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 1.60 2008/05/25 18:47:26 nth 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>
64 #include <machine/smp.h>
65 #include <machine/pcb.h>
66 #include <machine/pcb_ext.h>
67 #include <machine/vm86.h>
68 #include <machine/segments.h>
69 #include <machine/globaldata.h> /* npxthread */
72 #include <vm/vm_param.h>
74 #include <vm/vm_kern.h>
75 #include <vm/vm_page.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_extern.h>
80 #include <sys/thread2.h>
82 #include <bus/isa/i386/isa.h>
84 static void cpu_reset_real (void);
86 static void cpu_reset_proxy (void);
87 static u_int cpu_reset_proxyid
;
88 static volatile u_int cpu_reset_proxy_active
;
90 extern int _ucodesel
, _udatasel
;
94 * Finish a fork operation, with lwp lp2 nearly set up.
95 * Copy and update the pcb, set up the stack so that the child
96 * ready to run and return to user mode.
99 cpu_fork(struct lwp
*lp1
, struct lwp
*lp2
, int flags
)
103 if ((flags
& RFPROC
) == 0) {
104 if ((flags
& RFMEM
) == 0) {
105 /* unshare user LDT */
106 struct pcb
*pcb1
= lp1
->lwp_thread
->td_pcb
;
107 struct pcb_ldt
*pcb_ldt
= pcb1
->pcb_ldt
;
108 if (pcb_ldt
&& pcb_ldt
->ldt_refcnt
> 1) {
109 pcb_ldt
= user_ldt_alloc(pcb1
,pcb_ldt
->ldt_len
);
111 pcb1
->pcb_ldt
= pcb_ldt
;
119 /* Ensure that lp1's pcb is up to date. */
120 if (mdcpu
->gd_npxthread
== lp1
->lwp_thread
)
121 npxsave(lp1
->lwp_thread
->td_savefpu
);
125 * Copy lp1's PCB. This really only applies to the
126 * debug registers and FP state, but its faster to just copy the
127 * whole thing. Because we only save the PCB at switchout time,
128 * the register state may not be current.
130 pcb2
= lp2
->lwp_thread
->td_pcb
;
131 *pcb2
= *lp1
->lwp_thread
->td_pcb
;
134 * Create a new fresh stack for the new process.
135 * Copy the trap frame for the return to user mode as if from a
136 * syscall. This copies the user mode register values. The
137 * 16 byte offset saves space for vm86, and must match
138 * common_tss.esp0 (kernel stack pointer on entry from user mode)
140 * pcb_esp must allocate an additional call-return pointer below
141 * the trap frame which will be restored by cpu_restore from
142 * PCB_EIP, and the thread's td_sp pointer must allocate an
143 * additonal two worsd below the pcb_esp call-return pointer to
144 * hold the LWKT restore function pointer and eflags.
146 * The LWKT restore function pointer must be set to cpu_restore,
147 * which is our standard heavy weight process switch-in function.
148 * YYY eventually we should shortcut fork_return and fork_trampoline
149 * to use the LWKT restore function directly so we can get rid of
150 * all the extra crap we are setting up.
152 lp2
->lwp_md
.md_regs
= (struct trapframe
*)((char *)pcb2
- 16) - 1;
153 bcopy(lp1
->lwp_md
.md_regs
, lp2
->lwp_md
.md_regs
, sizeof(*lp2
->lwp_md
.md_regs
));
156 * Set registers for trampoline to user mode. Leave space for the
157 * return address on stack. These are the kernel mode register values.
159 pcb2
->pcb_cr3
= vtophys(vmspace_pmap(lp2
->lwp_proc
->p_vmspace
)->pm_pdir
);
161 pcb2
->pcb_esi
= (int)fork_return
; /* fork_trampoline argument */
163 pcb2
->pcb_esp
= (int)lp2
->lwp_md
.md_regs
- sizeof(void *);
164 pcb2
->pcb_ebx
= (int)lp2
; /* fork_trampoline argument */
165 pcb2
->pcb_eip
= (int)fork_trampoline
;
166 lp2
->lwp_thread
->td_sp
= (char *)(pcb2
->pcb_esp
- sizeof(void *));
167 *(u_int32_t
*)lp2
->lwp_thread
->td_sp
= PSL_USER
;
168 lp2
->lwp_thread
->td_sp
-= sizeof(void *);
169 *(void **)lp2
->lwp_thread
->td_sp
= (void *)cpu_heavy_restore
;
172 * pcb2->pcb_ldt: duplicated below, if necessary.
173 * pcb2->pcb_savefpu: cloned above.
174 * pcb2->pcb_flags: cloned above (always 0 here?).
175 * pcb2->pcb_onfault: cloned above (always NULL here?).
179 * XXX don't copy the i/o pages. this should probably be fixed.
183 /* Copy the LDT, if necessary. */
184 if (pcb2
->pcb_ldt
!= 0) {
186 pcb2
->pcb_ldt
->ldt_refcnt
++;
188 pcb2
->pcb_ldt
= user_ldt_alloc(pcb2
,
189 pcb2
->pcb_ldt
->ldt_len
);
192 bcopy(&lp1
->lwp_thread
->td_tls
, &lp2
->lwp_thread
->td_tls
,
193 sizeof(lp2
->lwp_thread
->td_tls
));
195 * Now, cpu_switch() can schedule the new lwp.
196 * pcb_esp is loaded pointing to the cpu_switch() stack frame
197 * containing the return address when exiting cpu_switch.
198 * This will normally be to fork_trampoline(), which will have
199 * %ebx loaded with the new lwp's pointer. fork_trampoline()
200 * will set up a stack to call fork_return(lp, frame); to complete
201 * the return to user-mode.
206 * Prepare new lwp to return to the address specified in params.
209 cpu_prepare_lwp(struct lwp
*lp
, struct lwp_params
*params
)
211 struct trapframe
*regs
= lp
->lwp_md
.md_regs
;
212 void *bad_return
= NULL
;
215 regs
->tf_eip
= (int)params
->func
;
216 regs
->tf_esp
= (int)params
->stack
;
217 /* Set up argument for function call */
218 regs
->tf_esp
-= sizeof(params
->arg
);
220 copyout(¶ms
->arg
, (void *)regs
->tf_esp
, sizeof(params
->arg
));
224 * Set up fake return address. As the lwp function may never return,
225 * we simply copy out a NULL pointer and force the lwp to receive
226 * a SIGSEGV if it returns anyways.
228 regs
->tf_esp
-= sizeof(void *);
229 error
= copyout(&bad_return
, (void *)regs
->tf_esp
, sizeof(bad_return
));
233 cpu_set_fork_handler(lp
,
234 (void (*)(void *, struct trapframe
*))generic_lwp_return
, lp
);
239 * Intercept the return address from a freshly forked process that has NOT
240 * been scheduled yet.
242 * This is needed to make kernel threads stay in kernel mode.
245 cpu_set_fork_handler(struct lwp
*lp
, void (*func
)(void *, struct trapframe
*),
249 * Note that the trap frame follows the args, so the function
250 * is really called like this: func(arg, frame);
252 lp
->lwp_thread
->td_pcb
->pcb_esi
= (int) func
; /* function */
253 lp
->lwp_thread
->td_pcb
->pcb_ebx
= (int) arg
; /* first arg */
257 cpu_set_thread_handler(thread_t td
, void (*rfunc
)(void), void *func
, void *arg
)
259 td
->td_pcb
->pcb_esi
= (int)func
;
260 td
->td_pcb
->pcb_ebx
= (int) arg
;
261 td
->td_switch
= cpu_lwkt_switch
;
262 td
->td_sp
-= sizeof(void *);
263 *(void **)td
->td_sp
= rfunc
; /* exit function on return */
264 td
->td_sp
-= sizeof(void *);
265 *(void **)td
->td_sp
= cpu_kthread_restore
;
271 struct thread
*td
= curthread
;
280 * If we were using a private TSS do a forced-switch to ourselves
281 * to switch back to the common TSS before freeing it.
284 if ((ext
= pcb
->pcb_ext
) != NULL
) {
289 kmem_free(&kernel_map
, (vm_offset_t
)ext
, ctob(IOPAGES
+ 1));
292 if (pcb
->pcb_flags
& PCB_DBREGS
) {
294 * disable all hardware breakpoints
297 pcb
->pcb_flags
&= ~PCB_DBREGS
;
299 td
->td_gd
->gd_cnt
.v_swtch
++;
301 crit_enter_quick(td
);
302 lwkt_deschedule_self(td
);
303 lwkt_remove_tdallq(td
);
308 * Terminate the current thread. The caller must have already acquired
309 * the thread's rwlock and placed it on a reap list or otherwise notified
310 * a reaper of its existance. We set a special assembly switch function which
311 * releases td_rwlock after it has cleaned up the MMU state and switched
314 * Must be caller from a critical section and with the thread descheduled.
317 cpu_thread_exit(void)
319 curthread
->td_switch
= cpu_exit_switch
;
320 curthread
->td_flags
|= TDF_EXITING
;
326 * Process Reaper. Called after the caller has acquired the thread's
327 * rwlock and removed it from the reap list.
330 cpu_proc_wait(struct proc
*p
)
332 /* drop per-process resources */
333 pmap_dispose_proc(p
);
338 setredzone(u_short
*pte
, caddr_t vaddr
)
340 /* eventually do this by setting up an expand-down stack segment
341 for ss0: selector, allowing stack access down to top of u.
342 this means though that protection violations need to be handled
343 thru a double fault exception that must do an integral task
344 switch to a known good context, within which a dump can be
345 taken. a sensible scheme might be to save the initial context
346 used by sched (that has physical memory mapped 1:1 at bottom)
347 and take the dump while still in mapped mode */
352 * Convert kernel VA to physical address
359 pa
= pmap_kextract((vm_offset_t
)addr
);
361 panic("kvtop: zero page frame");
366 * Force reset the processor by invalidating the entire address space!
371 cpu_reset_proxy(void)
375 cpu_reset_proxy_active
= 1;
376 while (cpu_reset_proxy_active
== 1)
377 ; /* Wait for other cpu to disable interupts */
378 saved_mp_lock
= mp_lock
;
379 mp_lock
= 0; /* BSP */
380 kprintf("cpu_reset_proxy: Grabbed mp lock for BSP\n");
381 cpu_reset_proxy_active
= 3;
382 while (cpu_reset_proxy_active
== 3)
383 ; /* Wait for other cpu to enable interrupts */
384 stop_cpus((1<<cpu_reset_proxyid
));
385 kprintf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid
);
395 if (smp_active_mask
== 1) {
401 kprintf("cpu_reset called on cpu#%d\n",mycpu
->gd_cpuid
);
403 map
= mycpu
->gd_other_cpus
& ~stopped_cpus
& smp_active_mask
;
406 kprintf("cpu_reset: Stopping other CPUs\n");
407 stop_cpus(map
); /* Stop all other CPUs */
410 if (mycpu
->gd_cpuid
== 0) {
415 /* We are not BSP (CPU #0) */
417 cpu_reset_proxyid
= mycpu
->gd_cpuid
;
418 cpustop_restartfunc
= cpu_reset_proxy
;
419 kprintf("cpu_reset: Restarting BSP\n");
420 started_cpus
= (1<<0); /* Restart CPU #0 */
423 while (cpu_reset_proxy_active
== 0 && cnt
< 10000000)
424 cnt
++; /* Wait for BSP to announce restart */
425 if (cpu_reset_proxy_active
== 0)
426 kprintf("cpu_reset: Failed to restart BSP\n");
427 __asm
__volatile("cli" : : : "memory");
428 cpu_reset_proxy_active
= 2;
430 while (cpu_reset_proxy_active
== 2 && cnt
< 10000000)
431 cnt
++; /* Do nothing */
432 if (cpu_reset_proxy_active
== 2) {
433 kprintf("cpu_reset: BSP did not grab mp lock\n");
434 cpu_reset_real(); /* XXX: Bogus ? */
436 cpu_reset_proxy_active
= 4;
437 __asm
__volatile("sti" : : : "memory");
451 * Attempt to do a CPU reset via the keyboard controller,
452 * do not turn of the GateA20, as any machine that fails
453 * to do the reset here would then end up in no man's land.
456 #if !defined(BROKEN_KEYBOARD_RESET)
457 outb(IO_KBD
+ 4, 0xFE);
458 DELAY(500000); /* wait 0.5 sec to see if that did it */
459 kprintf("Keyboard reset did not work, attempting CPU shutdown\n");
460 DELAY(1000000); /* wait 1 sec for kprintf to complete */
462 /* force a shutdown by unmapping entire address space ! */
463 bzero((caddr_t
) PTD
, PAGE_SIZE
);
465 /* "good night, sweet prince .... <THUNK!>" */
472 grow_stack(struct proc
*p
, u_int sp
)
476 rv
= vm_map_growstack (p
, sp
);
477 if (rv
!= KERN_SUCCESS
)
483 SYSCTL_DECL(_vm_stats_misc
);
485 static int cnt_prezero
;
487 SYSCTL_INT(_vm_stats_misc
, OID_AUTO
,
488 cnt_prezero
, CTLFLAG_RD
, &cnt_prezero
, 0, "");
491 swi_vm(void *arg
, void *frame
)
493 if (busdma_swi_pending
!= 0)
498 swi_vm_setup(void *arg
)
500 register_swi(SWI_VM
, swi_vm
, NULL
, "swi_vm", NULL
);
503 SYSINIT(vm_setup
, SI_BOOT2_MACHDEP
, SI_ORDER_ANY
, swi_vm_setup
, NULL
);
507 * Tell whether this address is in some physical memory region.
508 * Currently used by the kernel coredump code in order to avoid
509 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
510 * or other unpredictable behaviour.
514 is_physical_memory(vm_offset_t addr
)
518 /* The ISA ``memory hole''. */
519 if (addr
>= 0xa0000 && addr
< 0x100000)
524 * stuff other tests for known memory-mapped devices (PCI?)
532 * platform-specific vmspace initialization (nothing for i386)
535 cpu_vmspace_alloc(struct vmspace
*vm __unused
)
540 cpu_vmspace_free(struct vmspace
*vm __unused
)
545 * Used by /dev/kmem to determine if we can safely read or write
546 * the requested KVA range.
549 kvm_access_check(vm_offset_t saddr
, vm_offset_t eaddr
, int prot
)
553 if (saddr
< KvaStart
)
557 for (addr
= saddr
; addr
< eaddr
; addr
+= PAGE_SIZE
) {
558 if (pmap_extract(&kernel_pmap
, addr
) == 0)
561 if (!kernacc((caddr_t
)saddr
, eaddr
- saddr
, prot
))