kernel - Make pcb_onfault more robust.
[dragonfly.git] / sys / platform / pc32 / i386 / vm_machdep.c
CommitLineData
984263bc
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 $
42 */
43
1f2de5d4 44#include "use_npx.h"
984263bc
MD
45#include "opt_reset.h"
46
47#include <sys/param.h>
48#include <sys/systm.h>
49#include <sys/malloc.h>
50#include <sys/proc.h>
51#include <sys/buf.h>
ef0fdad1 52#include <sys/interrupt.h>
984263bc
MD
53#include <sys/vnode.h>
54#include <sys/vmmeter.h>
55#include <sys/kernel.h>
56#include <sys/sysctl.h>
57#include <sys/unistd.h>
58
59#include <machine/clock.h>
60#include <machine/cpu.h>
61#include <machine/md_var.h>
984263bc 62#include <machine/smp.h>
984263bc
MD
63#include <machine/pcb.h>
64#include <machine/pcb_ext.h>
65#include <machine/vm86.h>
4a22e893 66#include <machine/segments.h>
85100692 67#include <machine/globaldata.h> /* npxthread */
984263bc
MD
68
69#include <vm/vm.h>
70#include <vm/vm_param.h>
71#include <sys/lock.h>
72#include <vm/vm_kern.h>
73#include <vm/vm_page.h>
74#include <vm/vm_map.h>
75#include <vm/vm_extern.h>
76
77#include <sys/user.h>
684a93c4 78
ae8050a4 79#include <sys/thread2.h>
684a93c4 80#include <sys/mplock2.h>
984263bc 81
21ce0dfa 82#include <bus/isa/isa.h>
984263bc 83
3ae0cd58 84static void cpu_reset_real (void);
984263bc 85#ifdef SMP
3ae0cd58 86static void cpu_reset_proxy (void);
984263bc
MD
87static u_int cpu_reset_proxyid;
88static volatile u_int cpu_reset_proxy_active;
89#endif
90extern int _ucodesel, _udatasel;
91
984263bc
MD
92
93/*
bb3cd951 94 * Finish a fork operation, with lwp lp2 nearly set up.
984263bc
MD
95 * Copy and update the pcb, set up the stack so that the child
96 * ready to run and return to user mode.
97 */
98void
bb3cd951 99cpu_fork(struct lwp *lp1, struct lwp *lp2, int flags)
984263bc
MD
100{
101 struct pcb *pcb2;
102
103 if ((flags & RFPROC) == 0) {
984263bc
MD
104 if ((flags & RFMEM) == 0) {
105 /* unshare user LDT */
bb3cd951 106 struct pcb *pcb1 = lp1->lwp_thread->td_pcb;
984263bc
MD
107 struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt;
108 if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) {
109 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len);
110 user_ldt_free(pcb1);
111 pcb1->pcb_ldt = pcb_ldt;
112 set_user_ldt(pcb1);
113 }
114 }
984263bc
MD
115 return;
116 }
117
118#if NNPX > 0
bb3cd951
SS
119 /* Ensure that lp1's pcb is up to date. */
120 if (mdcpu->gd_npxthread == lp1->lwp_thread)
121 npxsave(lp1->lwp_thread->td_savefpu);
984263bc 122#endif
a96ef868
MD
123
124 /*
bb3cd951 125 * Copy lp1's PCB. This really only applies to the
a96ef868
MD
126 * debug registers and FP state, but its faster to just copy the
127 * whole thing. Because we only save the PCB at switchout time,
4e7c41c5 128 * the register state may not be current.
a96ef868 129 */
bb3cd951
SS
130 pcb2 = lp2->lwp_thread->td_pcb;
131 *pcb2 = *lp1->lwp_thread->td_pcb;
984263bc
MD
132
133 /*
134 * Create a new fresh stack for the new process.
135 * Copy the trap frame for the return to user mode as if from a
b7c628e4
MD
136 * syscall. This copies the user mode register values. The
137 * 16 byte offset saves space for vm86, and must match
138 * common_tss.esp0 (kernel stack pointer on entry from user mode)
8ad65e08
MD
139 *
140 * pcb_esp must allocate an additional call-return pointer below
d9eea1a5
MD
141 * the trap frame which will be restored by cpu_restore from
142 * PCB_EIP, and the thread's td_sp pointer must allocate an
143 * additonal two worsd below the pcb_esp call-return pointer to
144 * hold the LWKT restore function pointer and eflags.
8ad65e08
MD
145 *
146 * The LWKT restore function pointer must be set to cpu_restore,
147 * which is our standard heavy weight process switch-in function.
148 * YYY eventually we should shortcut fork_return and fork_trampoline
149 * to use the LWKT restore function directly so we can get rid of
150 * all the extra crap we are setting up.
984263bc 151 */
bb3cd951
SS
152 lp2->lwp_md.md_regs = (struct trapframe *)((char *)pcb2 - 16) - 1;
153 bcopy(lp1->lwp_md.md_regs, lp2->lwp_md.md_regs, sizeof(*lp2->lwp_md.md_regs));
984263bc
MD
154
155 /*
156 * Set registers for trampoline to user mode. Leave space for the
157 * return address on stack. These are the kernel mode register values.
158 */
bb3cd951 159 pcb2->pcb_cr3 = vtophys(vmspace_pmap(lp2->lwp_proc->p_vmspace)->pm_pdir);
984263bc
MD
160 pcb2->pcb_edi = 0;
161 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */
162 pcb2->pcb_ebp = 0;
bb3cd951
SS
163 pcb2->pcb_esp = (int)lp2->lwp_md.md_regs - sizeof(void *);
164 pcb2->pcb_ebx = (int)lp2; /* fork_trampoline argument */
984263bc 165 pcb2->pcb_eip = (int)fork_trampoline;
bb3cd951
SS
166 lp2->lwp_thread->td_sp = (char *)(pcb2->pcb_esp - sizeof(void *));
167 *(u_int32_t *)lp2->lwp_thread->td_sp = PSL_USER;
168 lp2->lwp_thread->td_sp -= sizeof(void *);
169 *(void **)lp2->lwp_thread->td_sp = (void *)cpu_heavy_restore;
a96ef868
MD
170
171 /*
984263bc
MD
172 * pcb2->pcb_ldt: duplicated below, if necessary.
173 * pcb2->pcb_savefpu: cloned above.
93ad6da2
MD
174 * pcb2->pcb_flags: cloned above (always 0 here).
175 * pcb2->pcb_onfault: cloned above (always NULL here).
176 * pcb2->pcb_onfault_sp:cloned above (don't care)
984263bc
MD
177 */
178
984263bc
MD
179 /*
180 * XXX don't copy the i/o pages. this should probably be fixed.
181 */
d8061892 182 pcb2->pcb_ext = NULL;
984263bc 183
984263bc 184 /* Copy the LDT, if necessary. */
d8061892 185 if (pcb2->pcb_ldt != NULL) {
984263bc
MD
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 }
bb3cd951
SS
193 bcopy(&lp1->lwp_thread->td_tls, &lp2->lwp_thread->td_tls,
194 sizeof(lp2->lwp_thread->td_tls));
984263bc 195 /*
08f2f1bb 196 * Now, cpu_switch() can schedule the new lwp.
984263bc
MD
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
08f2f1bb
SS
200 * %ebx loaded with the new lwp's pointer. fork_trampoline()
201 * will set up a stack to call fork_return(lp, frame); to complete
984263bc
MD
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
29adef62
NT
216 regs->tf_eip = (int)params->func;
217 regs->tf_esp = (int)params->stack;
91bd9c1e
SS
218 /* Set up argument for function call */
219 regs->tf_esp -= sizeof(params->arg);
29adef62
NT
220 error =
221 copyout(&params->arg, (void *)regs->tf_esp, sizeof(params->arg));
91bd9c1e
SS
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 *);
29adef62 230 error = copyout(&bad_return, (void *)regs->tf_esp, sizeof(bad_return));
91bd9c1e
SS
231 if (error)
232 return (error);
233
234 cpu_set_fork_handler(lp,
235 (void (*)(void *, struct trapframe *))generic_lwp_return, lp);
236 return (0);
237}
238
239/*
984263bc
MD
240 * Intercept the return address from a freshly forked process that has NOT
241 * been scheduled yet.
242 *
243 * This is needed to make kernel threads stay in kernel mode.
244 */
245void
91bd9c1e
SS
246cpu_set_fork_handler(struct lwp *lp, void (*func)(void *, struct trapframe *),
247 void *arg)
984263bc
MD
248{
249 /*
250 * Note that the trap frame follows the args, so the function
251 * is really called like this: func(arg, frame);
252 */
bb3cd951
SS
253 lp->lwp_thread->td_pcb->pcb_esi = (int) func; /* function */
254 lp->lwp_thread->td_pcb->pcb_ebx = (int) arg; /* first arg */
984263bc
MD
255}
256
257void
0cfcada1
MD
258cpu_set_thread_handler(thread_t td, void (*rfunc)(void), void *func, void *arg)
259{
260 td->td_pcb->pcb_esi = (int)func;
261 td->td_pcb->pcb_ebx = (int) arg;
262 td->td_switch = cpu_lwkt_switch;
263 td->td_sp -= sizeof(void *);
264 *(void **)td->td_sp = rfunc; /* exit function on return */
265 td->td_sp -= sizeof(void *);
266 *(void **)td->td_sp = cpu_kthread_restore;
267}
268
269void
c6880072 270cpu_lwp_exit(void)
984263bc 271{
99df837e 272 struct thread *td = curthread;
ae8050a4 273 struct pcb *pcb;
df011816 274 struct pcb_ext *ext;
99df837e 275
ae8050a4 276 /*
df011816
MD
277 * If we were using a private TSS do a forced-switch to ourselves
278 * to switch back to the common TSS before freeing it.
ae8050a4 279 */
99df837e 280 pcb = td->td_pcb;
df011816
MD
281 if ((ext = pcb->pcb_ext) != NULL) {
282 crit_enter();
283 pcb->pcb_ext = NULL;
eb673128 284 lwkt_switch_return(td->td_switch(td));
df011816 285 crit_exit();
e4846942 286 kmem_free(&kernel_map, (vm_offset_t)ext, ctob(IOPAGES + 1));
984263bc 287 }
984263bc 288 user_ldt_free(pcb);
984263bc
MD
289 if (pcb->pcb_flags & PCB_DBREGS) {
290 /*
291 * disable all hardware breakpoints
292 */
293 reset_dbregs();
294 pcb->pcb_flags &= ~PCB_DBREGS;
295 }
37af14fe 296 td->td_gd->gd_cnt.v_swtch++;
ae8050a4 297
37af14fe 298 crit_enter_quick(td);
09b62fa4
MD
299 if (td->td_flags & TDF_TSLEEPQ)
300 tsleep_remove(td);
37af14fe 301 lwkt_deschedule_self(td);
e56e4dea 302 lwkt_remove_tdallq(td);
99df837e
MD
303 cpu_thread_exit();
304}
305
306/*
307 * Terminate the current thread. The caller must have already acquired
308 * the thread's rwlock and placed it on a reap list or otherwise notified
309 * a reaper of its existance. We set a special assembly switch function which
310 * releases td_rwlock after it has cleaned up the MMU state and switched
311 * out the stack.
312 *
313 * Must be caller from a critical section and with the thread descheduled.
314 */
315void
316cpu_thread_exit(void)
317{
9e6e869e
MD
318#if NNPX > 0
319 npxexit();
320#endif
99df837e 321 curthread->td_switch = cpu_exit_switch;
c1102e9f 322 curthread->td_flags |= TDF_EXITING;
8ad65e08 323 lwkt_switch();
984263bc
MD
324 panic("cpu_exit");
325}
326
984263bc
MD
327#ifdef notyet
328static void
f123d5a1 329setredzone(u_short *pte, caddr_t vaddr)
984263bc
MD
330{
331/* eventually do this by setting up an expand-down stack segment
332 for ss0: selector, allowing stack access down to top of u.
333 this means though that protection violations need to be handled
334 thru a double fault exception that must do an integral task
335 switch to a known good context, within which a dump can be
336 taken. a sensible scheme might be to save the initial context
337 used by sched (that has physical memory mapped 1:1 at bottom)
338 and take the dump while still in mapped mode */
339}
340#endif
341
342/*
343 * Convert kernel VA to physical address
344 */
6ef943a3 345vm_paddr_t
984263bc
MD
346kvtop(void *addr)
347{
6ef943a3 348 vm_paddr_t pa;
984263bc 349
6ef943a3
MD
350 pa = pmap_kextract((vm_offset_t)addr);
351 if (pa == 0)
984263bc 352 panic("kvtop: zero page frame");
6ef943a3 353 return (pa);
984263bc
MD
354}
355
356/*
357 * Force reset the processor by invalidating the entire address space!
358 */
359
360#ifdef SMP
361static void
f123d5a1 362cpu_reset_proxy(void)
984263bc 363{
984263bc
MD
364 cpu_reset_proxy_active = 1;
365 while (cpu_reset_proxy_active == 1)
366 ; /* Wait for other cpu to disable interupts */
26be20a0 367 kprintf("cpu_reset_proxy: Grabbed mp lock for BSP\n");
984263bc
MD
368 cpu_reset_proxy_active = 3;
369 while (cpu_reset_proxy_active == 3)
370 ; /* Wait for other cpu to enable interrupts */
da23a592 371 stop_cpus(CPUMASK(cpu_reset_proxyid));
26be20a0 372 kprintf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
984263bc
MD
373 DELAY(1000000);
374 cpu_reset_real();
375}
376#endif
377
378void
f123d5a1 379cpu_reset(void)
984263bc
MD
380{
381#ifdef SMP
0f7a3396 382 if (smp_active_mask == 1) {
984263bc
MD
383 cpu_reset_real();
384 /* NOTREACHED */
385 } else {
da23a592 386 cpumask_t map;
984263bc 387 int cnt;
26be20a0 388 kprintf("cpu_reset called on cpu#%d\n",mycpu->gd_cpuid);
984263bc 389
0f7a3396 390 map = mycpu->gd_other_cpus & ~stopped_cpus & smp_active_mask;
984263bc
MD
391
392 if (map != 0) {
26be20a0 393 kprintf("cpu_reset: Stopping other CPUs\n");
984263bc
MD
394 stop_cpus(map); /* Stop all other CPUs */
395 }
396
72740893 397 if (mycpu->gd_cpuid == 0) {
984263bc
MD
398 DELAY(1000000);
399 cpu_reset_real();
400 /* NOTREACHED */
401 } else {
402 /* We are not BSP (CPU #0) */
403
72740893 404 cpu_reset_proxyid = mycpu->gd_cpuid;
984263bc 405 cpustop_restartfunc = cpu_reset_proxy;
26be20a0 406 kprintf("cpu_reset: Restarting BSP\n");
984263bc
MD
407 started_cpus = (1<<0); /* Restart CPU #0 */
408
409 cnt = 0;
410 while (cpu_reset_proxy_active == 0 && cnt < 10000000)
411 cnt++; /* Wait for BSP to announce restart */
412 if (cpu_reset_proxy_active == 0)
26be20a0 413 kprintf("cpu_reset: Failed to restart BSP\n");
984263bc
MD
414 __asm __volatile("cli" : : : "memory");
415 cpu_reset_proxy_active = 2;
416 cnt = 0;
417 while (cpu_reset_proxy_active == 2 && cnt < 10000000)
418 cnt++; /* Do nothing */
419 if (cpu_reset_proxy_active == 2) {
26be20a0 420 kprintf("cpu_reset: BSP did not grab mp lock\n");
984263bc
MD
421 cpu_reset_real(); /* XXX: Bogus ? */
422 }
423 cpu_reset_proxy_active = 4;
424 __asm __volatile("sti" : : : "memory");
425 while (1);
426 /* NOTREACHED */
427 }
428 }
429#else
430 cpu_reset_real();
431#endif
432}
433
434static void
f123d5a1 435cpu_reset_real(void)
984263bc 436{
984263bc
MD
437 /*
438 * Attempt to do a CPU reset via the keyboard controller,
439 * do not turn of the GateA20, as any machine that fails
440 * to do the reset here would then end up in no man's land.
441 */
442
443#if !defined(BROKEN_KEYBOARD_RESET)
444 outb(IO_KBD + 4, 0xFE);
445 DELAY(500000); /* wait 0.5 sec to see if that did it */
26be20a0
SW
446 kprintf("Keyboard reset did not work, attempting CPU shutdown\n");
447 DELAY(1000000); /* wait 1 sec for kprintf to complete */
984263bc 448#endif
984263bc
MD
449 /* force a shutdown by unmapping entire address space ! */
450 bzero((caddr_t) PTD, PAGE_SIZE);
451
452 /* "good night, sweet prince .... <THUNK!>" */
0f7a3396 453 cpu_invltlb();
984263bc
MD
454 /* NOTREACHED */
455 while(1);
456}
457
984263bc
MD
458SYSCTL_DECL(_vm_stats_misc);
459
26a0694b 460static void
477d3c1c 461swi_vm(void *arg, void *frame)
ef0fdad1 462{
984263bc
MD
463 if (busdma_swi_pending != 0)
464 busdma_swi();
465}
466
ef0fdad1
MD
467static void
468swi_vm_setup(void *arg)
469{
1da8d52f 470 register_swi(SWI_VM, swi_vm, NULL, "swi_vm", NULL, 0);
ef0fdad1
MD
471}
472
ba39e2e0 473SYSINIT(vm_setup, SI_BOOT2_MACHDEP, SI_ORDER_ANY, swi_vm_setup, NULL);
ef0fdad1 474
135d7199
MD
475/*
476 * platform-specific vmspace initialization (nothing for i386)
477 */
478void
479cpu_vmspace_alloc(struct vmspace *vm __unused)
480{
481}
482
483void
484cpu_vmspace_free(struct vmspace *vm __unused)
485{
486}
460426e6
MD
487
488/*
489 * Used by /dev/kmem to determine if we can safely read or write
490 * the requested KVA range.
491 */
492int
493kvm_access_check(vm_offset_t saddr, vm_offset_t eaddr, int prot)
494{
495 vm_offset_t addr;
496
497 if (saddr < KvaStart)
498 return EFAULT;
499 if (eaddr >= KvaEnd)
500 return EFAULT;
501 for (addr = saddr; addr < eaddr; addr += PAGE_SIZE) {
502 if (pmap_extract(&kernel_pmap, addr) == 0)
503 return EFAULT;
504 }
505 if (!kernacc((caddr_t)saddr, eaddr - saddr, prot))
506 return EFAULT;
507 return 0;
508}
509