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 $
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/malloc.h>
51 #include <sys/interrupt.h>
52 #include <sys/vnode.h>
53 #include <sys/vmmeter.h>
54 #include <sys/kernel.h>
55 #include <sys/sysctl.h>
56 #include <sys/unistd.h>
58 #include <machine/clock.h>
59 #include <machine/cpu.h>
60 #include <machine/md_var.h>
61 #include <machine/smp.h>
62 #include <machine/pcb.h>
63 #include <machine/pcb_ext.h>
64 #include <machine/vm86.h>
65 #include <machine/segments.h>
66 #include <machine/globaldata.h> /* npxthread */
69 #include <vm/vm_param.h>
71 #include <vm/vm_kern.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_map.h>
74 #include <vm/vm_extern.h>
78 #include <sys/thread2.h>
79 #include <sys/mplock2.h>
81 #include <bus/isa/isa.h>
86 char machine[] = MACHINE;
87 SYSCTL_STRING(_hw, HW_MACHINE, machine, CTLFLAG_RD,
88 machine, 0, "Machine class");
90 char cpu_vendor[] = "DragonFly"; /* XXX */
91 u_int cpu_vendor_id = 0; /* XXX */
92 u_int cpu_id = 0x80000000; /* XXX */
95 * Finish a fork operation, with lwp lp2 nearly set up.
96 * Copy and update the pcb, set up the stack so that the child
97 * ready to run and return to user mode.
100 cpu_fork(struct lwp *lp1, struct lwp *lp2, int flags)
104 if ((flags & RFPROC) == 0) {
105 if ((flags & RFMEM) == 0) {
106 /* unshare user LDT */
107 struct pcb *pcb1 = lp1->lwp_thread->td_pcb;
108 struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt;
109 if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) {
110 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len);
112 pcb1->pcb_ldt = pcb_ldt;
120 /* Ensure that lp1's pcb is up to date. */
121 if (mdcpu->gd_npxthread == lp1->lwp_thread)
122 npxsave(lp1->lwp_thread->td_savefpu);
126 * Copy lp1's PCB. This really only applies to the
127 * debug registers and FP state, but its faster to just copy the
128 * whole thing. Because we only save the PCB at switchout time,
129 * the register state may not be current.
131 pcb2 = lp2->lwp_thread->td_pcb;
132 *pcb2 = *lp1->lwp_thread->td_pcb;
135 * Create a new fresh stack for the new process.
136 * Copy the trap frame for the return to user mode as if from a
137 * syscall. This copies the user mode register values. The
138 * 16 byte offset saves space for vm86, and must match
139 * common_tss.esp0 (kernel stack pointer on entry from user mode)
141 * pcb_esp must allocate an additional call-return pointer below
142 * the trap frame which will be restored by cpu_restore from
143 * PCB_EIP, and the thread's td_sp pointer must allocate an
144 * additonal two worsd below the pcb_esp call-return pointer to
145 * hold the LWKT restore function pointer and eflags.
147 * The LWKT restore function pointer must be set to cpu_restore,
148 * which is our standard heavy weight process switch-in function.
149 * YYY eventually we should shortcut fork_return and fork_trampoline
150 * to use the LWKT restore function directly so we can get rid of
151 * all the extra crap we are setting up.
153 lp2->lwp_md.md_regs = (struct trapframe *)((char *)pcb2 - 16) - 1;
154 bcopy(lp1->lwp_md.md_regs, lp2->lwp_md.md_regs, sizeof(*lp2->lwp_md.md_regs));
157 * Set registers for trampoline to user mode. Leave space for the
158 * return address on stack. These are the kernel mode register values.
160 pcb2->pcb_unused01 = 0;
162 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */
164 pcb2->pcb_esp = (int)lp2->lwp_md.md_regs - sizeof(void *);
165 pcb2->pcb_ebx = (int)lp2; /* fork_trampoline argument */
166 pcb2->pcb_eip = (int)fork_trampoline;
167 lp2->lwp_thread->td_sp = (char *)(pcb2->pcb_esp - sizeof(void *));
168 *(u_int32_t *)lp2->lwp_thread->td_sp = PSL_USER;
169 lp2->lwp_thread->td_sp -= sizeof(void *);
170 *(void **)lp2->lwp_thread->td_sp = (void *)cpu_heavy_restore;
173 * pcb2->pcb_ldt: duplicated below, if necessary.
174 * pcb2->pcb_savefpu: cloned above.
175 * pcb2->pcb_flags: cloned above (always 0 here?).
176 * pcb2->pcb_onfault: cloned above (always NULL here?).
180 * XXX don't copy the i/o pages. this should probably be fixed.
184 /* Copy the LDT, if necessary. */
185 if (pcb2->pcb_ldt != 0) {
187 pcb2->pcb_ldt->ldt_refcnt++;
189 pcb2->pcb_ldt = user_ldt_alloc(pcb2,
190 pcb2->pcb_ldt->ldt_len);
193 bcopy(&lp1->lwp_thread->td_tls, &lp2->lwp_thread->td_tls,
194 sizeof(lp2->lwp_thread->td_tls));
196 * Now, cpu_switch() can schedule the new process.
197 * pcb_esp is loaded pointing to the cpu_switch() stack frame
198 * containing the return address when exiting cpu_switch.
199 * This will normally be to fork_trampoline(), which will have
200 * %ebx loaded with the new proc's pointer. fork_trampoline()
201 * will set up a stack to call fork_return(p, frame); to complete
202 * the return to user-mode.
207 * Prepare new lwp to return to the address specified in params.
210 cpu_prepare_lwp(struct lwp *lp, struct lwp_params *params)
212 struct trapframe *regs = lp->lwp_md.md_regs;
213 void *bad_return = NULL;
216 regs->tf_eip = (int)params->func;
217 regs->tf_esp = (int)params->stack;
218 /* Set up argument for function call */
219 regs->tf_esp -= sizeof(params->arg);
220 error = copyout(¶ms->arg, (void *)regs->tf_esp,
221 sizeof(params->arg));
225 * Set up fake return address. As the lwp function may never return,
226 * we simply copy out a NULL pointer and force the lwp to receive
227 * a SIGSEGV if it returns anyways.
229 regs->tf_esp -= sizeof(void *);
230 error = copyout(&bad_return, (void *)regs->tf_esp,
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");
369 SYSCTL_DECL(_vm_stats_misc);
372 * Tell whether this address is in some physical memory region.
373 * Currently used by the kernel coredump code in order to avoid
374 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
375 * or other unpredictable behaviour.
379 is_physical_memory(vm_offset_t addr)
385 * Used by /dev/kmem to determine if we can safely read or write
386 * the requested KVA range. Some portions of kernel memory are
387 * not governed by our virtual page table.
390 extern void _start(void);
393 kvm_access_check(vm_offset_t saddr, vm_offset_t eaddr, int prot)
397 if (saddr >= trunc_page((vm_offset_t)&_start) && eaddr <= round_page((vm_offset_t)&_end))
399 if (saddr < KvaStart)
403 for (addr = saddr; addr < eaddr; addr += PAGE_SIZE) {
404 if (pmap_extract(&kernel_pmap, addr) == 0)
407 if (!kernacc((caddr_t)saddr, eaddr - saddr, prot))