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
5 * Copyright (c) 2008 The DragonFly Project.
8 * This code is derived from software contributed to Berkeley by
9 * the Systems Programming Group of the University of Utah Computer
10 * Science Department, and William Jolitz.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. All advertising materials mentioning features or use of this software
21 * must display the following acknowledgement:
22 * This product includes software developed by the University of
23 * California, Berkeley and its contributors.
24 * 4. Neither the name of the University nor the names of its contributors
25 * may be used to endorse or promote products derived from this software
26 * without specific prior written permission.
28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40 * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91
41 * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
42 * $FreeBSD: src/sys/i386/i386/vm_machdep.c,v 1.132.2.9 2003/01/25 19:02:23 dillon Exp $
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/malloc.h>
50 #include <sys/interrupt.h>
51 #include <sys/vnode.h>
52 #include <sys/vmmeter.h>
53 #include <sys/kernel.h>
54 #include <sys/sysctl.h>
55 #include <sys/unistd.h>
57 #include <machine/clock.h>
58 #include <machine/cpu.h>
59 #include <machine/md_var.h>
60 #include <machine/smp.h>
61 #include <machine/pcb.h>
62 #include <machine/pcb_ext.h>
63 #include <machine/segments.h>
64 #include <machine/globaldata.h> /* npxthread */
67 #include <vm/vm_param.h>
69 #include <vm/vm_kern.h>
70 #include <vm/vm_page.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_extern.h>
74 #include <sys/thread2.h>
75 #include <sys/mplock2.h>
77 #include <bus/isa/isa.h>
79 static void cpu_reset_real (void);
81 * Finish a fork operation, with lwp lp2 nearly set up.
82 * Copy and update the pcb, set up the stack so that the child
83 * ready to run and return to user mode.
86 cpu_fork(struct lwp *lp1, struct lwp *lp2, int flags)
90 if ((flags & RFPROC) == 0) {
91 if ((flags & RFMEM) == 0) {
92 /* unshare user LDT */
93 struct pcb *pcb1 = lp1->lwp_thread->td_pcb;
94 struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt;
95 if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) {
96 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len);
98 pcb1->pcb_ldt = pcb_ldt;
105 /* Ensure that lp1's pcb is up to date. */
106 if (mdcpu->gd_npxthread == lp1->lwp_thread)
107 npxsave(lp1->lwp_thread->td_savefpu);
110 * Copy lp1's PCB. This really only applies to the
111 * debug registers and FP state, but its faster to just copy the
112 * whole thing. Because we only save the PCB at switchout time,
113 * the register state may not be current.
115 pcb2 = lp2->lwp_thread->td_pcb;
116 *pcb2 = *lp1->lwp_thread->td_pcb;
119 * Create a new fresh stack for the new process.
120 * Copy the trap frame for the return to user mode as if from a
121 * syscall. This copies the user mode register values.
123 * pcb_rsp must allocate an additional call-return pointer below
124 * the trap frame which will be restored by cpu_heavy_restore from
125 * PCB_RIP, and the thread's td_sp pointer must allocate an
126 * additonal two quadwords below the pcb_rsp call-return pointer to
127 * hold the LWKT restore function pointer and rflags.
129 * The LWKT restore function pointer must be set to cpu_heavy_restore,
130 * which is our standard heavy-weight process switch-in function.
131 * YYY eventually we should shortcut fork_return and fork_trampoline
132 * to use the LWKT restore function directly so we can get rid of
133 * all the extra crap we are setting up.
135 lp2->lwp_md.md_regs = (struct trapframe *)pcb2 - 1;
136 bcopy(lp1->lwp_md.md_regs, lp2->lwp_md.md_regs, sizeof(*lp2->lwp_md.md_regs));
139 * Set registers for trampoline to user mode. Leave space for the
140 * return address on stack. These are the kernel mode register values.
142 pcb2->pcb_cr3 = vtophys(vmspace_pmap(lp2->lwp_proc->p_vmspace)->pm_pml4);
143 pcb2->pcb_cr3 |= PG_RW | PG_U | PG_V;
144 pcb2->pcb_rbx = (unsigned long)fork_return; /* fork_trampoline argument */
146 pcb2->pcb_rsp = (unsigned long)lp2->lwp_md.md_regs - sizeof(void *);
147 pcb2->pcb_r12 = (unsigned long)lp2; /* fork_trampoline argument */
151 pcb2->pcb_rip = (unsigned long)fork_trampoline;
152 lp2->lwp_thread->td_sp = (char *)(pcb2->pcb_rsp - sizeof(void *));
153 *(u_int64_t *)lp2->lwp_thread->td_sp = PSL_USER;
154 lp2->lwp_thread->td_sp -= sizeof(void *);
155 *(void **)lp2->lwp_thread->td_sp = (void *)cpu_heavy_restore;
158 * pcb2->pcb_ldt: duplicated below, if necessary.
159 * pcb2->pcb_savefpu: cloned above.
160 * pcb2->pcb_flags: cloned above (always 0 here?).
161 * pcb2->pcb_onfault: cloned above (always NULL here).
162 * pcb2->pcb_onfault_sp:cloned above (dont care)
166 * XXX don't copy the i/o pages. this should probably be fixed.
168 pcb2->pcb_ext = NULL;
170 /* Copy the LDT, if necessary. */
171 if (pcb2->pcb_ldt != NULL) {
173 pcb2->pcb_ldt->ldt_refcnt++;
175 pcb2->pcb_ldt = user_ldt_alloc(pcb2,
176 pcb2->pcb_ldt->ldt_len);
179 bcopy(&lp1->lwp_thread->td_tls, &lp2->lwp_thread->td_tls,
180 sizeof(lp2->lwp_thread->td_tls));
182 * Now, cpu_switch() can schedule the new lwp.
183 * pcb_rsp is loaded pointing to the cpu_switch() stack frame
184 * containing the return address when exiting cpu_switch.
185 * This will normally be to fork_trampoline(), which will have
186 * %rbx loaded with the new lwp's pointer. fork_trampoline()
187 * will set up a stack to call fork_return(lp, frame); to complete
188 * the return to user-mode.
193 * Prepare new lwp to return to the address specified in params.
196 cpu_prepare_lwp(struct lwp *lp, struct lwp_params *params)
198 struct trapframe *regs = lp->lwp_md.md_regs;
199 void *bad_return = NULL;
202 regs->tf_rip = (long)params->func;
203 regs->tf_rsp = (long)params->stack;
204 /* Set up argument for function call */
205 regs->tf_rdi = (long)params->arg;
208 * Set up fake return address. As the lwp function may never return,
209 * we simply copy out a NULL pointer and force the lwp to receive
210 * a SIGSEGV if it returns anyways.
212 regs->tf_rsp -= sizeof(void *);
213 error = copyout(&bad_return, (void *)regs->tf_rsp, sizeof(bad_return));
217 cpu_set_fork_handler(lp,
218 (void (*)(void *, struct trapframe *))generic_lwp_return, lp);
223 * Intercept the return address from a freshly forked process that has NOT
224 * been scheduled yet.
226 * This is needed to make kernel threads stay in kernel mode.
229 cpu_set_fork_handler(struct lwp *lp, void (*func)(void *, struct trapframe *),
233 * Note that the trap frame follows the args, so the function
234 * is really called like this: func(arg, frame);
236 lp->lwp_thread->td_pcb->pcb_rbx = (long)func; /* function */
237 lp->lwp_thread->td_pcb->pcb_r12 = (long)arg; /* first arg */
241 cpu_set_thread_handler(thread_t td, void (*rfunc)(void), void *func, void *arg)
243 td->td_pcb->pcb_rbx = (long)func;
244 td->td_pcb->pcb_r12 = (long)arg;
245 td->td_switch = cpu_lwkt_switch;
246 td->td_sp -= sizeof(void *);
247 *(void **)td->td_sp = rfunc; /* exit function on return */
248 td->td_sp -= sizeof(void *);
249 *(void **)td->td_sp = cpu_kthread_restore;
255 struct thread *td = curthread;
260 /* Some i386 functionality was dropped */
261 KKASSERT(pcb->pcb_ext == NULL);
264 * disable all hardware breakpoints
266 if (pcb->pcb_flags & PCB_DBREGS) {
268 pcb->pcb_flags &= ~PCB_DBREGS;
270 td->td_gd->gd_cnt.v_swtch++;
272 crit_enter_quick(td);
273 if (td->td_flags & TDF_TSLEEPQ)
275 lwkt_deschedule_self(td);
276 lwkt_remove_tdallq(td);
281 * Terminate the current thread. The caller must have already acquired
282 * the thread's rwlock and placed it on a reap list or otherwise notified
283 * a reaper of its existance. We set a special assembly switch function which
284 * releases td_rwlock after it has cleaned up the MMU state and switched
287 * Must be caller from a critical section and with the thread descheduled.
290 cpu_thread_exit(void)
293 curthread->td_switch = cpu_exit_switch;
294 curthread->td_flags |= TDF_EXITING;
296 panic("cpu_thread_exit: lwkt_switch() unexpectedly returned");
309 * Attempt to do a CPU reset via the keyboard controller,
310 * do not turn of the GateA20, as any machine that fails
311 * to do the reset here would then end up in no man's land.
314 #if !defined(BROKEN_KEYBOARD_RESET)
315 outb(IO_KBD + 4, 0xFE);
316 DELAY(500000); /* wait 0.5 sec to see if that did it */
317 kprintf("Keyboard reset did not work, attempting CPU shutdown\n");
318 DELAY(1000000); /* wait 1 sec for kprintf to complete */
321 /* force a shutdown by unmapping entire address space ! */
322 bzero((caddr_t) PTD, PAGE_SIZE);
325 /* "good night, sweet prince .... <THUNK!>" */
332 * Convert kernel VA to physical address
339 pa = pmap_kextract((vm_offset_t)addr);
341 panic("kvtop: zero page frame");
346 swi_vm(void *arg, void *frame)
348 if (busdma_swi_pending != 0)
353 swi_vm_setup(void *arg)
355 register_swi(SWI_VM, swi_vm, NULL, "swi_vm", NULL, 0);
358 SYSINIT(vm_setup, SI_BOOT2_MACHDEP, SI_ORDER_ANY, swi_vm_setup, NULL);
361 * platform-specific vmspace initialization (nothing for x86_64)
364 cpu_vmspace_alloc(struct vmspace *vm __unused)
369 cpu_vmspace_free(struct vmspace *vm __unused)
374 kvm_access_check(vm_offset_t saddr, vm_offset_t eaddr, int prot)
378 if (saddr < KvaStart)
382 for (addr = saddr; addr < eaddr; addr += PAGE_SIZE) {
383 if (pmap_extract(&kernel_pmap, addr) == 0)
386 if (!kernacc((caddr_t)saddr, eaddr - saddr, prot))
393 void _test_frame_enter(struct trapframe *frame);
394 void _test_frame_exit(struct trapframe *frame);
397 _test_frame_enter(struct trapframe *frame)
399 thread_t td = curthread;
401 if (ISPL(frame->tf_cs) == SEL_UPL) {
402 KKASSERT(td->td_lwp);
403 KASSERT(td->td_lwp->lwp_md.md_regs == frame,
404 ("_test_frame_exit: Frame mismatch %p %p",
405 td->td_lwp->lwp_md.md_regs, frame));
406 td->td_lwp->lwp_saveusp = (void *)frame->tf_rsp;
407 td->td_lwp->lwp_saveupc = (void *)frame->tf_rip;
409 if ((char *)frame < td->td_kstack ||
410 (char *)frame > td->td_kstack + td->td_kstack_size) {
411 panic("_test_frame_exit: frame not on kstack %p kstack=%p",
412 frame, td->td_kstack);
417 _test_frame_exit(struct trapframe *frame)
419 thread_t td = curthread;
421 if (ISPL(frame->tf_cs) == SEL_UPL) {
422 KKASSERT(td->td_lwp);
423 KASSERT(td->td_lwp->lwp_md.md_regs == frame,
424 ("_test_frame_exit: Frame mismatch %p %p",
425 td->td_lwp->lwp_md.md_regs, frame));
426 if (td->td_lwp->lwp_saveusp != (void *)frame->tf_rsp) {
427 kprintf("_test_frame_exit: %s:%d usp mismatch %p/%p\n",
428 td->td_comm, td->td_proc->p_pid,
429 td->td_lwp->lwp_saveusp,
430 (void *)frame->tf_rsp);
432 if (td->td_lwp->lwp_saveupc != (void *)frame->tf_rip) {
433 kprintf("_test_frame_exit: %s:%d upc mismatch %p/%p\n",
434 td->td_comm, td->td_proc->p_pid,
435 td->td_lwp->lwp_saveupc,
436 (void *)frame->tf_rip);
440 * adulterate the fields to catch entries that
441 * don't run through test_frame_enter
443 td->td_lwp->lwp_saveusp =
444 (void *)~(intptr_t)td->td_lwp->lwp_saveusp;
445 td->td_lwp->lwp_saveupc =
446 (void *)~(intptr_t)td->td_lwp->lwp_saveupc;
448 if ((char *)frame < td->td_kstack ||
449 (char *)frame > td->td_kstack + td->td_kstack_size) {
450 panic("_test_frame_exit: frame not on kstack %p kstack=%p",
451 frame, td->td_kstack);