kernel - Fix floating point save state structure and minor npx issues
[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 $
21ce0dfa 42 * $DragonFly: src/sys/platform/pc32/i386/vm_machdep.c,v 1.61 2008/08/02 01:14:43 dillon Exp $
984263bc
MD
43 */
44
1f2de5d4
MD
45#include "use_npx.h"
46#include "use_isa.h"
984263bc
MD
47#include "opt_reset.h"
48
49#include <sys/param.h>
50#include <sys/systm.h>
51#include <sys/malloc.h>
52#include <sys/proc.h>
53#include <sys/buf.h>
ef0fdad1 54#include <sys/interrupt.h>
984263bc
MD
55#include <sys/vnode.h>
56#include <sys/vmmeter.h>
57#include <sys/kernel.h>
58#include <sys/sysctl.h>
59#include <sys/unistd.h>
60
61#include <machine/clock.h>
62#include <machine/cpu.h>
63#include <machine/md_var.h>
984263bc 64#include <machine/smp.h>
984263bc
MD
65#include <machine/pcb.h>
66#include <machine/pcb_ext.h>
67#include <machine/vm86.h>
4a22e893 68#include <machine/segments.h>
85100692 69#include <machine/globaldata.h> /* npxthread */
984263bc
MD
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
ae8050a4 81#include <sys/thread2.h>
684a93c4 82#include <sys/mplock2.h>
984263bc 83
21ce0dfa 84#include <bus/isa/isa.h>
984263bc 85
3ae0cd58 86static void cpu_reset_real (void);
984263bc 87#ifdef SMP
3ae0cd58 88static void cpu_reset_proxy (void);
984263bc
MD
89static u_int cpu_reset_proxyid;
90static volatile u_int cpu_reset_proxy_active;
91#endif
92extern int _ucodesel, _udatasel;
93
984263bc
MD
94
95/*
bb3cd951 96 * Finish a fork operation, with lwp lp2 nearly set up.
984263bc
MD
97 * Copy and update the pcb, set up the stack so that the child
98 * ready to run and return to user mode.
99 */
100void
bb3cd951 101cpu_fork(struct lwp *lp1, struct lwp *lp2, int flags)
984263bc
MD
102{
103 struct pcb *pcb2;
104
105 if ((flags & RFPROC) == 0) {
984263bc
MD
106 if ((flags & RFMEM) == 0) {
107 /* unshare user LDT */
bb3cd951 108 struct pcb *pcb1 = lp1->lwp_thread->td_pcb;
984263bc
MD
109 struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt;
110 if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) {
111 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len);
112 user_ldt_free(pcb1);
113 pcb1->pcb_ldt = pcb_ldt;
114 set_user_ldt(pcb1);
115 }
116 }
984263bc
MD
117 return;
118 }
119
120#if NNPX > 0
bb3cd951
SS
121 /* Ensure that lp1's pcb is up to date. */
122 if (mdcpu->gd_npxthread == lp1->lwp_thread)
123 npxsave(lp1->lwp_thread->td_savefpu);
984263bc 124#endif
a96ef868
MD
125
126 /*
bb3cd951 127 * Copy lp1's PCB. This really only applies to the
a96ef868
MD
128 * debug registers and FP state, but its faster to just copy the
129 * whole thing. Because we only save the PCB at switchout time,
4e7c41c5 130 * the register state may not be current.
a96ef868 131 */
bb3cd951
SS
132 pcb2 = lp2->lwp_thread->td_pcb;
133 *pcb2 = *lp1->lwp_thread->td_pcb;
984263bc
MD
134
135 /*
136 * Create a new fresh stack for the new process.
137 * Copy the trap frame for the return to user mode as if from a
b7c628e4
MD
138 * syscall. This copies the user mode register values. The
139 * 16 byte offset saves space for vm86, and must match
140 * common_tss.esp0 (kernel stack pointer on entry from user mode)
8ad65e08
MD
141 *
142 * pcb_esp must allocate an additional call-return pointer below
d9eea1a5
MD
143 * the trap frame which will be restored by cpu_restore from
144 * PCB_EIP, and the thread's td_sp pointer must allocate an
145 * additonal two worsd below the pcb_esp call-return pointer to
146 * hold the LWKT restore function pointer and eflags.
8ad65e08
MD
147 *
148 * The LWKT restore function pointer must be set to cpu_restore,
149 * which is our standard heavy weight process switch-in function.
150 * YYY eventually we should shortcut fork_return and fork_trampoline
151 * to use the LWKT restore function directly so we can get rid of
152 * all the extra crap we are setting up.
984263bc 153 */
bb3cd951
SS
154 lp2->lwp_md.md_regs = (struct trapframe *)((char *)pcb2 - 16) - 1;
155 bcopy(lp1->lwp_md.md_regs, lp2->lwp_md.md_regs, sizeof(*lp2->lwp_md.md_regs));
984263bc
MD
156
157 /*
158 * Set registers for trampoline to user mode. Leave space for the
159 * return address on stack. These are the kernel mode register values.
160 */
bb3cd951 161 pcb2->pcb_cr3 = vtophys(vmspace_pmap(lp2->lwp_proc->p_vmspace)->pm_pdir);
984263bc
MD
162 pcb2->pcb_edi = 0;
163 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */
164 pcb2->pcb_ebp = 0;
bb3cd951
SS
165 pcb2->pcb_esp = (int)lp2->lwp_md.md_regs - sizeof(void *);
166 pcb2->pcb_ebx = (int)lp2; /* fork_trampoline argument */
984263bc 167 pcb2->pcb_eip = (int)fork_trampoline;
bb3cd951
SS
168 lp2->lwp_thread->td_sp = (char *)(pcb2->pcb_esp - sizeof(void *));
169 *(u_int32_t *)lp2->lwp_thread->td_sp = PSL_USER;
170 lp2->lwp_thread->td_sp -= sizeof(void *);
171 *(void **)lp2->lwp_thread->td_sp = (void *)cpu_heavy_restore;
a96ef868
MD
172
173 /*
984263bc
MD
174 * pcb2->pcb_ldt: duplicated below, if necessary.
175 * pcb2->pcb_savefpu: cloned above.
176 * pcb2->pcb_flags: cloned above (always 0 here?).
177 * pcb2->pcb_onfault: cloned above (always NULL here?).
178 */
179
984263bc
MD
180 /*
181 * XXX don't copy the i/o pages. this should probably be fixed.
182 */
183 pcb2->pcb_ext = 0;
184
984263bc
MD
185 /* Copy the LDT, if necessary. */
186 if (pcb2->pcb_ldt != 0) {
187 if (flags & RFMEM) {
188 pcb2->pcb_ldt->ldt_refcnt++;
189 } else {
190 pcb2->pcb_ldt = user_ldt_alloc(pcb2,
191 pcb2->pcb_ldt->ldt_len);
192 }
193 }
bb3cd951
SS
194 bcopy(&lp1->lwp_thread->td_tls, &lp2->lwp_thread->td_tls,
195 sizeof(lp2->lwp_thread->td_tls));
984263bc 196 /*
08f2f1bb 197 * Now, cpu_switch() can schedule the new lwp.
984263bc
MD
198 * pcb_esp is loaded pointing to the cpu_switch() stack frame
199 * containing the return address when exiting cpu_switch.
200 * This will normally be to fork_trampoline(), which will have
08f2f1bb
SS
201 * %ebx loaded with the new lwp's pointer. fork_trampoline()
202 * will set up a stack to call fork_return(lp, frame); to complete
984263bc
MD
203 * the return to user-mode.
204 */
205}
206
207/*
91bd9c1e
SS
208 * Prepare new lwp to return to the address specified in params.
209 */
210int
211cpu_prepare_lwp(struct lwp *lp, struct lwp_params *params)
212{
213 struct trapframe *regs = lp->lwp_md.md_regs;
214 void *bad_return = NULL;
215 int error;
216
29adef62
NT
217 regs->tf_eip = (int)params->func;
218 regs->tf_esp = (int)params->stack;
91bd9c1e
SS
219 /* Set up argument for function call */
220 regs->tf_esp -= sizeof(params->arg);
29adef62
NT
221 error =
222 copyout(&params->arg, (void *)regs->tf_esp, sizeof(params->arg));
91bd9c1e
SS
223 if (error)
224 return (error);
225 /*
226 * Set up fake return address. As the lwp function may never return,
227 * we simply copy out a NULL pointer and force the lwp to receive
228 * a SIGSEGV if it returns anyways.
229 */
230 regs->tf_esp -= sizeof(void *);
29adef62 231 error = copyout(&bad_return, (void *)regs->tf_esp, sizeof(bad_return));
91bd9c1e
SS
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/*
984263bc
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)
984263bc
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 */
bb3cd951
SS
254 lp->lwp_thread->td_pcb->pcb_esi = (int) func; /* function */
255 lp->lwp_thread->td_pcb->pcb_ebx = (int) arg; /* first arg */
984263bc
MD
256}
257
258void
0cfcada1
MD
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)
984263bc 272{
99df837e 273 struct thread *td = curthread;
ae8050a4 274 struct pcb *pcb;
df011816 275 struct pcb_ext *ext;
99df837e 276
ae8050a4 277 /*
df011816
MD
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.
ae8050a4 280 */
99df837e 281 pcb = td->td_pcb;
df011816
MD
282 if ((ext = pcb->pcb_ext) != NULL) {
283 crit_enter();
284 pcb->pcb_ext = NULL;
eb673128 285 lwkt_switch_return(td->td_switch(td));
df011816 286 crit_exit();
e4846942 287 kmem_free(&kernel_map, (vm_offset_t)ext, ctob(IOPAGES + 1));
984263bc 288 }
984263bc 289 user_ldt_free(pcb);
984263bc
MD
290 if (pcb->pcb_flags & PCB_DBREGS) {
291 /*
292 * disable all hardware breakpoints
293 */
294 reset_dbregs();
295 pcb->pcb_flags &= ~PCB_DBREGS;
296 }
37af14fe 297 td->td_gd->gd_cnt.v_swtch++;
ae8050a4 298
37af14fe 299 crit_enter_quick(td);
09b62fa4
MD
300 if (td->td_flags & TDF_TSLEEPQ)
301 tsleep_remove(td);
37af14fe 302 lwkt_deschedule_self(td);
e56e4dea 303 lwkt_remove_tdallq(td);
99df837e
MD
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
99df837e 322 curthread->td_switch = cpu_exit_switch;
c1102e9f 323 curthread->td_flags |= TDF_EXITING;
8ad65e08 324 lwkt_switch();
984263bc
MD
325 panic("cpu_exit");
326}
327
984263bc
MD
328#ifdef notyet
329static void
f123d5a1 330setredzone(u_short *pte, caddr_t vaddr)
984263bc
MD
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 */
6ef943a3 346vm_paddr_t
984263bc
MD
347kvtop(void *addr)
348{
6ef943a3 349 vm_paddr_t pa;
984263bc 350
6ef943a3
MD
351 pa = pmap_kextract((vm_offset_t)addr);
352 if (pa == 0)
984263bc 353 panic("kvtop: zero page frame");
6ef943a3 354 return (pa);
984263bc
MD
355}
356
357/*
358 * Force reset the processor by invalidating the entire address space!
359 */
360
361#ifdef SMP
362static void
f123d5a1 363cpu_reset_proxy(void)
984263bc 364{
984263bc
MD
365 cpu_reset_proxy_active = 1;
366 while (cpu_reset_proxy_active == 1)
367 ; /* Wait for other cpu to disable interupts */
26be20a0 368 kprintf("cpu_reset_proxy: Grabbed mp lock for BSP\n");
984263bc
MD
369 cpu_reset_proxy_active = 3;
370 while (cpu_reset_proxy_active == 3)
371 ; /* Wait for other cpu to enable interrupts */
da23a592 372 stop_cpus(CPUMASK(cpu_reset_proxyid));
26be20a0 373 kprintf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
984263bc
MD
374 DELAY(1000000);
375 cpu_reset_real();
376}
377#endif
378
379void
f123d5a1 380cpu_reset(void)
984263bc
MD
381{
382#ifdef SMP
0f7a3396 383 if (smp_active_mask == 1) {
984263bc
MD
384 cpu_reset_real();
385 /* NOTREACHED */
386 } else {
da23a592 387 cpumask_t map;
984263bc 388 int cnt;
26be20a0 389 kprintf("cpu_reset called on cpu#%d\n",mycpu->gd_cpuid);
984263bc 390
0f7a3396 391 map = mycpu->gd_other_cpus & ~stopped_cpus & smp_active_mask;
984263bc
MD
392
393 if (map != 0) {
26be20a0 394 kprintf("cpu_reset: Stopping other CPUs\n");
984263bc
MD
395 stop_cpus(map); /* Stop all other CPUs */
396 }
397
72740893 398 if (mycpu->gd_cpuid == 0) {
984263bc
MD
399 DELAY(1000000);
400 cpu_reset_real();
401 /* NOTREACHED */
402 } else {
403 /* We are not BSP (CPU #0) */
404
72740893 405 cpu_reset_proxyid = mycpu->gd_cpuid;
984263bc 406 cpustop_restartfunc = cpu_reset_proxy;
26be20a0 407 kprintf("cpu_reset: Restarting BSP\n");
984263bc
MD
408 started_cpus = (1<<0); /* Restart CPU #0 */
409
410 cnt = 0;
411 while (cpu_reset_proxy_active == 0 && cnt < 10000000)
412 cnt++; /* Wait for BSP to announce restart */
413 if (cpu_reset_proxy_active == 0)
26be20a0 414 kprintf("cpu_reset: Failed to restart BSP\n");
984263bc
MD
415 __asm __volatile("cli" : : : "memory");
416 cpu_reset_proxy_active = 2;
417 cnt = 0;
418 while (cpu_reset_proxy_active == 2 && cnt < 10000000)
419 cnt++; /* Do nothing */
420 if (cpu_reset_proxy_active == 2) {
26be20a0 421 kprintf("cpu_reset: BSP did not grab mp lock\n");
984263bc
MD
422 cpu_reset_real(); /* XXX: Bogus ? */
423 }
424 cpu_reset_proxy_active = 4;
425 __asm __volatile("sti" : : : "memory");
426 while (1);
427 /* NOTREACHED */
428 }
429 }
430#else
431 cpu_reset_real();
432#endif
433}
434
435static void
f123d5a1 436cpu_reset_real(void)
984263bc 437{
984263bc
MD
438 /*
439 * Attempt to do a CPU reset via the keyboard controller,
440 * do not turn of the GateA20, as any machine that fails
441 * to do the reset here would then end up in no man's land.
442 */
443
444#if !defined(BROKEN_KEYBOARD_RESET)
445 outb(IO_KBD + 4, 0xFE);
446 DELAY(500000); /* wait 0.5 sec to see if that did it */
26be20a0
SW
447 kprintf("Keyboard reset did not work, attempting CPU shutdown\n");
448 DELAY(1000000); /* wait 1 sec for kprintf to complete */
984263bc 449#endif
984263bc
MD
450 /* force a shutdown by unmapping entire address space ! */
451 bzero((caddr_t) PTD, PAGE_SIZE);
452
453 /* "good night, sweet prince .... <THUNK!>" */
0f7a3396 454 cpu_invltlb();
984263bc
MD
455 /* NOTREACHED */
456 while(1);
457}
458
984263bc
MD
459SYSCTL_DECL(_vm_stats_misc);
460
26a0694b 461static void
477d3c1c 462swi_vm(void *arg, void *frame)
ef0fdad1 463{
984263bc
MD
464 if (busdma_swi_pending != 0)
465 busdma_swi();
466}
467
ef0fdad1
MD
468static void
469swi_vm_setup(void *arg)
470{
1da8d52f 471 register_swi(SWI_VM, swi_vm, NULL, "swi_vm", NULL, 0);
ef0fdad1
MD
472}
473
ba39e2e0 474SYSINIT(vm_setup, SI_BOOT2_MACHDEP, SI_ORDER_ANY, swi_vm_setup, NULL);
ef0fdad1
MD
475
476
984263bc
MD
477/*
478 * Tell whether this address is in some physical memory region.
479 * Currently used by the kernel coredump code in order to avoid
480 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
481 * or other unpredictable behaviour.
482 */
483
984263bc 484int
f123d5a1 485is_physical_memory(vm_offset_t addr)
984263bc
MD
486{
487
488#if NISA > 0
489 /* The ISA ``memory hole''. */
490 if (addr >= 0xa0000 && addr < 0x100000)
491 return 0;
492#endif
493
494 /*
495 * stuff other tests for known memory-mapped devices (PCI?)
496 * here
497 */
498
499 return 1;
500}
135d7199
MD
501
502/*
503 * platform-specific vmspace initialization (nothing for i386)
504 */
505void
506cpu_vmspace_alloc(struct vmspace *vm __unused)
507{
508}
509
510void
511cpu_vmspace_free(struct vmspace *vm __unused)
512{
513}
460426e6
MD
514
515/*
516 * Used by /dev/kmem to determine if we can safely read or write
517 * the requested KVA range.
518 */
519int
520kvm_access_check(vm_offset_t saddr, vm_offset_t eaddr, int prot)
521{
522 vm_offset_t addr;
523
524 if (saddr < KvaStart)
525 return EFAULT;
526 if (eaddr >= KvaEnd)
527 return EFAULT;
528 for (addr = saddr; addr < eaddr; addr += PAGE_SIZE) {
529 if (pmap_extract(&kernel_pmap, addr) == 0)
530 return EFAULT;
531 }
532 if (!kernacc((caddr_t)saddr, eaddr - saddr, prot))
533 return EFAULT;
534 return 0;
535}
536