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 1.61 2008/08/02 01:14:43 dillon 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>
81 #include <sys/thread2.h>
82 #include <sys/mplock2.h>
84 #include <bus/isa/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 lwp lp2 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(struct lwp
*lp1
, struct lwp
*lp2
, int flags
)
105 if ((flags
& RFPROC
) == 0) {
106 if ((flags
& RFMEM
) == 0) {
107 /* unshare user LDT */
108 struct pcb
*pcb1
= lp1
->lwp_thread
->td_pcb
;
109 struct pcb_ldt
*pcb_ldt
= pcb1
->pcb_ldt
;
110 if (pcb_ldt
&& pcb_ldt
->ldt_refcnt
> 1) {
111 pcb_ldt
= user_ldt_alloc(pcb1
,pcb_ldt
->ldt_len
);
113 pcb1
->pcb_ldt
= pcb_ldt
;
121 /* Ensure that lp1's pcb is up to date. */
122 if (mdcpu
->gd_npxthread
== lp1
->lwp_thread
)
123 npxsave(lp1
->lwp_thread
->td_savefpu
);
127 * Copy lp1's PCB. This really only applies to the
128 * debug registers and FP state, but its faster to just copy the
129 * whole thing. Because we only save the PCB at switchout time,
130 * the register state may not be current.
132 pcb2
= lp2
->lwp_thread
->td_pcb
;
133 *pcb2
= *lp1
->lwp_thread
->td_pcb
;
136 * Create a new fresh stack for the new process.
137 * Copy the trap frame for the return to user mode as if from a
138 * syscall. This copies the user mode register values. The
139 * 16 byte offset saves space for vm86, and must match
140 * common_tss.esp0 (kernel stack pointer on entry from user mode)
142 * pcb_esp must allocate an additional call-return pointer below
143 * the trap frame which will be restored by cpu_restore from
144 * PCB_EIP, and the thread's td_sp pointer must allocate an
145 * additonal two worsd below the pcb_esp call-return pointer to
146 * hold the LWKT restore function pointer and eflags.
148 * The LWKT restore function pointer must be set to cpu_restore,
149 * which is our standard heavy weight process switch-in function.
150 * YYY eventually we should shortcut fork_return and fork_trampoline
151 * to use the LWKT restore function directly so we can get rid of
152 * all the extra crap we are setting up.
154 lp2
->lwp_md
.md_regs
= (struct trapframe
*)((char *)pcb2
- 16) - 1;
155 bcopy(lp1
->lwp_md
.md_regs
, lp2
->lwp_md
.md_regs
, sizeof(*lp2
->lwp_md
.md_regs
));
158 * Set registers for trampoline to user mode. Leave space for the
159 * return address on stack. These are the kernel mode register values.
161 pcb2
->pcb_cr3
= vtophys(vmspace_pmap(lp2
->lwp_proc
->p_vmspace
)->pm_pdir
);
163 pcb2
->pcb_esi
= (int)fork_return
; /* fork_trampoline argument */
165 pcb2
->pcb_esp
= (int)lp2
->lwp_md
.md_regs
- sizeof(void *);
166 pcb2
->pcb_ebx
= (int)lp2
; /* fork_trampoline argument */
167 pcb2
->pcb_eip
= (int)fork_trampoline
;
168 lp2
->lwp_thread
->td_sp
= (char *)(pcb2
->pcb_esp
- sizeof(void *));
169 *(u_int32_t
*)lp2
->lwp_thread
->td_sp
= PSL_USER
;
170 lp2
->lwp_thread
->td_sp
-= sizeof(void *);
171 *(void **)lp2
->lwp_thread
->td_sp
= (void *)cpu_heavy_restore
;
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 * XXX don't copy the i/o pages. this should probably be fixed.
185 /* Copy the LDT, if necessary. */
186 if (pcb2
->pcb_ldt
!= 0) {
188 pcb2
->pcb_ldt
->ldt_refcnt
++;
190 pcb2
->pcb_ldt
= user_ldt_alloc(pcb2
,
191 pcb2
->pcb_ldt
->ldt_len
);
194 bcopy(&lp1
->lwp_thread
->td_tls
, &lp2
->lwp_thread
->td_tls
,
195 sizeof(lp2
->lwp_thread
->td_tls
));
197 * Now, cpu_switch() can schedule the new lwp.
198 * pcb_esp is loaded pointing to the cpu_switch() stack frame
199 * containing the return address when exiting cpu_switch.
200 * This will normally be to fork_trampoline(), which will have
201 * %ebx loaded with the new lwp's pointer. fork_trampoline()
202 * will set up a stack to call fork_return(lp, frame); to complete
203 * the return to user-mode.
208 * Prepare new lwp to return to the address specified in params.
211 cpu_prepare_lwp(struct lwp
*lp
, struct lwp_params
*params
)
213 struct trapframe
*regs
= lp
->lwp_md
.md_regs
;
214 void *bad_return
= NULL
;
217 regs
->tf_eip
= (int)params
->func
;
218 regs
->tf_esp
= (int)params
->stack
;
219 /* Set up argument for function call */
220 regs
->tf_esp
-= sizeof(params
->arg
);
222 copyout(¶ms
->arg
, (void *)regs
->tf_esp
, sizeof(params
->arg
));
226 * Set up fake return address. As the lwp function may never return,
227 * we simply copy out a NULL pointer and force the lwp to receive
228 * a SIGSEGV if it returns anyways.
230 regs
->tf_esp
-= sizeof(void *);
231 error
= copyout(&bad_return
, (void *)regs
->tf_esp
, sizeof(bad_return
));
235 cpu_set_fork_handler(lp
,
236 (void (*)(void *, struct trapframe
*))generic_lwp_return
, lp
);
241 * Intercept the return address from a freshly forked process that has NOT
242 * been scheduled yet.
244 * This is needed to make kernel threads stay in kernel mode.
247 cpu_set_fork_handler(struct lwp
*lp
, void (*func
)(void *, struct trapframe
*),
251 * Note that the trap frame follows the args, so the function
252 * is really called like this: func(arg, frame);
254 lp
->lwp_thread
->td_pcb
->pcb_esi
= (int) func
; /* function */
255 lp
->lwp_thread
->td_pcb
->pcb_ebx
= (int) arg
; /* first arg */
259 cpu_set_thread_handler(thread_t td
, void (*rfunc
)(void), void *func
, void *arg
)
261 td
->td_pcb
->pcb_esi
= (int)func
;
262 td
->td_pcb
->pcb_ebx
= (int) arg
;
263 td
->td_switch
= cpu_lwkt_switch
;
264 td
->td_sp
-= sizeof(void *);
265 *(void **)td
->td_sp
= rfunc
; /* exit function on return */
266 td
->td_sp
-= sizeof(void *);
267 *(void **)td
->td_sp
= cpu_kthread_restore
;
273 struct thread
*td
= curthread
;
282 * If we were using a private TSS do a forced-switch to ourselves
283 * to switch back to the common TSS before freeing it.
286 if ((ext
= pcb
->pcb_ext
) != NULL
) {
291 kmem_free(&kernel_map
, (vm_offset_t
)ext
, ctob(IOPAGES
+ 1));
294 if (pcb
->pcb_flags
& PCB_DBREGS
) {
296 * disable all hardware breakpoints
299 pcb
->pcb_flags
&= ~PCB_DBREGS
;
301 td
->td_gd
->gd_cnt
.v_swtch
++;
303 crit_enter_quick(td
);
304 if (td
->td_flags
& TDF_TSLEEPQ
)
306 lwkt_deschedule_self(td
);
307 lwkt_remove_tdallq(td
);
312 * Terminate the current thread. The caller must have already acquired
313 * the thread's rwlock and placed it on a reap list or otherwise notified
314 * a reaper of its existance. We set a special assembly switch function which
315 * releases td_rwlock after it has cleaned up the MMU state and switched
318 * Must be caller from a critical section and with the thread descheduled.
321 cpu_thread_exit(void)
323 curthread
->td_switch
= cpu_exit_switch
;
324 curthread
->td_flags
|= TDF_EXITING
;
330 * Process Reaper. Called after the caller has acquired the thread's
331 * rwlock and removed it from the reap list.
334 cpu_proc_wait(struct proc
*p
)
336 /* drop per-process resources */
337 pmap_dispose_proc(p
);
342 setredzone(u_short
*pte
, caddr_t vaddr
)
344 /* eventually do this by setting up an expand-down stack segment
345 for ss0: selector, allowing stack access down to top of u.
346 this means though that protection violations need to be handled
347 thru a double fault exception that must do an integral task
348 switch to a known good context, within which a dump can be
349 taken. a sensible scheme might be to save the initial context
350 used by sched (that has physical memory mapped 1:1 at bottom)
351 and take the dump while still in mapped mode */
356 * Convert kernel VA to physical address
363 pa
= pmap_kextract((vm_offset_t
)addr
);
365 panic("kvtop: zero page frame");
370 * Force reset the processor by invalidating the entire address space!
375 cpu_reset_proxy(void)
379 cpu_reset_proxy_active
= 1;
380 while (cpu_reset_proxy_active
== 1)
381 ; /* Wait for other cpu to disable interupts */
382 saved_mp_lock
= mp_lock
;
383 mp_lock
= 0; /* BSP */
384 kprintf("cpu_reset_proxy: Grabbed mp lock for BSP\n");
385 cpu_reset_proxy_active
= 3;
386 while (cpu_reset_proxy_active
== 3)
387 ; /* Wait for other cpu to enable interrupts */
388 stop_cpus((1<<cpu_reset_proxyid
));
389 kprintf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid
);
399 if (smp_active_mask
== 1) {
405 kprintf("cpu_reset called on cpu#%d\n",mycpu
->gd_cpuid
);
407 map
= mycpu
->gd_other_cpus
& ~stopped_cpus
& smp_active_mask
;
410 kprintf("cpu_reset: Stopping other CPUs\n");
411 stop_cpus(map
); /* Stop all other CPUs */
414 if (mycpu
->gd_cpuid
== 0) {
419 /* We are not BSP (CPU #0) */
421 cpu_reset_proxyid
= mycpu
->gd_cpuid
;
422 cpustop_restartfunc
= cpu_reset_proxy
;
423 kprintf("cpu_reset: Restarting BSP\n");
424 started_cpus
= (1<<0); /* Restart CPU #0 */
427 while (cpu_reset_proxy_active
== 0 && cnt
< 10000000)
428 cnt
++; /* Wait for BSP to announce restart */
429 if (cpu_reset_proxy_active
== 0)
430 kprintf("cpu_reset: Failed to restart BSP\n");
431 __asm
__volatile("cli" : : : "memory");
432 cpu_reset_proxy_active
= 2;
434 while (cpu_reset_proxy_active
== 2 && cnt
< 10000000)
435 cnt
++; /* Do nothing */
436 if (cpu_reset_proxy_active
== 2) {
437 kprintf("cpu_reset: BSP did not grab mp lock\n");
438 cpu_reset_real(); /* XXX: Bogus ? */
440 cpu_reset_proxy_active
= 4;
441 __asm
__volatile("sti" : : : "memory");
455 * Attempt to do a CPU reset via the keyboard controller,
456 * do not turn of the GateA20, as any machine that fails
457 * to do the reset here would then end up in no man's land.
460 #if !defined(BROKEN_KEYBOARD_RESET)
461 outb(IO_KBD
+ 4, 0xFE);
462 DELAY(500000); /* wait 0.5 sec to see if that did it */
463 kprintf("Keyboard reset did not work, attempting CPU shutdown\n");
464 DELAY(1000000); /* wait 1 sec for kprintf to complete */
466 /* force a shutdown by unmapping entire address space ! */
467 bzero((caddr_t
) PTD
, PAGE_SIZE
);
469 /* "good night, sweet prince .... <THUNK!>" */
476 grow_stack(struct proc
*p
, vm_offset_t sp
)
480 rv
= vm_map_growstack (p
, sp
);
481 if (rv
!= KERN_SUCCESS
)
487 SYSCTL_DECL(_vm_stats_misc
);
489 static int cnt_prezero
;
491 SYSCTL_INT(_vm_stats_misc
, OID_AUTO
,
492 cnt_prezero
, CTLFLAG_RD
, &cnt_prezero
, 0, "");
495 swi_vm(void *arg
, void *frame
)
497 if (busdma_swi_pending
!= 0)
502 swi_vm_setup(void *arg
)
504 register_swi(SWI_VM
, swi_vm
, NULL
, "swi_vm", NULL
);
507 SYSINIT(vm_setup
, SI_BOOT2_MACHDEP
, SI_ORDER_ANY
, swi_vm_setup
, NULL
);
511 * Tell whether this address is in some physical memory region.
512 * Currently used by the kernel coredump code in order to avoid
513 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
514 * or other unpredictable behaviour.
518 is_physical_memory(vm_offset_t addr
)
522 /* The ISA ``memory hole''. */
523 if (addr
>= 0xa0000 && addr
< 0x100000)
528 * stuff other tests for known memory-mapped devices (PCI?)
536 * platform-specific vmspace initialization (nothing for i386)
539 cpu_vmspace_alloc(struct vmspace
*vm __unused
)
544 cpu_vmspace_free(struct vmspace
*vm __unused
)
549 * Used by /dev/kmem to determine if we can safely read or write
550 * the requested KVA range.
553 kvm_access_check(vm_offset_t saddr
, vm_offset_t eaddr
, int prot
)
557 if (saddr
< KvaStart
)
561 for (addr
= saddr
; addr
< eaddr
; addr
+= PAGE_SIZE
) {
562 if (pmap_extract(&kernel_pmap
, addr
) == 0)
565 if (!kernacc((caddr_t
)saddr
, eaddr
- saddr
, prot
))