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 1.34 2005/02/27 10:57:24 swildner Exp $
47 #include "opt_reset.h"
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/malloc.h>
54 #include <sys/interrupt.h>
55 #include <sys/vnode.h>
56 #include <sys/vmmeter.h>
57 #include <sys/kernel.h>
58 #include <sys/sysctl.h>
59 #include <sys/unistd.h>
61 #include <machine/clock.h>
62 #include <machine/cpu.h>
63 #include <machine/md_var.h>
65 #include <machine/smp.h>
67 #include <machine/pcb.h>
68 #include <machine/pcb_ext.h>
69 #include <machine/vm86.h>
70 #include <machine/globaldata.h> /* npxthread */
71 #include <machine/ipl.h> /* SWI_ */
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>
82 #include <sys/thread2.h>
84 #include <bus/isa/i386/isa.h>
86 static void cpu_reset_real (void);
88 static void cpu_reset_proxy (void);
89 static u_int cpu_reset_proxyid;
90 static volatile u_int cpu_reset_proxy_active;
92 extern int _ucodesel, _udatasel;
96 * Finish a fork operation, with process p2 nearly set up.
97 * Copy and update the pcb, set up the stack so that the child
98 * ready to run and return to user mode.
101 cpu_fork(p1, p2, flags)
102 struct proc *p1, *p2;
107 if ((flags & RFPROC) == 0) {
108 if ((flags & RFMEM) == 0) {
109 /* unshare user LDT */
110 struct pcb *pcb1 = p1->p_thread->td_pcb;
111 struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt;
112 if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) {
113 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len);
115 pcb1->pcb_ldt = pcb_ldt;
123 /* Ensure that p1's pcb is up to date. */
124 if (mdcpu->gd_npxthread == p1->p_thread)
125 npxsave(p1->p_thread->td_savefpu);
129 pcb2 = p2->p_thread->td_pcb;
130 *pcb2 = *p1->p_thread->td_pcb;
133 * Create a new fresh stack for the new process.
134 * Copy the trap frame for the return to user mode as if from a
135 * syscall. This copies the user mode register values. The
136 * 16 byte offset saves space for vm86, and must match
137 * common_tss.esp0 (kernel stack pointer on entry from user mode)
139 * pcb_esp must allocate an additional call-return pointer below
140 * the trap frame which will be restored by cpu_restore from
141 * PCB_EIP, and the thread's td_sp pointer must allocate an
142 * additonal two worsd below the pcb_esp call-return pointer to
143 * hold the LWKT restore function pointer and eflags.
145 * The LWKT restore function pointer must be set to cpu_restore,
146 * which is our standard heavy weight process switch-in function.
147 * YYY eventually we should shortcut fork_return and fork_trampoline
148 * to use the LWKT restore function directly so we can get rid of
149 * all the extra crap we are setting up.
151 p2->p_md.md_regs = (struct trapframe *)((char *)pcb2 - 16) - 1;
152 bcopy(p1->p_md.md_regs, p2->p_md.md_regs, sizeof(*p2->p_md.md_regs));
155 * Set registers for trampoline to user mode. Leave space for the
156 * return address on stack. These are the kernel mode register values.
158 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
160 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */
162 pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *);
163 pcb2->pcb_ebx = (int)p2; /* fork_trampoline argument */
164 pcb2->pcb_eip = (int)fork_trampoline;
165 p2->p_thread->td_sp = (char *)(pcb2->pcb_esp - sizeof(void *));
166 *(u_int32_t *)p2->p_thread->td_sp = PSL_USER;
167 p2->p_thread->td_sp -= sizeof(void *);
168 *(void **)p2->p_thread->td_sp = (void *)cpu_heavy_restore;
170 * pcb2->pcb_ldt: duplicated below, if necessary.
171 * pcb2->pcb_savefpu: cloned above.
172 * pcb2->pcb_flags: cloned above (always 0 here?).
173 * pcb2->pcb_onfault: cloned above (always NULL here?).
177 * XXX don't copy the i/o pages. this should probably be fixed.
181 /* Copy the LDT, if necessary. */
182 if (pcb2->pcb_ldt != 0) {
184 pcb2->pcb_ldt->ldt_refcnt++;
186 pcb2->pcb_ldt = user_ldt_alloc(pcb2,
187 pcb2->pcb_ldt->ldt_len);
190 bcopy(&p1->p_thread->td_tls, &p2->p_thread->td_tls,
191 sizeof(p2->p_thread->td_tls));
193 * Now, cpu_switch() can schedule the new process.
194 * pcb_esp is loaded pointing to the cpu_switch() stack frame
195 * containing the return address when exiting cpu_switch.
196 * This will normally be to fork_trampoline(), which will have
197 * %ebx loaded with the new proc's pointer. fork_trampoline()
198 * will set up a stack to call fork_return(p, frame); to complete
199 * the return to user-mode.
204 * Intercept the return address from a freshly forked process that has NOT
205 * been scheduled yet.
207 * This is needed to make kernel threads stay in kernel mode.
210 cpu_set_fork_handler(p, func, arg)
212 void (*func) (void *);
216 * Note that the trap frame follows the args, so the function
217 * is really called like this: func(arg, frame);
219 p->p_thread->td_pcb->pcb_esi = (int) func; /* function */
220 p->p_thread->td_pcb->pcb_ebx = (int) arg; /* first arg */
224 cpu_set_thread_handler(thread_t td, void (*rfunc)(void), void *func, void *arg)
226 td->td_pcb->pcb_esi = (int)func;
227 td->td_pcb->pcb_ebx = (int) arg;
228 td->td_switch = cpu_lwkt_switch;
229 td->td_sp -= sizeof(void *);
230 *(void **)td->td_sp = rfunc; /* exit function on return */
231 td->td_sp -= sizeof(void *);
232 *(void **)td->td_sp = cpu_kthread_restore;
238 struct thread *td = curthread;
243 KKASSERT(td->td_proc);
244 npxexit(td->td_proc);
251 if (pcb->pcb_ext != 0) {
253 * XXX do we need to move the TSS off the allocated pages
254 * before freeing them? (not done here)
256 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
261 if (pcb->pcb_flags & PCB_DBREGS) {
263 * disable all hardware breakpoints
266 pcb->pcb_flags &= ~PCB_DBREGS;
268 td->td_gd->gd_cnt.v_swtch++;
270 crit_enter_quick(td);
271 lwkt_deschedule_self(td);
276 * Terminate the current thread. The caller must have already acquired
277 * the thread's rwlock and placed it on a reap list or otherwise notified
278 * a reaper of its existance. We set a special assembly switch function which
279 * releases td_rwlock after it has cleaned up the MMU state and switched
282 * Must be caller from a critical section and with the thread descheduled.
285 cpu_thread_exit(void)
287 curthread->td_switch = cpu_exit_switch;
293 * Process Reaper. Called after the caller has acquired the thread's
294 * rwlock and removed it from the reap list.
297 cpu_proc_wait(struct proc *p)
301 /* drop per-process resources */
302 td = pmap_dispose_proc(p);
304 lwkt_free_thread(td);
308 * Dump the machine specific header information at the start of a core dump.
311 cpu_coredump(struct thread *td, struct vnode *vp, struct ucred *cred)
313 struct proc *p = td->td_proc;
318 tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK);
322 bzero(tempuser, ctob(UPAGES));
323 bcopy(p->p_addr, tempuser, sizeof(struct user));
324 bcopy(p->p_md.md_regs,
325 tempuser + ((caddr_t) p->p_md.md_regs - (caddr_t) p->p_addr),
326 sizeof(struct trapframe));
327 bcopy(p->p_thread->td_pcb, tempuser + ((char *)p->p_thread->td_pcb - (char *)p->p_addr), sizeof(struct pcb));
329 error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, ctob(UPAGES),
330 (off_t)0, UIO_SYSSPACE, IO_UNIT, cred, (int *)NULL, td);
332 free(tempuser, M_TEMP);
339 setredzone(pte, vaddr)
343 /* eventually do this by setting up an expand-down stack segment
344 for ss0: selector, allowing stack access down to top of u.
345 this means though that protection violations need to be handled
346 thru a double fault exception that must do an integral task
347 switch to a known good context, within which a dump can be
348 taken. a sensible scheme might be to save the initial context
349 used by sched (that has physical memory mapped 1:1 at bottom)
350 and take the dump while still in mapped mode */
355 * Convert kernel VA to physical address
362 pa = pmap_kextract((vm_offset_t)addr);
364 panic("kvtop: zero page frame");
369 * Force reset the processor by invalidating the entire address space!
378 cpu_reset_proxy_active = 1;
379 while (cpu_reset_proxy_active == 1)
380 ; /* Wait for other cpu to disable interupts */
381 saved_mp_lock = mp_lock;
382 mp_lock = 0; /* BSP */
383 printf("cpu_reset_proxy: Grabbed mp lock for BSP\n");
384 cpu_reset_proxy_active = 3;
385 while (cpu_reset_proxy_active == 3)
386 ; /* Wait for other cpu to enable interrupts */
387 stop_cpus((1<<cpu_reset_proxyid));
388 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
398 if (smp_active_mask == 1) {
404 printf("cpu_reset called on cpu#%d\n",mycpu->gd_cpuid);
406 map = mycpu->gd_other_cpus & ~stopped_cpus & smp_active_mask;
409 printf("cpu_reset: Stopping other CPUs\n");
410 stop_cpus(map); /* Stop all other CPUs */
413 if (mycpu->gd_cpuid == 0) {
418 /* We are not BSP (CPU #0) */
420 cpu_reset_proxyid = mycpu->gd_cpuid;
421 cpustop_restartfunc = cpu_reset_proxy;
422 printf("cpu_reset: Restarting BSP\n");
423 started_cpus = (1<<0); /* Restart CPU #0 */
426 while (cpu_reset_proxy_active == 0 && cnt < 10000000)
427 cnt++; /* Wait for BSP to announce restart */
428 if (cpu_reset_proxy_active == 0)
429 printf("cpu_reset: Failed to restart BSP\n");
430 __asm __volatile("cli" : : : "memory");
431 cpu_reset_proxy_active = 2;
433 while (cpu_reset_proxy_active == 2 && cnt < 10000000)
434 cnt++; /* Do nothing */
435 if (cpu_reset_proxy_active == 2) {
436 printf("cpu_reset: BSP did not grab mp lock\n");
437 cpu_reset_real(); /* XXX: Bogus ? */
439 cpu_reset_proxy_active = 4;
440 __asm __volatile("sti" : : : "memory");
454 * Attempt to do a CPU reset via the keyboard controller,
455 * do not turn of the GateA20, as any machine that fails
456 * to do the reset here would then end up in no man's land.
459 #if !defined(BROKEN_KEYBOARD_RESET)
460 outb(IO_KBD + 4, 0xFE);
461 DELAY(500000); /* wait 0.5 sec to see if that did it */
462 printf("Keyboard reset did not work, attempting CPU shutdown\n");
463 DELAY(1000000); /* wait 1 sec for printf to complete */
465 /* force a shutdown by unmapping entire address space ! */
466 bzero((caddr_t) PTD, PAGE_SIZE);
468 /* "good night, sweet prince .... <THUNK!>" */
481 rv = vm_map_growstack (p, sp);
482 if (rv != KERN_SUCCESS)
488 SYSCTL_DECL(_vm_stats_misc);
490 static int cnt_prezero;
492 SYSCTL_INT(_vm_stats_misc, OID_AUTO,
493 cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, "");
496 * Implement the pre-zeroed page mechanism.
497 * This routine is called from the idle loop.
500 #define ZIDLE_LO(v) ((v) * 2 / 3)
501 #define ZIDLE_HI(v) ((v) * 4 / 5)
506 static int free_rover;
507 static int zero_state;
511 * Attempt to maintain approximately 1/2 of our free pages in a
512 * PG_ZERO'd state. Add some hysteresis to (attempt to) avoid
513 * generally zeroing a page when the system is near steady-state.
514 * Otherwise we might get 'flutter' during disk I/O / IPC or
515 * fast sleeps. We also do not want to be continuously zeroing
516 * pages because doing so may flush our L1 and L2 caches too much.
519 if (zero_state && vm_page_zero_count >= ZIDLE_LO(vmstats.v_free_count))
521 if (vm_page_zero_count >= ZIDLE_HI(vmstats.v_free_count))
528 __asm __volatile("sti" : : : "memory");
530 m = vm_page_list_find(PQ_FREE, free_rover, FALSE);
531 if (m != NULL && (m->flags & PG_ZERO) == 0) {
532 vm_page_queues[m->queue].lcnt--;
533 TAILQ_REMOVE(&vm_page_queues[m->queue].pl, m, pageq);
536 pmap_zero_page(VM_PAGE_TO_PHYS(m));
538 vm_page_flag_set(m, PG_ZERO);
539 m->queue = PQ_FREE + m->pc;
540 vm_page_queues[m->queue].lcnt++;
541 TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m,
543 ++vm_page_zero_count;
545 if (vm_page_zero_count >= ZIDLE_HI(vmstats.v_free_count))
548 free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK;
550 __asm __volatile("cli" : : : "memory");
559 * We have to enable interrupts for a moment if the try_mplock fails
560 * in order to potentially take an IPI. XXX this should be in
563 __asm __volatile("sti; nop; cli" : : : "memory");
570 if (busdma_swi_pending != 0)
575 swi_vm_setup(void *arg)
577 register_swi(SWI_VM, swi_vm, NULL, "swi_vm", NULL);
580 SYSINIT(vm_setup, SI_SUB_CPU, SI_ORDER_ANY, swi_vm_setup, NULL);
584 * Tell whether this address is in some physical memory region.
585 * Currently used by the kernel coredump code in order to avoid
586 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
587 * or other unpredictable behaviour.
591 is_physical_memory(addr)
596 /* The ISA ``memory hole''. */
597 if (addr >= 0xa0000 && addr < 0x100000)
602 * stuff other tests for known memory-mapped devices (PCI?)