kernel - Fix floating point save state structure and minor npx issues
[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 $
6f7b98e0
MD
42 */
43
44#include "use_npx.h"
6f7b98e0
MD
45
46#include <sys/param.h>
47#include <sys/systm.h>
48#include <sys/malloc.h>
49#include <sys/proc.h>
50#include <sys/buf.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>
57
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 */
67
68#include <vm/vm.h>
69#include <vm/vm_param.h>
70#include <sys/lock.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>
75
76#include <sys/user.h>
684a93c4 77
6f7b98e0 78#include <sys/thread2.h>
684a93c4 79#include <sys/mplock2.h>
6f7b98e0 80
d7510ae6 81#include <bus/isa/isa.h>
6f7b98e0
MD
82
83#include <stdio.h>
84#include <stdlib.h>
85
0955fd91 86char machine[] = MACHINE;
6f7b98e0
MD
87SYSCTL_STRING(_hw, HW_MACHINE, machine, CTLFLAG_RD,
88 machine, 0, "Machine class");
89
9ad680a3 90char cpu_vendor[] = "DragonFly"; /* XXX */
dfa6eaf6 91u_int cpu_vendor_id = 0; /* XXX */
9ad680a3
MD
92u_int cpu_id = 0x80000000; /* XXX */
93
6f7b98e0
MD
94/*
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.
98 */
99void
100cpu_fork(struct lwp *lp1, struct lwp *lp2, int flags)
101{
102 struct pcb *pcb2;
103
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);
111 user_ldt_free(pcb1);
112 pcb1->pcb_ldt = pcb_ldt;
113 set_user_ldt(pcb1);
114 }
115 }
116 return;
117 }
118
119#if NNPX > 0
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);
123#endif
124
125 /*
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,
4e7c41c5 129 * the register state may not be current.
6f7b98e0
MD
130 */
131 pcb2 = lp2->lwp_thread->td_pcb;
132 *pcb2 = *lp1->lwp_thread->td_pcb;
133
134 /*
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)
140 *
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.
146 *
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.
152 */
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));
155
156 /*
157 * Set registers for trampoline to user mode. Leave space for the
158 * return address on stack. These are the kernel mode register values.
159 */
431d0fef 160 pcb2->pcb_unused01 = 0;
6f7b98e0
MD
161 pcb2->pcb_edi = 0;
162 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */
163 pcb2->pcb_ebp = 0;
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;
171
172 /*
6f7b98e0
MD
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?).
177 */
178
179 /*
180 * XXX don't copy the i/o pages. this should probably be fixed.
181 */
182 pcb2->pcb_ext = 0;
183
184 /* Copy the LDT, if necessary. */
185 if (pcb2->pcb_ldt != 0) {
186 if (flags & RFMEM) {
187 pcb2->pcb_ldt->ldt_refcnt++;
188 } else {
189 pcb2->pcb_ldt = user_ldt_alloc(pcb2,
190 pcb2->pcb_ldt->ldt_len);
191 }
192 }
193 bcopy(&lp1->lwp_thread->td_tls, &lp2->lwp_thread->td_tls,
194 sizeof(lp2->lwp_thread->td_tls));
195 /*
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.
203 */
204}
205
206/*
91bd9c1e
SS
207 * Prepare new lwp to return to the address specified in params.
208 */
209int
210cpu_prepare_lwp(struct lwp *lp, struct lwp_params *params)
211{
212 struct trapframe *regs = lp->lwp_md.md_regs;
213 void *bad_return = NULL;
214 int error;
215
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(&params->arg, (void *)regs->tf_esp,
221 sizeof(params->arg));
222 if (error)
223 return (error);
224 /*
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.
228 */
229 regs->tf_esp -= sizeof(void *);
230 error = copyout(&bad_return, (void *)regs->tf_esp,
231 sizeof(bad_return));
232 if (error)
233 return (error);
234
235 cpu_set_fork_handler(lp,
236 (void (*)(void *, struct trapframe *))generic_lwp_return, lp);
237 return (0);
238}
239
240/*
6f7b98e0
MD
241 * Intercept the return address from a freshly forked process that has NOT
242 * been scheduled yet.
243 *
244 * This is needed to make kernel threads stay in kernel mode.
245 */
246void
91bd9c1e
SS
247cpu_set_fork_handler(struct lwp *lp, void (*func)(void *, struct trapframe *),
248 void *arg)
6f7b98e0
MD
249{
250 /*
251 * Note that the trap frame follows the args, so the function
252 * is really called like this: func(arg, frame);
253 */
254 lp->lwp_thread->td_pcb->pcb_esi = (int) func; /* function */
255 lp->lwp_thread->td_pcb->pcb_ebx = (int) arg; /* first arg */
256}
257
258void
259cpu_set_thread_handler(thread_t td, void (*rfunc)(void), void *func, void *arg)
260{
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;
268}
269
270void
c6880072 271cpu_lwp_exit(void)
6f7b98e0
MD
272{
273 struct thread *td = curthread;
274 struct pcb *pcb;
275 struct pcb_ext *ext;
276
6f7b98e0
MD
277 /*
278 * If we were using a private TSS do a forced-switch to ourselves
279 * to switch back to the common TSS before freeing it.
280 */
281 pcb = td->td_pcb;
282 if ((ext = pcb->pcb_ext) != NULL) {
283 crit_enter();
284 pcb->pcb_ext = NULL;
eb673128 285 lwkt_switch_return(td->td_switch(td));
6f7b98e0
MD
286 crit_exit();
287 kmem_free(&kernel_map, (vm_offset_t)ext, ctob(IOPAGES + 1));
288 }
289 user_ldt_free(pcb);
290 if (pcb->pcb_flags & PCB_DBREGS) {
291 /*
292 * disable all hardware breakpoints
293 */
294 reset_dbregs();
295 pcb->pcb_flags &= ~PCB_DBREGS;
296 }
297 td->td_gd->gd_cnt.v_swtch++;
298
299 crit_enter_quick(td);
09b62fa4
MD
300 if (td->td_flags & TDF_TSLEEPQ)
301 tsleep_remove(td);
6f7b98e0
MD
302 lwkt_deschedule_self(td);
303 lwkt_remove_tdallq(td);
304 cpu_thread_exit();
305}
306
307/*
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
312 * out the stack.
313 *
314 * Must be caller from a critical section and with the thread descheduled.
315 */
316void
317cpu_thread_exit(void)
318{
9e6e869e
MD
319#if NNPX > 0
320 npxexit();
321#endif
6f7b98e0
MD
322 curthread->td_switch = cpu_exit_switch;
323 curthread->td_flags |= TDF_EXITING;
324 lwkt_switch();
325 panic("cpu_exit");
326}
327
6f7b98e0
MD
328#ifdef notyet
329static void
330setredzone(u_short *pte, caddr_t vaddr)
331{
332/* eventually do this by setting up an expand-down stack segment
333 for ss0: selector, allowing stack access down to top of u.
334 this means though that protection violations need to be handled
335 thru a double fault exception that must do an integral task
336 switch to a known good context, within which a dump can be
337 taken. a sensible scheme might be to save the initial context
338 used by sched (that has physical memory mapped 1:1 at bottom)
339 and take the dump while still in mapped mode */
340}
341#endif
342
343/*
344 * Convert kernel VA to physical address
345 */
346vm_paddr_t
347kvtop(void *addr)
348{
349 vm_paddr_t pa;
350
351 pa = pmap_kextract((vm_offset_t)addr);
352 if (pa == 0)
353 panic("kvtop: zero page frame");
354 return (pa);
355}
356
6f7b98e0
MD
357SYSCTL_DECL(_vm_stats_misc);
358
6f7b98e0
MD
359/*
360 * Tell whether this address is in some physical memory region.
361 * Currently used by the kernel coredump code in order to avoid
362 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
363 * or other unpredictable behaviour.
364 */
365
366int
367is_physical_memory(vm_offset_t addr)
368{
369 return 1;
370}
371
460426e6
MD
372/*
373 * Used by /dev/kmem to determine if we can safely read or write
374 * the requested KVA range. Some portions of kernel memory are
375 * not governed by our virtual page table.
376 */
377extern int32_t _end;
378extern void _start(void);
379
380int
381kvm_access_check(vm_offset_t saddr, vm_offset_t eaddr, int prot)
382{
383 vm_offset_t addr;
384
fd1501d5 385 if (saddr >= trunc_page((vm_offset_t)&_start) && eaddr <= round_page((vm_offset_t)&_end))
460426e6
MD
386 return 0;
387 if (saddr < KvaStart)
388 return EFAULT;
389 if (eaddr >= KvaEnd)
390 return EFAULT;
391 for (addr = saddr; addr < eaddr; addr += PAGE_SIZE) {
392 if (pmap_extract(&kernel_pmap, addr) == 0)
393 return EFAULT;
394 }
395 if (!kernacc((caddr_t)saddr, eaddr - saddr, prot))
396 return EFAULT;
397 return 0;
398}
399