dsched - Periph.: call dsched_exit on thread exit
[dragonfly.git] / sys / platform / vkernel / i386 / vm_machdep.c
CommitLineData
6f7b98e0
MD
1/*-
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 * All rights reserved.
6 *
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.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
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.
26 *
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
37 * SUCH DAMAGE.
38 *
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 $
d7510ae6 42 * $DragonFly: src/sys/platform/vkernel/i386/vm_machdep.c,v 1.10 2008/08/02 05:22:21 dillon Exp $
6f7b98e0
MD
43 */
44
45#include "use_npx.h"
46#include "use_isa.h"
47
48#include <sys/param.h>
49#include <sys/systm.h>
50#include <sys/malloc.h>
51#include <sys/proc.h>
52#include <sys/buf.h>
53#include <sys/interrupt.h>
54#include <sys/vnode.h>
55#include <sys/vmmeter.h>
56#include <sys/kernel.h>
57#include <sys/sysctl.h>
58#include <sys/unistd.h>
f8abf63c 59#include <sys/dsched.h>
6f7b98e0
MD
60
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 */
70
71#include <vm/vm.h>
72#include <vm/vm_param.h>
73#include <sys/lock.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>
78
79#include <sys/user.h>
684a93c4 80
6f7b98e0 81#include <sys/thread2.h>
684a93c4 82#include <sys/mplock2.h>
6f7b98e0 83
d7510ae6 84#include <bus/isa/isa.h>
6f7b98e0
MD
85
86#include <stdio.h>
87#include <stdlib.h>
88
0955fd91 89char machine[] = MACHINE;
6f7b98e0
MD
90SYSCTL_STRING(_hw, HW_MACHINE, machine, CTLFLAG_RD,
91 machine, 0, "Machine class");
92
9ad680a3
MD
93char cpu_vendor[] = "DragonFly"; /* XXX */
94u_int cpu_id = 0x80000000; /* XXX */
95
6f7b98e0
MD
96/*
97 * Finish a fork operation, with lwp lp2 nearly set up.
98 * Copy and update the pcb, set up the stack so that the child
99 * ready to run and return to user mode.
100 */
101void
102cpu_fork(struct lwp *lp1, struct lwp *lp2, int flags)
103{
104 struct pcb *pcb2;
105
106 if ((flags & RFPROC) == 0) {
107 if ((flags & RFMEM) == 0) {
108 /* unshare user LDT */
109 struct pcb *pcb1 = lp1->lwp_thread->td_pcb;
110 struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt;
111 if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) {
112 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len);
113 user_ldt_free(pcb1);
114 pcb1->pcb_ldt = pcb_ldt;
115 set_user_ldt(pcb1);
116 }
117 }
118 return;
119 }
120
121#if NNPX > 0
122 /* Ensure that lp1's pcb is up to date. */
123 if (mdcpu->gd_npxthread == lp1->lwp_thread)
124 npxsave(lp1->lwp_thread->td_savefpu);
125#endif
126
127 /*
128 * Copy lp1's PCB. This really only applies to the
129 * debug registers and FP state, but its faster to just copy the
130 * whole thing. Because we only save the PCB at switchout time,
4e7c41c5 131 * the register state may not be current.
6f7b98e0
MD
132 */
133 pcb2 = lp2->lwp_thread->td_pcb;
134 *pcb2 = *lp1->lwp_thread->td_pcb;
135
136 /*
137 * Create a new fresh stack for the new process.
138 * Copy the trap frame for the return to user mode as if from a
139 * syscall. This copies the user mode register values. The
140 * 16 byte offset saves space for vm86, and must match
141 * common_tss.esp0 (kernel stack pointer on entry from user mode)
142 *
143 * pcb_esp must allocate an additional call-return pointer below
144 * the trap frame which will be restored by cpu_restore from
145 * PCB_EIP, and the thread's td_sp pointer must allocate an
146 * additonal two worsd below the pcb_esp call-return pointer to
147 * hold the LWKT restore function pointer and eflags.
148 *
149 * The LWKT restore function pointer must be set to cpu_restore,
150 * which is our standard heavy weight process switch-in function.
151 * YYY eventually we should shortcut fork_return and fork_trampoline
152 * to use the LWKT restore function directly so we can get rid of
153 * all the extra crap we are setting up.
154 */
155 lp2->lwp_md.md_regs = (struct trapframe *)((char *)pcb2 - 16) - 1;
156 bcopy(lp1->lwp_md.md_regs, lp2->lwp_md.md_regs, sizeof(*lp2->lwp_md.md_regs));
157
158 /*
159 * Set registers for trampoline to user mode. Leave space for the
160 * return address on stack. These are the kernel mode register values.
161 */
431d0fef 162 pcb2->pcb_unused01 = 0;
6f7b98e0
MD
163 pcb2->pcb_edi = 0;
164 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */
165 pcb2->pcb_ebp = 0;
166 pcb2->pcb_esp = (int)lp2->lwp_md.md_regs - sizeof(void *);
167 pcb2->pcb_ebx = (int)lp2; /* fork_trampoline argument */
168 pcb2->pcb_eip = (int)fork_trampoline;
169 lp2->lwp_thread->td_sp = (char *)(pcb2->pcb_esp - sizeof(void *));
170 *(u_int32_t *)lp2->lwp_thread->td_sp = PSL_USER;
171 lp2->lwp_thread->td_sp -= sizeof(void *);
172 *(void **)lp2->lwp_thread->td_sp = (void *)cpu_heavy_restore;
173
174 /*
6f7b98e0
MD
175 * pcb2->pcb_ldt: duplicated below, if necessary.
176 * pcb2->pcb_savefpu: cloned above.
177 * pcb2->pcb_flags: cloned above (always 0 here?).
178 * pcb2->pcb_onfault: cloned above (always NULL here?).
179 */
180
181 /*
182 * XXX don't copy the i/o pages. this should probably be fixed.
183 */
184 pcb2->pcb_ext = 0;
185
186 /* Copy the LDT, if necessary. */
187 if (pcb2->pcb_ldt != 0) {
188 if (flags & RFMEM) {
189 pcb2->pcb_ldt->ldt_refcnt++;
190 } else {
191 pcb2->pcb_ldt = user_ldt_alloc(pcb2,
192 pcb2->pcb_ldt->ldt_len);
193 }
194 }
195 bcopy(&lp1->lwp_thread->td_tls, &lp2->lwp_thread->td_tls,
196 sizeof(lp2->lwp_thread->td_tls));
197 /*
198 * Now, cpu_switch() can schedule the new process.
199 * pcb_esp is loaded pointing to the cpu_switch() stack frame
200 * containing the return address when exiting cpu_switch.
201 * This will normally be to fork_trampoline(), which will have
202 * %ebx loaded with the new proc's pointer. fork_trampoline()
203 * will set up a stack to call fork_return(p, frame); to complete
204 * the return to user-mode.
205 */
206}
207
208/*
91bd9c1e
SS
209 * Prepare new lwp to return to the address specified in params.
210 */
211int
212cpu_prepare_lwp(struct lwp *lp, struct lwp_params *params)
213{
214 struct trapframe *regs = lp->lwp_md.md_regs;
215 void *bad_return = NULL;
216 int error;
217
218 regs->tf_eip = (int)params->func;
219 regs->tf_esp = (int)params->stack;
220 /* Set up argument for function call */
221 regs->tf_esp -= sizeof(params->arg);
222 error = copyout(&params->arg, (void *)regs->tf_esp,
223 sizeof(params->arg));
224 if (error)
225 return (error);
226 /*
227 * Set up fake return address. As the lwp function may never return,
228 * we simply copy out a NULL pointer and force the lwp to receive
229 * a SIGSEGV if it returns anyways.
230 */
231 regs->tf_esp -= sizeof(void *);
232 error = copyout(&bad_return, (void *)regs->tf_esp,
233 sizeof(bad_return));
234 if (error)
235 return (error);
236
237 cpu_set_fork_handler(lp,
238 (void (*)(void *, struct trapframe *))generic_lwp_return, lp);
239 return (0);
240}
241
242/*
6f7b98e0
MD
243 * Intercept the return address from a freshly forked process that has NOT
244 * been scheduled yet.
245 *
246 * This is needed to make kernel threads stay in kernel mode.
247 */
248void
91bd9c1e
SS
249cpu_set_fork_handler(struct lwp *lp, void (*func)(void *, struct trapframe *),
250 void *arg)
6f7b98e0
MD
251{
252 /*
253 * Note that the trap frame follows the args, so the function
254 * is really called like this: func(arg, frame);
255 */
256 lp->lwp_thread->td_pcb->pcb_esi = (int) func; /* function */
257 lp->lwp_thread->td_pcb->pcb_ebx = (int) arg; /* first arg */
258}
259
260void
261cpu_set_thread_handler(thread_t td, void (*rfunc)(void), void *func, void *arg)
262{
263 td->td_pcb->pcb_esi = (int)func;
264 td->td_pcb->pcb_ebx = (int) arg;
265 td->td_switch = cpu_lwkt_switch;
266 td->td_sp -= sizeof(void *);
267 *(void **)td->td_sp = rfunc; /* exit function on return */
268 td->td_sp -= sizeof(void *);
269 *(void **)td->td_sp = cpu_kthread_restore;
270}
271
272void
c6880072 273cpu_lwp_exit(void)
6f7b98e0
MD
274{
275 struct thread *td = curthread;
276 struct pcb *pcb;
277 struct pcb_ext *ext;
278
279#if NNPX > 0
280 npxexit();
281#endif /* NNPX */
282
283 /*
284 * If we were using a private TSS do a forced-switch to ourselves
285 * to switch back to the common TSS before freeing it.
286 */
287 pcb = td->td_pcb;
288 if ((ext = pcb->pcb_ext) != NULL) {
289 crit_enter();
290 pcb->pcb_ext = NULL;
291 td->td_switch(td);
292 crit_exit();
293 kmem_free(&kernel_map, (vm_offset_t)ext, ctob(IOPAGES + 1));
294 }
295 user_ldt_free(pcb);
296 if (pcb->pcb_flags & PCB_DBREGS) {
297 /*
298 * disable all hardware breakpoints
299 */
300 reset_dbregs();
301 pcb->pcb_flags &= ~PCB_DBREGS;
302 }
303 td->td_gd->gd_cnt.v_swtch++;
304
f8abf63c 305 dsched_exit_thread(td);
6f7b98e0 306 crit_enter_quick(td);
09b62fa4
MD
307 if (td->td_flags & TDF_TSLEEPQ)
308 tsleep_remove(td);
6f7b98e0
MD
309 lwkt_deschedule_self(td);
310 lwkt_remove_tdallq(td);
311 cpu_thread_exit();
312}
313
314/*
315 * Terminate the current thread. The caller must have already acquired
316 * the thread's rwlock and placed it on a reap list or otherwise notified
317 * a reaper of its existance. We set a special assembly switch function which
318 * releases td_rwlock after it has cleaned up the MMU state and switched
319 * out the stack.
320 *
321 * Must be caller from a critical section and with the thread descheduled.
322 */
323void
324cpu_thread_exit(void)
325{
326 curthread->td_switch = cpu_exit_switch;
327 curthread->td_flags |= TDF_EXITING;
328 lwkt_switch();
329 panic("cpu_exit");
330}
331
332/*
333 * Process Reaper. Called after the caller has acquired the thread's
334 * rwlock and removed it from the reap list.
335 */
336void
337cpu_proc_wait(struct proc *p)
338{
6f7b98e0 339 /* drop per-process resources */
c6880072 340 pmap_dispose_proc(p);
6f7b98e0 341}
6f7b98e0
MD
342
343#ifdef notyet
344static void
345setredzone(u_short *pte, caddr_t vaddr)
346{
347/* eventually do this by setting up an expand-down stack segment
348 for ss0: selector, allowing stack access down to top of u.
349 this means though that protection violations need to be handled
350 thru a double fault exception that must do an integral task
351 switch to a known good context, within which a dump can be
352 taken. a sensible scheme might be to save the initial context
353 used by sched (that has physical memory mapped 1:1 at bottom)
354 and take the dump while still in mapped mode */
355}
356#endif
357
358/*
359 * Convert kernel VA to physical address
360 */
361vm_paddr_t
362kvtop(void *addr)
363{
364 vm_paddr_t pa;
365
366 pa = pmap_kextract((vm_offset_t)addr);
367 if (pa == 0)
368 panic("kvtop: zero page frame");
369 return (pa);
370}
371
6f7b98e0
MD
372SYSCTL_DECL(_vm_stats_misc);
373
374static int cnt_prezero;
375
376SYSCTL_INT(_vm_stats_misc, OID_AUTO,
377 cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, "");
378
379/*
380 * Tell whether this address is in some physical memory region.
381 * Currently used by the kernel coredump code in order to avoid
382 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
383 * or other unpredictable behaviour.
384 */
385
386int
387is_physical_memory(vm_offset_t addr)
388{
389 return 1;
390}
391
460426e6
MD
392/*
393 * Used by /dev/kmem to determine if we can safely read or write
394 * the requested KVA range. Some portions of kernel memory are
395 * not governed by our virtual page table.
396 */
397extern int32_t _end;
398extern void _start(void);
399
400int
401kvm_access_check(vm_offset_t saddr, vm_offset_t eaddr, int prot)
402{
403 vm_offset_t addr;
404
fd1501d5 405 if (saddr >= trunc_page((vm_offset_t)&_start) && eaddr <= round_page((vm_offset_t)&_end))
460426e6
MD
406 return 0;
407 if (saddr < KvaStart)
408 return EFAULT;
409 if (eaddr >= KvaEnd)
410 return EFAULT;
411 for (addr = saddr; addr < eaddr; addr += PAGE_SIZE) {
412 if (pmap_extract(&kernel_pmap, addr) == 0)
413 return EFAULT;
414 }
415 if (!kernacc((caddr_t)saddr, eaddr - saddr, prot))
416 return EFAULT;
417 return 0;
418}
419