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("$FreeBSD$");
45
46 #include "opt_isa.h"
47 #include "opt_cpu.h"
48 #include "opt_compat.h"
49
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/bio.h>
53 #include <sys/buf.h>
54 #include <sys/kse.h>
55 #include <sys/kernel.h>
56 #include <sys/ktr.h>
57 #include <sys/lock.h>
58 #include <sys/malloc.h>
59 #include <sys/mbuf.h>
60 #include <sys/mutex.h>
61 #include <sys/pioctl.h>
62 #include <sys/proc.h>
63 #include <sys/sf_buf.h>
64 #include <sys/smp.h>
65 #include <sys/sysctl.h>
66 #include <sys/unistd.h>
67 #include <sys/vnode.h>
68 #include <sys/vmmeter.h>
69
70 #include <machine/cpu.h>
71 #include <machine/md_var.h>
72 #include <machine/pcb.h>
73 #include <machine/specialreg.h>
74
75 #include <vm/vm.h>
76 #include <vm/vm_extern.h>
77 #include <vm/vm_kern.h>
78 #include <vm/vm_page.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_param.h>
81
82 #include <amd64/isa/isa.h>
83
84 #ifdef COMPAT_IA32
85
86 extern struct sysentvec ia32_freebsd_sysvec;
87
88 #endif
89
90 static void cpu_reset_real(void);
91 #ifdef SMP
92 static void cpu_reset_proxy(void);
93 static u_int cpu_reset_proxyid;
94 static volatile u_int cpu_reset_proxy_active;
95 #endif
96
97 /*
98 * Finish a fork operation, with process p2 nearly set up.
99 * Copy and update the pcb, set up the stack so that the child
100 * ready to run and return to user mode.
101 */
102 void
103 cpu_fork(td1, p2, td2, flags)
104 register struct thread *td1;
105 register struct proc *p2;
106 struct thread *td2;
107 int flags;
108 {
109 register struct proc *p1;
110 struct pcb *pcb2;
111 struct mdproc *mdp2;
112 pmap_t pmap2;
113
114 p1 = td1->td_proc;
115 if ((flags & RFPROC) == 0)
116 return;
117
118 /* Ensure that p1's pcb is up to date. */
119 fpuexit(td1);
120
121 /* Point the pcb to the top of the stack */
122 pcb2 = (struct pcb *)(td2->td_kstack +
123 td2->td_kstack_pages * PAGE_SIZE) - 1;
124 td2->td_pcb = pcb2;
125
126 /* Copy p1's pcb */
127 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
128
129 /* Point mdproc and then copy over td1's contents */
130 mdp2 = &p2->p_md;
131 bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
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
136 * syscall. This copies most of the user mode register values.
137 */
138 td2->td_frame = (struct trapframe *)td2->td_pcb - 1;
139 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
140
141 td2->td_frame->tf_rax = 0; /* Child returns zero */
142 td2->td_frame->tf_rflags &= ~PSL_C; /* success */
143 td2->td_frame->tf_rdx = 1;
144
145 /*
146 * If the parent process has the trap bit set (i.e. a debugger had
147 * single stepped the process to the system call), we need to clear
148 * the trap flag from the new frame unless the debugger had set PF_FORK
149 * on the parent. Otherwise, the child will receive a (likely
150 * unexpected) SIGTRAP when it executes the first instruction after
151 * returning to userland.
152 */
153 if ((p1->p_pfsflags & PF_FORK) == 0)
154 td2->td_frame->tf_rflags &= ~PSL_T;
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 */
160 pmap2 = vmspace_pmap(p2->p_vmspace);
161 pcb2->pcb_cr3 = DMAP_TO_PHYS((vm_offset_t)pmap2->pm_pml4);
162 pcb2->pcb_r12 = (register_t)fork_return; /* fork_trampoline argument */
163 pcb2->pcb_rbp = 0;
164 pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *);
165 pcb2->pcb_rbx = (register_t)td2; /* fork_trampoline argument */
166 pcb2->pcb_rip = (register_t)fork_trampoline;
167 /*-
168 * pcb2->pcb_dr*: cloned above.
169 * pcb2->pcb_savefpu: cloned above.
170 * pcb2->pcb_flags: cloned above.
171 * pcb2->pcb_onfault: cloned above (always NULL here?).
172 * pcb2->pcb_[fg]sbase: cloned above
173 */
174
175 /* Setup to release spin count in fork_exit(). */
176 td2->td_md.md_spinlock_count = 1;
177 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
178
179 /*
180 * Now, cpu_switch() can schedule the new process.
181 * pcb_rsp is loaded pointing to the cpu_switch() stack frame
182 * containing the return address when exiting cpu_switch.
183 * This will normally be to fork_trampoline(), which will have
184 * %ebx loaded with the new proc's pointer. fork_trampoline()
185 * will set up a stack to call fork_return(p, frame); to complete
186 * the return to user-mode.
187 */
188 }
189
190 /*
191 * Intercept the return address from a freshly forked process that has NOT
192 * been scheduled yet.
193 *
194 * This is needed to make kernel threads stay in kernel mode.
195 */
196 void
197 cpu_set_fork_handler(td, func, arg)
198 struct thread *td;
199 void (*func)(void *);
200 void *arg;
201 {
202 /*
203 * Note that the trap frame follows the args, so the function
204 * is really called like this: func(arg, frame);
205 */
206 td->td_pcb->pcb_r12 = (long) func; /* function */
207 td->td_pcb->pcb_rbx = (long) arg; /* first arg */
208 }
209
210 void
211 cpu_exit(struct thread *td)
212 {
213 }
214
215 void
216 cpu_thread_exit(struct thread *td)
217 {
218
219 if (td == PCPU_GET(fpcurthread))
220 fpudrop();
221
222 /* Disable any hardware breakpoints. */
223 if (td->td_pcb->pcb_flags & PCB_DBREGS) {
224 reset_dbregs();
225 td->td_pcb->pcb_flags &= ~PCB_DBREGS;
226 }
227 }
228
229 void
230 cpu_thread_clean(struct thread *td)
231 {
232 }
233
234 void
235 cpu_thread_swapin(struct thread *td)
236 {
237 }
238
239 void
240 cpu_thread_swapout(struct thread *td)
241 {
242 }
243
244 void
245 cpu_thread_alloc(struct thread *td)
246 {
247
248 td->td_pcb = (struct pcb *)(td->td_kstack +
249 td->td_kstack_pages * PAGE_SIZE) - 1;
250 td->td_frame = (struct trapframe *)td->td_pcb - 1;
251 }
252
253 void
254 cpu_thread_free(struct thread *td)
255 {
256 }
257
258 /*
259 * Initialize machine state (pcb and trap frame) for a new thread about to
260 * upcall. Put enough state in the new thread's PCB to get it to go back
261 * userret(), where we can intercept it again to set the return (upcall)
262 * Address and stack, along with those from upcals that are from other sources
263 * such as those generated in thread_userret() itself.
264 */
265 void
266 cpu_set_upcall(struct thread *td, struct thread *td0)
267 {
268 struct pcb *pcb2;
269
270 /* Point the pcb to the top of the stack. */
271 pcb2 = td->td_pcb;
272
273 /*
274 * Copy the upcall pcb. This loads kernel regs.
275 * Those not loaded individually below get their default
276 * values here.
277 *
278 * XXXKSE It might be a good idea to simply skip this as
279 * the values of the other registers may be unimportant.
280 * This would remove any requirement for knowing the KSE
281 * at this time (see the matching comment below for
282 * more analysis) (need a good safe default).
283 */
284 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
285 pcb2->pcb_flags &= ~PCB_FPUINITDONE;
286
287 /*
288 * Create a new fresh stack for the new thread.
289 * Don't forget to set this stack value into whatever supplies
290 * the address for the fault handlers.
291 * The contexts are filled in at the time we actually DO the
292 * upcall as only then do we know which KSE we got.
293 */
294 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
295
296 /*
297 * Set registers for trampoline to user mode. Leave space for the
298 * return address on stack. These are the kernel mode register values.
299 */
300 pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pml4);
301 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */
302 pcb2->pcb_rbp = 0;
303 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */
304 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */
305 pcb2->pcb_rip = (register_t)fork_trampoline;
306 /*
307 * If we didn't copy the pcb, we'd need to do the following registers:
308 * pcb2->pcb_dr*: cloned above.
309 * pcb2->pcb_savefpu: cloned above.
310 * pcb2->pcb_onfault: cloned above (always NULL here?).
311 * pcb2->pcb_[fg]sbase: cloned above
312 */
313
314 /* Setup to release spin count in fork_exit(). */
315 td->td_md.md_spinlock_count = 1;
316 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
317 }
318
319 /*
320 * Set that machine state for performing an upcall that has to
321 * be done in thread_userret() so that those upcalls generated
322 * in thread_userret() itself can be done as well.
323 */
324 void
325 cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg,
326 stack_t *stack)
327 {
328
329 /*
330 * Do any extra cleaning that needs to be done.
331 * The thread may have optional components
332 * that are not present in a fresh thread.
333 * This may be a recycled thread so make it look
334 * as though it's newly allocated.
335 */
336 cpu_thread_clean(td);
337
338 #ifdef COMPAT_IA32
339 if (td->td_proc->p_sysent == &ia32_freebsd_sysvec) {
340 /*
341 * Set the trap frame to point at the beginning of the uts
342 * function.
343 */
344 td->td_frame->tf_rbp = 0;
345 td->td_frame->tf_rsp =
346 (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
347 td->td_frame->tf_rip = (uintptr_t)entry;
348
349 /*
350 * Pass the address of the mailbox for this kse to the uts
351 * function as a parameter on the stack.
352 */
353 suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)),
354 (uint32_t)(uintptr_t)arg);
355
356 return;
357 }
358 #endif
359
360 /*
361 * Set the trap frame to point at the beginning of the uts
362 * function.
363 */
364 td->td_frame->tf_rbp = 0;
365 td->td_frame->tf_rsp =
366 ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f;
367 td->td_frame->tf_rsp -= 8;
368 td->td_frame->tf_rip = (register_t)entry;
369
370 /*
371 * Pass the address of the mailbox for this kse to the uts
372 * function as a parameter on the stack.
373 */
374 td->td_frame->tf_rdi = (register_t)arg;
375 }
376
377 int
378 cpu_set_user_tls(struct thread *td, void *tls_base)
379 {
380
381 if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS)
382 return (EINVAL);
383
384 #ifdef COMPAT_IA32
385 if (td->td_proc->p_sysent == &ia32_freebsd_sysvec) {
386 if (td == curthread) {
387 critical_enter();
388 td->td_pcb->pcb_gsbase = (register_t)tls_base;
389 wrmsr(MSR_KGSBASE, td->td_pcb->pcb_gsbase);
390 critical_exit();
391 } else {
392 td->td_pcb->pcb_gsbase = (register_t)tls_base;
393 }
394 return (0);
395 }
396 #endif
397 if (td == curthread) {
398 critical_enter();
399 td->td_pcb->pcb_fsbase = (register_t)tls_base;
400 wrmsr(MSR_FSBASE, td->td_pcb->pcb_fsbase);
401 critical_exit();
402 } else {
403 td->td_pcb->pcb_fsbase = (register_t)tls_base;
404 }
405 return (0);
406 }
407
408 #ifdef SMP
409 static void
410 cpu_reset_proxy()
411 {
412
413 cpu_reset_proxy_active = 1;
414 while (cpu_reset_proxy_active == 1)
415 ; /* Wait for other cpu to see that we've started */
416 stop_cpus((1<<cpu_reset_proxyid));
417 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
418 DELAY(1000000);
419 cpu_reset_real();
420 }
421 #endif
422
423 void
424 cpu_reset()
425 {
426 #ifdef SMP
427 u_int cnt, map;
428
429 if (smp_active) {
430 map = PCPU_GET(other_cpus) & ~stopped_cpus;
431 if (map != 0) {
432 printf("cpu_reset: Stopping other CPUs\n");
433 stop_cpus(map);
434 }
435
436 if (PCPU_GET(cpuid) != 0) {
437 cpu_reset_proxyid = PCPU_GET(cpuid);
438 cpustop_restartfunc = cpu_reset_proxy;
439 cpu_reset_proxy_active = 0;
440 printf("cpu_reset: Restarting BSP\n");
441
442 /* Restart CPU #0. */
443 atomic_store_rel_int(&started_cpus, 1 << 0);
444
445 cnt = 0;
446 while (cpu_reset_proxy_active == 0 && cnt < 10000000)
447 cnt++; /* Wait for BSP to announce restart */
448 if (cpu_reset_proxy_active == 0)
449 printf("cpu_reset: Failed to restart BSP\n");
450 enable_intr();
451 cpu_reset_proxy_active = 2;
452
453 while (1);
454 /* NOTREACHED */
455 }
456
457 DELAY(1000000);
458 }
459 #endif
460 cpu_reset_real();
461 /* NOTREACHED */
462 }
463
464 static void
465 cpu_reset_real()
466 {
467 struct region_descriptor null_idt;
468 int b;
469
470 disable_intr();
471
472 /*
473 * Attempt to do a CPU reset via the keyboard controller,
474 * do not turn off GateA20, as any machine that fails
475 * to do the reset here would then end up in no man's land.
476 */
477 outb(IO_KBD + 4, 0xFE);
478 DELAY(500000); /* wait 0.5 sec to see if that did it */
479
480 /*
481 * Attempt to force a reset via the Reset Control register at
482 * I/O port 0xcf9. Bit 2 forces a system reset when it
483 * transitions from 0 to 1. Bit 1 selects the type of reset
484 * to attempt: 0 selects a "soft" reset, and 1 selects a
485 * "hard" reset. We try a "hard" reset. The first write sets
486 * bit 1 to select a "hard" reset and clears bit 2. The
487 * second write forces a 0 -> 1 transition in bit 2 to trigger
488 * a reset.
489 */
490 outb(0xcf9, 0x2);
491 outb(0xcf9, 0x6);
492 DELAY(500000); /* wait 0.5 sec to see if that did it */
493
494 /*
495 * Attempt to force a reset via the Fast A20 and Init register
496 * at I/O port 0x92. Bit 1 serves as an alternate A20 gate.
497 * Bit 0 asserts INIT# when set to 1. We are careful to only
498 * preserve bit 1 while setting bit 0. We also must clear bit
499 * 0 before setting it if it isn't already clear.
500 */
501 b = inb(0x92);
502 if (b != 0xff) {
503 if ((b & 0x1) != 0)
504 outb(0x92, b & 0xfe);
505 outb(0x92, b | 0x1);
506 DELAY(500000); /* wait 0.5 sec to see if that did it */
507 }
508
509 printf("No known reset method worked, attempting CPU shutdown\n");
510 DELAY(1000000); /* wait 1 sec for printf to complete */
511
512 /* Wipe the IDT. */
513 null_idt.rd_limit = 0;
514 null_idt.rd_base = 0;
515 lidt(&null_idt);
516
517 /* "good night, sweet prince .... <THUNK!>" */
518 breakpoint();
519
520 /* NOTREACHED */
521 while(1);
522 }
523
524 /*
525 * Allocate an sf_buf for the given vm_page. On this machine, however, there
526 * is no sf_buf object. Instead, an opaque pointer to the given vm_page is
527 * returned.
528 */
529 struct sf_buf *
530 sf_buf_alloc(struct vm_page *m, int pri)
531 {
532
533 return ((struct sf_buf *)m);
534 }
535
536 /*
537 * Free the sf_buf. In fact, do nothing because there are no resources
538 * associated with the sf_buf.
539 */
540 void
541 sf_buf_free(struct sf_buf *sf)
542 {
543 }
544
545 /*
546 * Software interrupt handler for queued VM system processing.
547 */
548 void
549 swi_vm(void *dummy)
550 {
551 if (busdma_swi_pending != 0)
552 busdma_swi();
553 }
554
555 /*
556 * Tell whether this address is in some physical memory region.
557 * Currently used by the kernel coredump code in order to avoid
558 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
559 * or other unpredictable behaviour.
560 */
561
562 int
563 is_physical_memory(vm_paddr_t addr)
564 {
565
566 #ifdef DEV_ISA
567 /* The ISA ``memory hole''. */
568 if (addr >= 0xa0000 && addr < 0x100000)
569 return 0;
570 #endif
571
572 /*
573 * stuff other tests for known memory-mapped devices (PCI?)
574 * here
575 */
576
577 return 1;
578 }
Cache object: 2ecff5bcf22edce0fee70e5ad0c16ac3
|