_lock = mp_lock;
348 printf("cpu_reset_proxy: Grabbed mp lock for BSP\n");
349 cpu_reset_proxy_active = 3;
350 while (cpu_reset_proxy_active == 3)
351 ; /* Wait for other cpu to enable interrupts */
352 stop_cpus((1<<cpu_reset_proxyid));
353 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
363 if (smp_active == 0) {
370 printf("cpu_reset called on cpu#%d\n",cpuid);
372 map = other_cpus & ~ stopped_cpus;
375 printf("cpu_reset: Stopping other CPUs\n");
376 stop_cpus(map); /* Stop all other CPUs */
384 /* We are not BSP (CPU #0) */
386 cpu_reset_proxyid = cpuid;
387 cpustop_restartfunc = cpu_reset_proxy;
388 printf("cpu_reset: Restarting BSP\n");
389 started_cpus = (1<<0); /* Restart CPU #0 */
392 while (cpu_reset_proxy_active == 0 && cnt < 10000000)
393 cnt++; /* Wait for BSP to announce restart */
394 if (cpu_reset_proxy_active == 0)
395 printf("cpu_reset: Failed to restart BSP\n");
396 __asm __volatile("cli" : : : "memory");
397 cpu_reset_proxy_active = 2;
399 while (cpu_reset_proxy_active == 2 && cnt < 10000000)
400 cnt++; /* Do nothing */
401 if (cpu_reset_proxy_active == 2) {
402 printf("cpu_reset: BSP did not grab mp lock\n");
403 cpu_reset_real(); /* XXX: Bogus ? */
405 cpu_reset_proxy_active = 4;
406 __asm __volatile("sti" : : : "memory");
422 * Attempt to do a CPU reset via CPU reset port.
425 if ((inb(0x35) & 0xa0) != 0xa0) {
426 outb(0x37, 0x0f); /* SHUT0 = 0. */
427 outb(0x37, 0x0b); /* SHUT1 = 0. */
429 outb(0xf0, 0x00); /* Reset. */
432 * Attempt to do a CPU reset via the keyboard controller,
433 * do not turn of the GateA20, as any machine that fails
434 * to do the reset here would then end up in no man's land.
437 #if !defined(BROKEN_KEYBOARD_RESET)
438 outb(IO_KBD + 4, 0xFE);
439 DELAY(500000); /* wait 0.5 sec to see if that did it */
440 printf("Keyboard reset did not work, attempting CPU shutdown\n");
441 DELAY(1000000); /* wait 1 sec for printf to complete */
444 /* force a shutdown by unmapping entire address space ! */
445 bzero((caddr_t) PTD, PAGE_SIZE);
447 /* "good night, sweet prince .... <THUNK!>" */
460 rv = vm_map_growstack (p, sp);
461 if (rv != KERN_SUCCESS)
467 SYSCTL_DECL(_vm_stats_misc);
469 static int cnt_prezero;
471 SYSCTL_INT(_vm_stats_misc, OID_AUTO,
472 cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, "");
475 * Implement the pre-zeroed page mechanism.
476 * This routine is called from the idle loop.
479 #define ZIDLE_LO(v) ((v) * 2 / 3)
480 #define ZIDLE_HI(v) ((v) * 4 / 5)
485 static int free_rover;
486 static int zero_state;
491 * Attempt to maintain approximately 1/2 of our free pages in a
492 * PG_ZERO'd state. Add some hysteresis to (attempt to) avoid
493 * generally zeroing a page when the system is near steady-state.
494 * Otherwise we might get 'flutter' during disk I/O / IPC or
495 * fast sleeps. We also do not want to be continuously zeroing
496 * pages because doing so may flush our L1 and L2 caches too much.
499 if (zero_state && vm_page_zero_count >= ZIDLE_LO(cnt.v_free_count))
501 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count))
508 __asm __volatile("sti" : : : "memory");
510 m = vm_page_list_find(PQ_FREE, free_rover, FALSE);
511 if (m != NULL && (m->flags & PG_ZERO) == 0) {
512 vm_page_queues[m->queue].lcnt--;
513 TAILQ_REMOVE(&vm_page_queues[m->queue].pl, m, pageq);
516 pmap_zero_page(VM_PAGE_TO_PHYS(m));
518 vm_page_flag_set(m, PG_ZERO);
519 m->queue = PQ_FREE + m->pc;
520 vm_page_queues[m->queue].lcnt++;
521 TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m,
523 ++vm_page_zero_count;
525 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count))
528 free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK;
530 __asm __volatile("cli" : : : "memory");
539 * We have to enable interrupts for a moment if the try_mplock fails
540 * in order to potentially take an IPI. XXX this should be in
543 __asm __volatile("sti; nop; cli" : : : "memory");
548 * Software interrupt handler for queued VM system processing.
553 if (busdma_swi_pending != 0)
558 * Tell whether this address is in some physical memory region.
559 * Currently used by the kernel coredump code in order to avoid
560 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
561 * or other unpredictable behaviour.
567 is_physical_memory(addr)
572 /* The ISA ``memory hole''. */
573 if (addr >= 0xa0000 && addr < 0x100000)
578 * stuff other tests for known memory-mapped devices (PCI?)