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/platform/pc32/i386/vm_machdep.c,v 1.7 2003/06/20 02:09:50 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_thread->td_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. */
145 if (npxthread == p1->p_thread)
146 npxsave(&p1->p_thread->td_pcb->pcb_save);
150 *p2->p_thread->td_pcb = *p1->p_thread->td_pcb;
151 pcb2 = p2->p_thread->td_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. The
157 * 16 byte offset saves space for vm86, and must match
158 * common_tss.esp0 (kernel stack pointer on entry from user mode)
160 * pcb_esp must allocate an additional call-return pointer below
161 * the trap frame which will be restored by cpu_restore, and the
162 * thread's td_sp pointer must allocate an additonal call-return
163 * pointer below the pcb_esp call-return pointer to hold the LWKT
164 * restore function pointer.
166 * The LWKT restore function pointer must be set to cpu_restore,
167 * which is our standard heavy weight process switch-in function.
168 * YYY eventually we should shortcut fork_return and fork_trampoline
169 * to use the LWKT restore function directly so we can get rid of
170 * all the extra crap we are setting up.
172 p2->p_md.md_regs = (struct trapframe *)
173 ((char *)p2->p_thread->td_pcb - 16) - 1;
174 bcopy(p1->p_md.md_regs, p2->p_md.md_regs, sizeof(*p2->p_md.md_regs));
177 * Set registers for trampoline to user mode. Leave space for the
178 * return address on stack. These are the kernel mode register values.
180 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
182 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */
184 pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *);
185 pcb2->pcb_ebx = (int)p2; /* fork_trampoline argument */
186 pcb2->pcb_eip = (int)fork_trampoline;
187 p2->p_thread->td_sp = (char *)(pcb2->pcb_esp - sizeof(void *));
188 *(void **)p2->p_thread->td_sp = (void *)cpu_heavy_restore;
190 * pcb2->pcb_ldt: duplicated below, if necessary.
191 * pcb2->pcb_savefpu: cloned above.
192 * pcb2->pcb_flags: cloned above (always 0 here?).
193 * pcb2->pcb_onfault: cloned above (always NULL here?).
197 pcb2->pcb_mpnest = 1;
200 * XXX don't copy the i/o pages. this should probably be fixed.
205 /* Copy the LDT, if necessary. */
206 if (pcb2->pcb_ldt != 0) {
208 pcb2->pcb_ldt->ldt_refcnt++;
210 pcb2->pcb_ldt = user_ldt_alloc(pcb2,
211 pcb2->pcb_ldt->ldt_len);
217 * Now, cpu_switch() can schedule the new process.
218 * pcb_esp is loaded pointing to the cpu_switch() stack frame
219 * containing the return address when exiting cpu_switch.
220 * This will normally be to fork_trampoline(), which will have
221 * %ebx loaded with the new proc's pointer. fork_trampoline()
222 * will set up a stack to call fork_return(p, frame); to complete
223 * the return to user-mode.
228 * Intercept the return address from a freshly forked process that has NOT
229 * been scheduled yet.
231 * This is needed to make kernel threads stay in kernel mode.
234 cpu_set_fork_handler(p, func, arg)
236 void (*func) __P((void *));
240 * Note that the trap frame follows the args, so the function
241 * is really called like this: func(arg, frame);
243 p->p_thread->td_pcb->pcb_esi = (int) func; /* function */
244 p->p_thread->td_pcb->pcb_ebx = (int) arg; /* first arg */
249 register struct proc *p;
251 struct pcb *pcb = p->p_thread->td_pcb;
254 * This allows us to disassociate curproc from the thread.
256 if (p->p_thread->td_switch == cpu_heavy_switch)
257 p->p_thread->td_switch = cpu_exit_switch;
258 p->p_thread->td_proc = NULL;
263 if (pcb->pcb_ext != 0) {
265 * XXX do we need to move the TSS off the allocated pages
266 * before freeing them? (not done here)
268 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
275 if (pcb->pcb_flags & PCB_DBREGS) {
277 * disable all hardware breakpoints
280 pcb->pcb_flags &= ~PCB_DBREGS;
293 /* drop per-process resources */
294 td = pmap_dispose_proc(p);
296 pmap_dispose_thread(td);
300 * Dump the machine specific header information at the start of a core dump.
303 cpu_coredump(p, vp, cred)
311 tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK);
315 bzero(tempuser, ctob(UPAGES));
316 bcopy(p->p_addr, tempuser, sizeof(struct user));
317 bcopy(p->p_md.md_regs,
318 tempuser + ((caddr_t) p->p_md.md_regs - (caddr_t) p->p_addr),
319 sizeof(struct trapframe));
320 bcopy(p->p_thread->td_pcb, tempuser + ((char *)p->p_thread->td_pcb - (char *)p->p_addr), sizeof(struct pcb));
322 error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, ctob(UPAGES),
323 (off_t)0, UIO_SYSSPACE, IO_UNIT, cred, (int *)NULL, p);
325 free(tempuser, M_TEMP);
332 setredzone(pte, vaddr)
336 /* eventually do this by setting up an expand-down stack segment
337 for ss0: selector, allowing stack access down to top of u.
338 this means though that protection violations need to be handled
339 thru a double fault exception that must do an integral task
340 switch to a known good context, within which a dump can be
341 taken. a sensible scheme might be to save the initial context
342 used by sched (that has physical memory mapped 1:1 at bottom)
343 and take the dump while still in mapped mode */
348 * Convert kernel VA to physical address
355 va = pmap_kextract((vm_offset_t)addr);
357 panic("kvtop: zero page frame");
362 * Force reset the processor by invalidating the entire address space!
371 cpu_reset_proxy_active = 1;
372 while (cpu_reset_proxy_active == 1)
373 ; /* Wait for other cpu to disable interupts */
374 saved_mp_lock = mp_lock;
376 printf("cpu_reset_proxy: Grabbed mp lock for BSP\n");
377 cpu_reset_proxy_active = 3;
378 while (cpu_reset_proxy_active == 3)
379 ; /* Wait for other cpu to enable interrupts */
380 stop_cpus((1<<cpu_reset_proxyid));
381 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
391 if (smp_active == 0) {
398 printf("cpu_reset called on cpu#%d\n",cpuid);
400 map = other_cpus & ~ stopped_cpus;
403 printf("cpu_reset: Stopping other CPUs\n");
404 stop_cpus(map); /* Stop all other CPUs */
412 /* We are not BSP (CPU #0) */
414 cpu_reset_proxyid = cpuid;
415 cpustop_restartfunc = cpu_reset_proxy;
416 printf("cpu_reset: Restarting BSP\n");
417 started_cpus = (1<<0); /* Restart CPU #0 */
420 while (cpu_reset_proxy_active == 0 && cnt < 10000000)
421 cnt++; /* Wait for BSP to announce restart */
422 if (cpu_reset_proxy_active == 0)
423 printf("cpu_reset: Failed to restart BSP\n");
424 __asm __volatile("cli" : : : "memory");
425 cpu_reset_proxy_active = 2;
427 while (cpu_reset_proxy_active == 2 && cnt < 10000000)
428 cnt++; /* Do nothing */
429 if (cpu_reset_proxy_active == 2) {
430 printf("cpu_reset: BSP did not grab mp lock\n");
431 cpu_reset_real(); /* XXX: Bogus ? */
433 cpu_reset_proxy_active = 4;
434 __asm __volatile("sti" : : : "memory");
450 * Attempt to do a CPU reset via CPU reset port.
453 if ((inb(0x35) & 0xa0) != 0xa0) {
454 outb(0x37, 0x0f); /* SHUT0 = 0. */
455 outb(0x37, 0x0b); /* SHUT1 = 0. */
457 outb(0xf0, 0x00); /* Reset. */
460 * Attempt to do a CPU reset via the keyboard controller,
461 * do not turn of the GateA20, as any machine that fails
462 * to do the reset here would then end up in no man's land.
465 #if !defined(BROKEN_KEYBOARD_RESET)
466 outb(IO_KBD + 4, 0xFE);
467 DELAY(500000); /* wait 0.5 sec to see if that did it */
468 printf("Keyboard reset did not work, attempting CPU shutdown\n");
469 DELAY(1000000); /* wait 1 sec for printf to complete */
472 /* force a shutdown by unmapping entire address space ! */
473 bzero((caddr_t) PTD, PAGE_SIZE);
475 /* "good night, sweet prince .... <THUNK!>" */
488 rv = vm_map_growstack (p, sp);
489 if (rv != KERN_SUCCESS)
495 SYSCTL_DECL(_vm_stats_misc);
497 static int cnt_prezero;
499 SYSCTL_INT(_vm_stats_misc, OID_AUTO,
500 cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, "");
503 * Implement the pre-zeroed page mechanism.
504 * This routine is called from the idle loop.
507 #define ZIDLE_LO(v) ((v) * 2 / 3)
508 #define ZIDLE_HI(v) ((v) * 4 / 5)
513 static int free_rover;
514 static int zero_state;
519 * Attempt to maintain approximately 1/2 of our free pages in a
520 * PG_ZERO'd state. Add some hysteresis to (attempt to) avoid
521 * generally zeroing a page when the system is near steady-state.
522 * Otherwise we might get 'flutter' during disk I/O / IPC or
523 * fast sleeps. We also do not want to be continuously zeroing
524 * pages because doing so may flush our L1 and L2 caches too much.
527 if (zero_state && vm_page_zero_count >= ZIDLE_LO(cnt.v_free_count))
529 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count))
536 __asm __volatile("sti" : : : "memory");
538 m = vm_page_list_find(PQ_FREE, free_rover, FALSE);
539 if (m != NULL && (m->flags & PG_ZERO) == 0) {
540 vm_page_queues[m->queue].lcnt--;
541 TAILQ_REMOVE(&vm_page_queues[m->queue].pl, m, pageq);
544 pmap_zero_page(VM_PAGE_TO_PHYS(m));
546 vm_page_flag_set(m, PG_ZERO);
547 m->queue = PQ_FREE + m->pc;
548 vm_page_queues[m->queue].lcnt++;
549 TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m,
551 ++vm_page_zero_count;
553 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count))
556 free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK;
558 __asm __volatile("cli" : : : "memory");
567 * We have to enable interrupts for a moment if the try_mplock fails
568 * in order to potentially take an IPI. XXX this should be in
571 __asm __volatile("sti; nop; cli" : : : "memory");
576 * Software interrupt handler for queued VM system processing.
581 if (busdma_swi_pending != 0)
586 * Tell whether this address is in some physical memory region.
587 * Currently used by the kernel coredump code in order to avoid
588 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
589 * or other unpredictable behaviour.
595 is_physical_memory(addr)
600 /* The ISA ``memory hole''. */
601 if (addr >= 0xa0000 && addr < 0x100000)
606 * stuff other tests for known memory-mapped devices (PCI?)