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 */
42
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD: releng/8.1/sys/i386/i386/vm_machdep.c 200443 2009-12-12 20:06:25Z kib $");
45
46 #include "opt_isa.h"
47 #include "opt_npx.h"
48 #include "opt_reset.h"
49 #include "opt_cpu.h"
50 #include "opt_xbox.h"
51
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/bio.h>
55 #include <sys/buf.h>
56 #include <sys/kernel.h>
57 #include <sys/ktr.h>
58 #include <sys/lock.h>
59 #include <sys/malloc.h>
60 #include <sys/mbuf.h>
61 #include <sys/mutex.h>
62 #include <sys/pioctl.h>
63 #include <sys/proc.h>
64 #include <sys/sysent.h>
65 #include <sys/sf_buf.h>
66 #include <sys/smp.h>
67 #include <sys/sched.h>
68 #include <sys/sysctl.h>
69 #include <sys/unistd.h>
70 #include <sys/vnode.h>
71 #include <sys/vmmeter.h>
72
73 #include <machine/cpu.h>
74 #include <machine/cputypes.h>
75 #include <machine/md_var.h>
76 #include <machine/pcb.h>
77 #include <machine/pcb_ext.h>
78 #include <machine/smp.h>
79 #include <machine/vm86.h>
80
81 #ifdef CPU_ELAN
82 #include <machine/elan_mmcr.h>
83 #endif
84
85 #include <vm/vm.h>
86 #include <vm/vm_extern.h>
87 #include <vm/vm_kern.h>
88 #include <vm/vm_page.h>
89 #include <vm/vm_map.h>
90 #include <vm/vm_param.h>
91
92 #ifdef XEN
93 #include <xen/hypervisor.h>
94 #endif
95 #ifdef PC98
96 #include <pc98/cbus/cbus.h>
97 #else
98 #include <i386/isa/isa.h>
99 #endif
100
101 #ifdef XBOX
102 #include <machine/xbox.h>
103 #endif
104
105 #ifndef NSFBUFS
106 #define NSFBUFS (512 + maxusers * 16)
107 #endif
108
109 static void cpu_reset_real(void);
110 #ifdef SMP
111 static void cpu_reset_proxy(void);
112 static u_int cpu_reset_proxyid;
113 static volatile u_int cpu_reset_proxy_active;
114 #endif
115 static void sf_buf_init(void *arg);
116 SYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL);
117
118 LIST_HEAD(sf_head, sf_buf);
119
120 /*
121 * A hash table of active sendfile(2) buffers
122 */
123 static struct sf_head *sf_buf_active;
124 static u_long sf_buf_hashmask;
125
126 #define SF_BUF_HASH(m) (((m) - vm_page_array) & sf_buf_hashmask)
127
128 static TAILQ_HEAD(, sf_buf) sf_buf_freelist;
129 static u_int sf_buf_alloc_want;
130
131 /*
132 * A lock used to synchronize access to the hash table and free list
133 */
134 static struct mtx sf_buf_lock;
135
136 extern int _ucodesel, _udatasel;
137
138 /*
139 * Finish a fork operation, with process p2 nearly set up.
140 * Copy and update the pcb, set up the stack so that the child
141 * ready to run and return to user mode.
142 */
143 void
144 cpu_fork(td1, p2, td2, flags)
145 register struct thread *td1;
146 register struct proc *p2;
147 struct thread *td2;
148 int flags;
149 {
150 register struct proc *p1;
151 struct pcb *pcb2;
152 struct mdproc *mdp2;
153 #ifdef DEV_NPX
154 register_t savecrit;
155 #endif
156
157 p1 = td1->td_proc;
158 if ((flags & RFPROC) == 0) {
159 if ((flags & RFMEM) == 0) {
160 /* unshare user LDT */
161 struct mdproc *mdp1 = &p1->p_md;
162 struct proc_ldt *pldt, *pldt1;
163
164 mtx_lock_spin(&dt_lock);
165 if ((pldt1 = mdp1->md_ldt) != NULL &&
166 pldt1->ldt_refcnt > 1) {
167 pldt = user_ldt_alloc(mdp1, pldt1->ldt_len);
168 if (pldt == NULL)
169 panic("could not copy LDT");
170 mdp1->md_ldt = pldt;
171 set_user_ldt(mdp1);
172 user_ldt_deref(pldt1);
173 } else
174 mtx_unlock_spin(&dt_lock);
175 }
176 return;
177 }
178
179 /* Ensure that p1's pcb is up to date. */
180 if (td1 == curthread)
181 td1->td_pcb->pcb_gs = rgs();
182 #ifdef DEV_NPX
183 savecrit = intr_disable();
184 if (PCPU_GET(fpcurthread) == td1)
185 npxsave(&td1->td_pcb->pcb_save);
186 intr_restore(savecrit);
187 #endif
188
189 /* Point the pcb to the top of the stack */
190 pcb2 = (struct pcb *)(td2->td_kstack +
191 td2->td_kstack_pages * PAGE_SIZE) - 1;
192 td2->td_pcb = pcb2;
193
194 /* Copy p1's pcb */
195 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
196
197 /* Point mdproc and then copy over td1's contents */
198 mdp2 = &p2->p_md;
199 bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
200
201 /*
202 * Create a new fresh stack for the new process.
203 * Copy the trap frame for the return to user mode as if from a
204 * syscall. This copies most of the user mode register values.
205 * The -16 is so we can expand the trapframe if we go to vm86.
206 */
207 td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1;
208 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
209
210 td2->td_frame->tf_eax = 0; /* Child returns zero */
211 td2->td_frame->tf_eflags &= ~PSL_C; /* success */
212 td2->td_frame->tf_edx = 1;
213
214 /*
215 * If the parent process has the trap bit set (i.e. a debugger had
216 * single stepped the process to the system call), we need to clear
217 * the trap flag from the new frame unless the debugger had set PF_FORK
218 * on the parent. Otherwise, the child will receive a (likely
219 * unexpected) SIGTRAP when it executes the first instruction after
220 * returning to userland.
221 */
222 if ((p1->p_pfsflags & PF_FORK) == 0)
223 td2->td_frame->tf_eflags &= ~PSL_T;
224
225 /*
226 * Set registers for trampoline to user mode. Leave space for the
227 * return address on stack. These are the kernel mode register values.
228 */
229 #ifdef PAE
230 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdpt);
231 #else
232 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
233 #endif
234 pcb2->pcb_edi = 0;
235 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */
236 pcb2->pcb_ebp = 0;
237 pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
238 pcb2->pcb_ebx = (int)td2; /* fork_trampoline argument */
239 pcb2->pcb_eip = (int)fork_trampoline;
240 pcb2->pcb_psl = PSL_KERNEL; /* ints disabled */
241 /*-
242 * pcb2->pcb_dr*: cloned above.
243 * pcb2->pcb_savefpu: cloned above.
244 * pcb2->pcb_flags: cloned above.
245 * pcb2->pcb_onfault: cloned above (always NULL here?).
246 * pcb2->pcb_gs: cloned above.
247 * pcb2->pcb_ext: cleared below.
248 */
249
250 /*
251 * XXX don't copy the i/o pages. this should probably be fixed.
252 */
253 pcb2->pcb_ext = 0;
254
255 /* Copy the LDT, if necessary. */
256 mtx_lock_spin(&dt_lock);
257 if (mdp2->md_ldt != NULL) {
258 if (flags & RFMEM) {
259 mdp2->md_ldt->ldt_refcnt++;
260 } else {
261 mdp2->md_ldt = user_ldt_alloc(mdp2,
262 mdp2->md_ldt->ldt_len);
263 if (mdp2->md_ldt == NULL)
264 panic("could not copy LDT");
265 }
266 }
267 mtx_unlock_spin(&dt_lock);
268
269 /* Setup to release spin count in fork_exit(). */
270 td2->td_md.md_spinlock_count = 1;
271 /*
272 * XXX XEN need to check on PSL_USER is handled
273 */
274 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
275 /*
276 * Now, cpu_switch() can schedule the new process.
277 * pcb_esp is loaded pointing to the cpu_switch() stack frame
278 * containing the return address when exiting cpu_switch.
279 * This will normally be to fork_trampoline(), which will have
280 * %ebx loaded with the new proc's pointer. fork_trampoline()
281 * will set up a stack to call fork_return(p, frame); to complete
282 * the return to user-mode.
283 */
284 }
285
286 /*
287 * Intercept the return address from a freshly forked process that has NOT
288 * been scheduled yet.
289 *
290 * This is needed to make kernel threads stay in kernel mode.
291 */
292 void
293 cpu_set_fork_handler(td, func, arg)
294 struct thread *td;
295 void (*func)(void *);
296 void *arg;
297 {
298 /*
299 * Note that the trap frame follows the args, so the function
300 * is really called like this: func(arg, frame);
301 */
302 td->td_pcb->pcb_esi = (int) func; /* function */
303 td->td_pcb->pcb_ebx = (int) arg; /* first arg */
304 }
305
306 void
307 cpu_exit(struct thread *td)
308 {
309
310 /*
311 * If this process has a custom LDT, release it. Reset pc->pcb_gs
312 * and %gs before we free it in case they refer to an LDT entry.
313 */
314 mtx_lock_spin(&dt_lock);
315 if (td->td_proc->p_md.md_ldt) {
316 td->td_pcb->pcb_gs = _udatasel;
317 load_gs(_udatasel);
318 user_ldt_free(td);
319 } else
320 mtx_unlock_spin(&dt_lock);
321 }
322
323 void
324 cpu_thread_exit(struct thread *td)
325 {
326
327 #ifdef DEV_NPX
328 if (td == PCPU_GET(fpcurthread))
329 npxdrop();
330 #endif
331
332 /* Disable any hardware breakpoints. */
333 if (td->td_pcb->pcb_flags & PCB_DBREGS) {
334 reset_dbregs();
335 td->td_pcb->pcb_flags &= ~PCB_DBREGS;
336 }
337 }
338
339 void
340 cpu_thread_clean(struct thread *td)
341 {
342 struct pcb *pcb;
343
344 pcb = td->td_pcb;
345 if (pcb->pcb_ext != NULL) {
346 /* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */
347 /*
348 * XXX do we need to move the TSS off the allocated pages
349 * before freeing them? (not done here)
350 */
351 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext,
352 ctob(IOPAGES + 1));
353 pcb->pcb_ext = NULL;
354 }
355 }
356
357 void
358 cpu_thread_swapin(struct thread *td)
359 {
360 }
361
362 void
363 cpu_thread_swapout(struct thread *td)
364 {
365 }
366
367 void
368 cpu_thread_alloc(struct thread *td)
369 {
370
371 td->td_pcb = (struct pcb *)(td->td_kstack +
372 td->td_kstack_pages * PAGE_SIZE) - 1;
373 td->td_frame = (struct trapframe *)((caddr_t)td->td_pcb - 16) - 1;
374 td->td_pcb->pcb_ext = NULL;
375 }
376
377 void
378 cpu_thread_free(struct thread *td)
379 {
380
381 cpu_thread_clean(td);
382 }
383
384 void
385 cpu_set_syscall_retval(struct thread *td, int error)
386 {
387
388 switch (error) {
389 case 0:
390 td->td_frame->tf_eax = td->td_retval[0];
391 td->td_frame->tf_edx = td->td_retval[1];
392 td->td_frame->tf_eflags &= ~PSL_C;
393 break;
394
395 case ERESTART:
396 /*
397 * Reconstruct pc, assuming lcall $X,y is 7 bytes, int
398 * 0x80 is 2 bytes. We saved this in tf_err.
399 */
400 td->td_frame->tf_eip -= td->td_frame->tf_err;
401 break;
402
403 case EJUSTRETURN:
404 break;
405
406 default:
407 if (td->td_proc->p_sysent->sv_errsize) {
408 if (error >= td->td_proc->p_sysent->sv_errsize)
409 error = -1; /* XXX */
410 else
411 error = td->td_proc->p_sysent->sv_errtbl[error];
412 }
413 td->td_frame->tf_eax = error;
414 td->td_frame->tf_eflags |= PSL_C;
415 break;
416 }
417 }
418
419 /*
420 * Initialize machine state (pcb and trap frame) for a new thread about to
421 * upcall. Put enough state in the new thread's PCB to get it to go back
422 * userret(), where we can intercept it again to set the return (upcall)
423 * Address and stack, along with those from upcals that are from other sources
424 * such as those generated in thread_userret() itself.
425 */
426 void
427 cpu_set_upcall(struct thread *td, struct thread *td0)
428 {
429 struct pcb *pcb2;
430
431 /* Point the pcb to the top of the stack. */
432 pcb2 = td->td_pcb;
433
434 /*
435 * Copy the upcall pcb. This loads kernel regs.
436 * Those not loaded individually below get their default
437 * values here.
438 */
439 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
440 pcb2->pcb_flags &= ~(PCB_NPXTRAP|PCB_NPXINITDONE);
441
442 /*
443 * Create a new fresh stack for the new thread.
444 */
445 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
446
447 /* If the current thread has the trap bit set (i.e. a debugger had
448 * single stepped the process to the system call), we need to clear
449 * the trap flag from the new frame. Otherwise, the new thread will
450 * receive a (likely unexpected) SIGTRAP when it executes the first
451 * instruction after returning to userland.
452 */
453 td->td_frame->tf_eflags &= ~PSL_T;
454
455 /*
456 * Set registers for trampoline to user mode. Leave space for the
457 * return address on stack. These are the kernel mode register values.
458 */
459 pcb2->pcb_edi = 0;
460 pcb2->pcb_esi = (int)fork_return; /* trampoline arg */
461 pcb2->pcb_ebp = 0;
462 pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */
463 pcb2->pcb_ebx = (int)td; /* trampoline arg */
464 pcb2->pcb_eip = (int)fork_trampoline;
465 pcb2->pcb_psl &= ~(PSL_I); /* interrupts must be disabled */
466 pcb2->pcb_gs = rgs();
467 /*
468 * If we didn't copy the pcb, we'd need to do the following registers:
469 * pcb2->pcb_cr3: cloned above.
470 * pcb2->pcb_dr*: cloned above.
471 * pcb2->pcb_savefpu: cloned above.
472 * pcb2->pcb_flags: cloned above.
473 * pcb2->pcb_onfault: cloned above (always NULL here?).
474 * pcb2->pcb_gs: cloned above.
475 * pcb2->pcb_ext: cleared below.
476 */
477 pcb2->pcb_ext = NULL;
478
479 /* Setup to release spin count in fork_exit(). */
480 td->td_md.md_spinlock_count = 1;
481 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
482 }
483
484 /*
485 * Set that machine state for performing an upcall that has to
486 * be done in thread_userret() so that those upcalls generated
487 * in thread_userret() itself can be done as well.
488 */
489 void
490 cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg,
491 stack_t *stack)
492 {
493
494 /*
495 * Do any extra cleaning that needs to be done.
496 * The thread may have optional components
497 * that are not present in a fresh thread.
498 * This may be a recycled thread so make it look
499 * as though it's newly allocated.
500 */
501 cpu_thread_clean(td);
502
503 /*
504 * Set the trap frame to point at the beginning of the uts
505 * function.
506 */
507 td->td_frame->tf_ebp = 0;
508 td->td_frame->tf_esp =
509 (((int)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
510 td->td_frame->tf_eip = (int)entry;
511
512 /*
513 * Pass the address of the mailbox for this kse to the uts
514 * function as a parameter on the stack.
515 */
516 suword((void *)(td->td_frame->tf_esp + sizeof(void *)),
517 (int)arg);
518 }
519
520 int
521 cpu_set_user_tls(struct thread *td, void *tls_base)
522 {
523 struct segment_descriptor sd;
524 uint32_t base;
525
526 /*
527 * Construct a descriptor and store it in the pcb for
528 * the next context switch. Also store it in the gdt
529 * so that the load of tf_fs into %fs will activate it
530 * at return to userland.
531 */
532 base = (uint32_t)tls_base;
533 sd.sd_lobase = base & 0xffffff;
534 sd.sd_hibase = (base >> 24) & 0xff;
535 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */
536 sd.sd_hilimit = 0xf;
537 sd.sd_type = SDT_MEMRWA;
538 sd.sd_dpl = SEL_UPL;
539 sd.sd_p = 1;
540 sd.sd_xx = 0;
541 sd.sd_def32 = 1;
542 sd.sd_gran = 1;
543 critical_enter();
544 /* set %gs */
545 td->td_pcb->pcb_gsd = sd;
546 if (td == curthread) {
547 PCPU_GET(fsgs_gdt)[1] = sd;
548 load_gs(GSEL(GUGS_SEL, SEL_UPL));
549 }
550 critical_exit();
551 return (0);
552 }
553
554 /*
555 * Convert kernel VA to physical address
556 */
557 vm_paddr_t
558 kvtop(void *addr)
559 {
560 vm_paddr_t pa;
561
562 pa = pmap_kextract((vm_offset_t)addr);
563 if (pa == 0)
564 panic("kvtop: zero page frame");
565 return (pa);
566 }
567
568 #ifdef SMP
569 static void
570 cpu_reset_proxy()
571 {
572
573 cpu_reset_proxy_active = 1;
574 while (cpu_reset_proxy_active == 1)
575 ; /* Wait for other cpu to see that we've started */
576 stop_cpus((1<<cpu_reset_proxyid));
577 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
578 DELAY(1000000);
579 cpu_reset_real();
580 }
581 #endif
582
583 void
584 cpu_reset()
585 {
586 #ifdef XBOX
587 if (arch_i386_is_xbox) {
588 /* Kick the PIC16L, it can reboot the box */
589 pic16l_reboot();
590 for (;;);
591 }
592 #endif
593
594 #ifdef SMP
595 u_int cnt, map;
596
597 if (smp_active) {
598 map = PCPU_GET(other_cpus) & ~stopped_cpus;
599 if (map != 0) {
600 printf("cpu_reset: Stopping other CPUs\n");
601 stop_cpus(map);
602 }
603
604 if (PCPU_GET(cpuid) != 0) {
605 cpu_reset_proxyid = PCPU_GET(cpuid);
606 cpustop_restartfunc = cpu_reset_proxy;
607 cpu_reset_proxy_active = 0;
608 printf("cpu_reset: Restarting BSP\n");
609
610 /* Restart CPU #0. */
611 /* XXX: restart_cpus(1 << 0); */
612 atomic_store_rel_int(&started_cpus, (1 << 0));
613
614 cnt = 0;
615 while (cpu_reset_proxy_active == 0 && cnt < 10000000)
616 cnt++; /* Wait for BSP to announce restart */
617 if (cpu_reset_proxy_active == 0)
618 printf("cpu_reset: Failed to restart BSP\n");
619 enable_intr();
620 cpu_reset_proxy_active = 2;
621
622 while (1);
623 /* NOTREACHED */
624 }
625
626 DELAY(1000000);
627 }
628 #endif
629 cpu_reset_real();
630 /* NOTREACHED */
631 }
632
633 static void
634 cpu_reset_real()
635 {
636 struct region_descriptor null_idt;
637 #ifndef PC98
638 int b;
639 #endif
640
641 disable_intr();
642 #ifdef XEN
643 if (smp_processor_id() == 0)
644 HYPERVISOR_shutdown(SHUTDOWN_reboot);
645 else
646 HYPERVISOR_shutdown(SHUTDOWN_poweroff);
647 #endif
648 #ifdef CPU_ELAN
649 if (elan_mmcr != NULL)
650 elan_mmcr->RESCFG = 1;
651 #endif
652
653 if (cpu == CPU_GEODE1100) {
654 /* Attempt Geode's own reset */
655 outl(0xcf8, 0x80009044ul);
656 outl(0xcfc, 0xf);
657 }
658
659 #ifdef PC98
660 /*
661 * Attempt to do a CPU reset via CPU reset port.
662 */
663 if ((inb(0x35) & 0xa0) != 0xa0) {
664 outb(0x37, 0x0f); /* SHUT0 = 0. */
665 outb(0x37, 0x0b); /* SHUT1 = 0. */
666 }
667 outb(0xf0, 0x00); /* Reset. */
668 #else
669 #if !defined(BROKEN_KEYBOARD_RESET)
670 /*
671 * Attempt to do a CPU reset via the keyboard controller,
672 * do not turn off GateA20, as any machine that fails
673 * to do the reset here would then end up in no man's land.
674 */
675 outb(IO_KBD + 4, 0xFE);
676 DELAY(500000); /* wait 0.5 sec to see if that did it */
677 #endif
678
679 /*
680 * Attempt to force a reset via the Reset Control register at
681 * I/O port 0xcf9. Bit 2 forces a system reset when it
682 * transitions from 0 to 1. Bit 1 selects the type of reset
683 * to attempt: 0 selects a "soft" reset, and 1 selects a
684 * "hard" reset. We try a "hard" reset. The first write sets
685 * bit 1 to select a "hard" reset and clears bit 2. The
686 * second write forces a 0 -> 1 transition in bit 2 to trigger
687 * a reset.
688 */
689 outb(0xcf9, 0x2);
690 outb(0xcf9, 0x6);
691 DELAY(500000); /* wait 0.5 sec to see if that did it */
692
693 /*
694 * Attempt to force a reset via the Fast A20 and Init register
695 * at I/O port 0x92. Bit 1 serves as an alternate A20 gate.
696 * Bit 0 asserts INIT# when set to 1. We are careful to only
697 * preserve bit 1 while setting bit 0. We also must clear bit
698 * 0 before setting it if it isn't already clear.
699 */
700 b = inb(0x92);
701 if (b != 0xff) {
702 if ((b & 0x1) != 0)
703 outb(0x92, b & 0xfe);
704 outb(0x92, b | 0x1);
705 DELAY(500000); /* wait 0.5 sec to see if that did it */
706 }
707 #endif /* PC98 */
708
709 printf("No known reset method worked, attempting CPU shutdown\n");
710 DELAY(1000000); /* wait 1 sec for printf to complete */
711
712 /* Wipe the IDT. */
713 null_idt.rd_limit = 0;
714 null_idt.rd_base = 0;
715 lidt(&null_idt);
716
717 /* "good night, sweet prince .... <THUNK!>" */
718 breakpoint();
719
720 /* NOTREACHED */
721 while(1);
722 }
723
724 /*
725 * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-))
726 */
727 static void
728 sf_buf_init(void *arg)
729 {
730 struct sf_buf *sf_bufs;
731 vm_offset_t sf_base;
732 int i;
733
734 nsfbufs = NSFBUFS;
735 TUNABLE_INT_FETCH("kern.ipc.nsfbufs", &nsfbufs);
736
737 sf_buf_active = hashinit(nsfbufs, M_TEMP, &sf_buf_hashmask);
738 TAILQ_INIT(&sf_buf_freelist);
739 sf_base = kmem_alloc_nofault(kernel_map, nsfbufs * PAGE_SIZE);
740 sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP,
741 M_NOWAIT | M_ZERO);
742 for (i = 0; i < nsfbufs; i++) {
743 sf_bufs[i].kva = sf_base + i * PAGE_SIZE;
744 TAILQ_INSERT_TAIL(&sf_buf_freelist, &sf_bufs[i], free_entry);
745 }
746 sf_buf_alloc_want = 0;
747 mtx_init(&sf_buf_lock, "sf_buf", NULL, MTX_DEF);
748 }
749
750 /*
751 * Invalidate the cache lines that may belong to the page, if
752 * (possibly old) mapping of the page by sf buffer exists. Returns
753 * TRUE when mapping was found and cache invalidated.
754 */
755 boolean_t
756 sf_buf_invalidate_cache(vm_page_t m)
757 {
758 struct sf_head *hash_list;
759 struct sf_buf *sf;
760 boolean_t ret;
761
762 hash_list = &sf_buf_active[SF_BUF_HASH(m)];
763 ret = FALSE;
764 mtx_lock(&sf_buf_lock);
765 LIST_FOREACH(sf, hash_list, list_entry) {
766 if (sf->m == m) {
767 /*
768 * Use pmap_qenter to update the pte for
769 * existing mapping, in particular, the PAT
770 * settings are recalculated.
771 */
772 pmap_qenter(sf->kva, &m, 1);
773 pmap_invalidate_cache_range(sf->kva, sf->kva +
774 PAGE_SIZE);
775 ret = TRUE;
776 break;
777 }
778 }
779 mtx_unlock(&sf_buf_lock);
780 return (ret);
781 }
782
783 /*
784 * Get an sf_buf from the freelist. May block if none are available.
785 */
786 struct sf_buf *
787 sf_buf_alloc(struct vm_page *m, int flags)
788 {
789 pt_entry_t opte, *ptep;
790 struct sf_head *hash_list;
791 struct sf_buf *sf;
792 #ifdef SMP
793 cpumask_t cpumask, other_cpus;
794 #endif
795 int error;
796
797 KASSERT(curthread->td_pinned > 0 || (flags & SFB_CPUPRIVATE) == 0,
798 ("sf_buf_alloc(SFB_CPUPRIVATE): curthread not pinned"));
799 hash_list = &sf_buf_active[SF_BUF_HASH(m)];
800 mtx_lock(&sf_buf_lock);
801 LIST_FOREACH(sf, hash_list, list_entry) {
802 if (sf->m == m) {
803 sf->ref_count++;
804 if (sf->ref_count == 1) {
805 TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
806 nsfbufsused++;
807 nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
808 }
809 #ifdef SMP
810 goto shootdown;
811 #else
812 goto done;
813 #endif
814 }
815 }
816 while ((sf = TAILQ_FIRST(&sf_buf_freelist)) == NULL) {
817 if (flags & SFB_NOWAIT)
818 goto done;
819 sf_buf_alloc_want++;
820 mbstat.sf_allocwait++;
821 error = msleep(&sf_buf_freelist, &sf_buf_lock,
822 (flags & SFB_CATCH) ? PCATCH | PVM : PVM, "sfbufa", 0);
823 sf_buf_alloc_want--;
824
825 /*
826 * If we got a signal, don't risk going back to sleep.
827 */
828 if (error)
829 goto done;
830 }
831 TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
832 if (sf->m != NULL)
833 LIST_REMOVE(sf, list_entry);
834 LIST_INSERT_HEAD(hash_list, sf, list_entry);
835 sf->ref_count = 1;
836 sf->m = m;
837 nsfbufsused++;
838 nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
839
840 /*
841 * Update the sf_buf's virtual-to-physical mapping, flushing the
842 * virtual address from the TLB. Since the reference count for
843 * the sf_buf's old mapping was zero, that mapping is not
844 * currently in use. Consequently, there is no need to exchange
845 * the old and new PTEs atomically, even under PAE.
846 */
847 ptep = vtopte(sf->kva);
848 opte = *ptep;
849 #ifdef XEN
850 PT_SET_MA(sf->kva, xpmap_ptom(VM_PAGE_TO_PHYS(m)) | pgeflag
851 | PG_RW | PG_V | pmap_cache_bits(m->md.pat_mode, 0));
852 #else
853 *ptep = VM_PAGE_TO_PHYS(m) | pgeflag | PG_RW | PG_V |
854 pmap_cache_bits(m->md.pat_mode, 0);
855 #endif
856
857 /*
858 * Avoid unnecessary TLB invalidations: If the sf_buf's old
859 * virtual-to-physical mapping was not used, then any processor
860 * that has invalidated the sf_buf's virtual address from its TLB
861 * since the last used mapping need not invalidate again.
862 */
863 #ifdef SMP
864 if ((opte & (PG_V | PG_A)) == (PG_V | PG_A))
865 sf->cpumask = 0;
866 shootdown:
867 sched_pin();
868 cpumask = PCPU_GET(cpumask);
869 if ((sf->cpumask & cpumask) == 0) {
870 sf->cpumask |= cpumask;
871 invlpg(sf->kva);
872 }
873 if ((flags & SFB_CPUPRIVATE) == 0) {
874 other_cpus = PCPU_GET(other_cpus) & ~sf->cpumask;
875 if (other_cpus != 0) {
876 sf->cpumask |= other_cpus;
877 smp_masked_invlpg(other_cpus, sf->kva);
878 }
879 }
880 sched_unpin();
881 #else
882 if ((opte & (PG_V | PG_A)) == (PG_V | PG_A))
883 pmap_invalidate_page(kernel_pmap, sf->kva);
884 #endif
885 done:
886 mtx_unlock(&sf_buf_lock);
887 return (sf);
888 }
889
890 /*
891 * Remove a reference from the given sf_buf, adding it to the free
892 * list when its reference count reaches zero. A freed sf_buf still,
893 * however, retains its virtual-to-physical mapping until it is
894 * recycled or reactivated by sf_buf_alloc(9).
895 */
896 void
897 sf_buf_free(struct sf_buf *sf)
898 {
899
900 mtx_lock(&sf_buf_lock);
901 sf->ref_count--;
902 if (sf->ref_count == 0) {
903 TAILQ_INSERT_TAIL(&sf_buf_freelist, sf, free_entry);
904 nsfbufsused--;
905 #ifdef XEN
906 /*
907 * Xen doesn't like having dangling R/W mappings
908 */
909 pmap_qremove(sf->kva, 1);
910 sf->m = NULL;
911 LIST_REMOVE(sf, list_entry);
912 #endif
913 if (sf_buf_alloc_want > 0)
914 wakeup_one(&sf_buf_freelist);
915 }
916 mtx_unlock(&sf_buf_lock);
917 }
918
919 /*
920 * Software interrupt handler for queued VM system processing.
921 */
922 void
923 swi_vm(void *dummy)
924 {
925 if (busdma_swi_pending != 0)
926 busdma_swi();
927 }
928
929 /*
930 * Tell whether this address is in some physical memory region.
931 * Currently used by the kernel coredump code in order to avoid
932 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
933 * or other unpredictable behaviour.
934 */
935
936 int
937 is_physical_memory(vm_paddr_t addr)
938 {
939
940 #ifdef DEV_ISA
941 /* The ISA ``memory hole''. */
942 if (addr >= 0xa0000 && addr < 0x100000)
943 return 0;
944 #endif
945
946 /*
947 * stuff other tests for known memory-mapped devices (PCI?)
948 * here
949 */
950
951 return 1;
952 }
Cache object: e1fa6bb417405f8d482b5e131f26bda3
|