1 /*-
2 * SPDX-License-Identifier: BSD-4-Clause
3 *
4 * Copyright (c) 1982, 1986 The Regents of the University of California.
5 * Copyright (c) 1989, 1990 William Jolitz
6 * Copyright (c) 1994 John Dyson
7 * All rights reserved.
8 *
9 * This code is derived from software contributed to Berkeley by
10 * the Systems Programming Group of the University of Utah Computer
11 * Science Department, and William Jolitz.
12 *
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. All advertising materials mentioning features or use of this software
22 * must display the following acknowledgement:
23 * This product includes software developed by the University of
24 * California, Berkeley and its contributors.
25 * 4. Neither the name of the University nor the names of its contributors
26 * may be used to endorse or promote products derived from this software
27 * without specific prior written permission.
28 *
29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 * SUCH DAMAGE.
40 *
41 * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91
42 * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
43 */
44
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD: releng/12.0/sys/amd64/amd64/vm_machdep.c 339349 2018-10-13 21:18:31Z mjg $");
47
48 #include "opt_isa.h"
49 #include "opt_cpu.h"
50
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/bio.h>
54 #include <sys/buf.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/smp.h>
64 #include <sys/sysctl.h>
65 #include <sys/sysent.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/smp.h>
74 #include <machine/specialreg.h>
75 #include <machine/tss.h>
76
77 #include <vm/vm.h>
78 #include <vm/vm_extern.h>
79 #include <vm/vm_kern.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_param.h>
83
84 _Static_assert(OFFSETOF_CURTHREAD == offsetof(struct pcpu, pc_curthread),
85 "OFFSETOF_CURTHREAD does not correspond with offset of pc_curthread.");
86 _Static_assert(OFFSETOF_CURPCB == offsetof(struct pcpu, pc_curpcb),
87 "OFFSETOF_CURPCB does not correspond with offset of pc_curpcb.");
88 _Static_assert(OFFSETOF_MONITORBUF == offsetof(struct pcpu, pc_monitorbuf),
89 "OFFSETOF_MONINORBUF does not correspond with offset of pc_monitorbuf.");
90
91 struct savefpu *
92 get_pcb_user_save_td(struct thread *td)
93 {
94 vm_offset_t p;
95
96 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
97 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN);
98 KASSERT((p % XSAVE_AREA_ALIGN) == 0, ("Unaligned pcb_user_save area"));
99 return ((struct savefpu *)p);
100 }
101
102 struct savefpu *
103 get_pcb_user_save_pcb(struct pcb *pcb)
104 {
105 vm_offset_t p;
106
107 p = (vm_offset_t)(pcb + 1);
108 return ((struct savefpu *)p);
109 }
110
111 struct pcb *
112 get_pcb_td(struct thread *td)
113 {
114 vm_offset_t p;
115
116 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
117 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN) -
118 sizeof(struct pcb);
119 return ((struct pcb *)p);
120 }
121
122 void *
123 alloc_fpusave(int flags)
124 {
125 void *res;
126 struct savefpu_ymm *sf;
127
128 res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags);
129 if (use_xsave) {
130 sf = (struct savefpu_ymm *)res;
131 bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd));
132 sf->sv_xstate.sx_hd.xstate_bv = xsave_mask;
133 }
134 return (res);
135 }
136
137 /*
138 * Finish a fork operation, with process p2 nearly set up.
139 * Copy and update the pcb, set up the stack so that the child
140 * ready to run and return to user mode.
141 */
142 void
143 cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
144 {
145 struct proc *p1;
146 struct pcb *pcb2;
147 struct mdproc *mdp1, *mdp2;
148 struct proc_ldt *pldt;
149
150 p1 = td1->td_proc;
151 if ((flags & RFPROC) == 0) {
152 if ((flags & RFMEM) == 0) {
153 /* unshare user LDT */
154 mdp1 = &p1->p_md;
155 mtx_lock(&dt_lock);
156 if ((pldt = mdp1->md_ldt) != NULL &&
157 pldt->ldt_refcnt > 1 &&
158 user_ldt_alloc(p1, 1) == NULL)
159 panic("could not copy LDT");
160 mtx_unlock(&dt_lock);
161 }
162 return;
163 }
164
165 /* Ensure that td1's pcb is up to date. */
166 fpuexit(td1);
167 update_pcb_bases(td1->td_pcb);
168
169 /* Point the pcb to the top of the stack */
170 pcb2 = get_pcb_td(td2);
171 td2->td_pcb = pcb2;
172
173 /* Copy td1's pcb */
174 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
175
176 /* Properly initialize pcb_save */
177 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
178 bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2),
179 cpu_max_ext_state_size);
180
181 /* Point mdproc and then copy over td1's contents */
182 mdp2 = &p2->p_md;
183 bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
184
185 /*
186 * Create a new fresh stack for the new process.
187 * Copy the trap frame for the return to user mode as if from a
188 * syscall. This copies most of the user mode register values.
189 */
190 td2->td_frame = (struct trapframe *)td2->td_pcb - 1;
191 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
192
193 td2->td_frame->tf_rax = 0; /* Child returns zero */
194 td2->td_frame->tf_rflags &= ~PSL_C; /* success */
195 td2->td_frame->tf_rdx = 1;
196
197 /*
198 * If the parent process has the trap bit set (i.e. a debugger had
199 * single stepped the process to the system call), we need to clear
200 * the trap flag from the new frame unless the debugger had set PF_FORK
201 * on the parent. Otherwise, the child will receive a (likely
202 * unexpected) SIGTRAP when it executes the first instruction after
203 * returning to userland.
204 */
205 if ((p1->p_pfsflags & PF_FORK) == 0)
206 td2->td_frame->tf_rflags &= ~PSL_T;
207
208 /*
209 * Set registers for trampoline to user mode. Leave space for the
210 * return address on stack. These are the kernel mode register values.
211 */
212 pcb2->pcb_r12 = (register_t)fork_return; /* fork_trampoline argument */
213 pcb2->pcb_rbp = 0;
214 pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *);
215 pcb2->pcb_rbx = (register_t)td2; /* fork_trampoline argument */
216 pcb2->pcb_rip = (register_t)fork_trampoline;
217 /*-
218 * pcb2->pcb_dr*: cloned above.
219 * pcb2->pcb_savefpu: cloned above.
220 * pcb2->pcb_flags: cloned above.
221 * pcb2->pcb_onfault: cloned above (always NULL here?).
222 * pcb2->pcb_[fg]sbase: cloned above
223 */
224
225 /* Setup to release spin count in fork_exit(). */
226 td2->td_md.md_spinlock_count = 1;
227 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
228 td2->td_md.md_invl_gen.gen = 0;
229
230 /* As an i386, do not copy io permission bitmap. */
231 pcb2->pcb_tssp = NULL;
232
233 /* New segment registers. */
234 set_pcb_flags_raw(pcb2, PCB_FULL_IRET);
235
236 /* Copy the LDT, if necessary. */
237 mdp1 = &td1->td_proc->p_md;
238 mdp2 = &p2->p_md;
239 if (mdp1->md_ldt == NULL) {
240 mdp2->md_ldt = NULL;
241 return;
242 }
243 mtx_lock(&dt_lock);
244 if (mdp1->md_ldt != NULL) {
245 if (flags & RFMEM) {
246 mdp1->md_ldt->ldt_refcnt++;
247 mdp2->md_ldt = mdp1->md_ldt;
248 bcopy(&mdp1->md_ldt_sd, &mdp2->md_ldt_sd, sizeof(struct
249 system_segment_descriptor));
250 } else {
251 mdp2->md_ldt = NULL;
252 mdp2->md_ldt = user_ldt_alloc(p2, 0);
253 if (mdp2->md_ldt == NULL)
254 panic("could not copy LDT");
255 amd64_set_ldt_data(td2, 0, max_ldt_segment,
256 (struct user_segment_descriptor *)
257 mdp1->md_ldt->ldt_base);
258 }
259 } else
260 mdp2->md_ldt = NULL;
261 mtx_unlock(&dt_lock);
262
263 /*
264 * Now, cpu_switch() can schedule the new process.
265 * pcb_rsp is loaded pointing to the cpu_switch() stack frame
266 * containing the return address when exiting cpu_switch.
267 * This will normally be to fork_trampoline(), which will have
268 * %ebx loaded with the new proc's pointer. fork_trampoline()
269 * will set up a stack to call fork_return(p, frame); to complete
270 * the return to user-mode.
271 */
272 }
273
274 /*
275 * Intercept the return address from a freshly forked process that has NOT
276 * been scheduled yet.
277 *
278 * This is needed to make kernel threads stay in kernel mode.
279 */
280 void
281 cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg)
282 {
283 /*
284 * Note that the trap frame follows the args, so the function
285 * is really called like this: func(arg, frame);
286 */
287 td->td_pcb->pcb_r12 = (long) func; /* function */
288 td->td_pcb->pcb_rbx = (long) arg; /* first arg */
289 }
290
291 void
292 cpu_exit(struct thread *td)
293 {
294
295 /*
296 * If this process has a custom LDT, release it.
297 */
298 if (td->td_proc->p_md.md_ldt != NULL)
299 user_ldt_free(td);
300 }
301
302 void
303 cpu_thread_exit(struct thread *td)
304 {
305 struct pcb *pcb;
306
307 critical_enter();
308 if (td == PCPU_GET(fpcurthread))
309 fpudrop();
310 critical_exit();
311
312 pcb = td->td_pcb;
313
314 /* Disable any hardware breakpoints. */
315 if (pcb->pcb_flags & PCB_DBREGS) {
316 reset_dbregs();
317 clear_pcb_flags(pcb, PCB_DBREGS);
318 }
319 }
320
321 void
322 cpu_thread_clean(struct thread *td)
323 {
324 struct pcb *pcb;
325
326 pcb = td->td_pcb;
327
328 /*
329 * Clean TSS/iomap
330 */
331 if (pcb->pcb_tssp != NULL) {
332 pmap_pti_remove_kva((vm_offset_t)pcb->pcb_tssp,
333 (vm_offset_t)pcb->pcb_tssp + ctob(IOPAGES + 1));
334 kmem_free((vm_offset_t)pcb->pcb_tssp, ctob(IOPAGES + 1));
335 pcb->pcb_tssp = NULL;
336 }
337 }
338
339 void
340 cpu_thread_swapin(struct thread *td)
341 {
342 }
343
344 void
345 cpu_thread_swapout(struct thread *td)
346 {
347 }
348
349 void
350 cpu_thread_alloc(struct thread *td)
351 {
352 struct pcb *pcb;
353 struct xstate_hdr *xhdr;
354
355 td->td_pcb = pcb = get_pcb_td(td);
356 td->td_frame = (struct trapframe *)pcb - 1;
357 pcb->pcb_save = get_pcb_user_save_pcb(pcb);
358 if (use_xsave) {
359 xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
360 bzero(xhdr, sizeof(*xhdr));
361 xhdr->xstate_bv = xsave_mask;
362 }
363 }
364
365 void
366 cpu_thread_free(struct thread *td)
367 {
368
369 cpu_thread_clean(td);
370 }
371
372 void
373 cpu_set_syscall_retval(struct thread *td, int error)
374 {
375 struct trapframe *frame;
376
377 frame = td->td_frame;
378 if (__predict_true(error == 0)) {
379 frame->tf_rax = td->td_retval[0];
380 frame->tf_rdx = td->td_retval[1];
381 frame->tf_rflags &= ~PSL_C;
382 return;
383 }
384
385 switch (error) {
386 case ERESTART:
387 /*
388 * Reconstruct pc, we know that 'syscall' is 2 bytes,
389 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
390 * We saved this in tf_err.
391 * %r10 (which was holding the value of %rcx) is restored
392 * for the next iteration.
393 * %r10 restore is only required for freebsd/amd64 processes,
394 * but shall be innocent for any ia32 ABI.
395 *
396 * Require full context restore to get the arguments
397 * in the registers reloaded at return to usermode.
398 */
399 frame->tf_rip -= frame->tf_err;
400 frame->tf_r10 = frame->tf_rcx;
401 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
402 break;
403
404 case EJUSTRETURN:
405 break;
406
407 default:
408 frame->tf_rax = SV_ABI_ERRNO(td->td_proc, error);
409 frame->tf_rflags |= PSL_C;
410 break;
411 }
412 }
413
414 /*
415 * Initialize machine state, mostly pcb and trap frame for a new
416 * thread, about to return to userspace. Put enough state in the new
417 * thread's PCB to get it to go back to the fork_return(), which
418 * finalizes the thread state and handles peculiarities of the first
419 * return to userspace for the new thread.
420 */
421 void
422 cpu_copy_thread(struct thread *td, struct thread *td0)
423 {
424 struct pcb *pcb2;
425
426 /* Point the pcb to the top of the stack. */
427 pcb2 = td->td_pcb;
428
429 /*
430 * Copy the upcall pcb. This loads kernel regs.
431 * Those not loaded individually below get their default
432 * values here.
433 */
434 update_pcb_bases(td0->td_pcb);
435 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
436 clear_pcb_flags(pcb2, PCB_FPUINITDONE | PCB_USERFPUINITDONE |
437 PCB_KERNFPU);
438 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
439 bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save,
440 cpu_max_ext_state_size);
441 set_pcb_flags_raw(pcb2, PCB_FULL_IRET);
442
443 /*
444 * Create a new fresh stack for the new thread.
445 */
446 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
447
448 /* If the current thread has the trap bit set (i.e. a debugger had
449 * single stepped the process to the system call), we need to clear
450 * the trap flag from the new frame. Otherwise, the new thread will
451 * receive a (likely unexpected) SIGTRAP when it executes the first
452 * instruction after returning to userland.
453 */
454 td->td_frame->tf_rflags &= ~PSL_T;
455
456 /*
457 * Set registers for trampoline to user mode. Leave space for the
458 * return address on stack. These are the kernel mode register values.
459 */
460 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */
461 pcb2->pcb_rbp = 0;
462 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */
463 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */
464 pcb2->pcb_rip = (register_t)fork_trampoline;
465 /*
466 * If we didn't copy the pcb, we'd need to do the following registers:
467 * pcb2->pcb_dr*: cloned above.
468 * pcb2->pcb_savefpu: cloned above.
469 * pcb2->pcb_onfault: cloned above (always NULL here?).
470 * pcb2->pcb_[fg]sbase: cloned above
471 */
472
473 /* Setup to release spin count in fork_exit(). */
474 td->td_md.md_spinlock_count = 1;
475 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
476 }
477
478 /*
479 * Set that machine state for performing an upcall that starts
480 * the entry function with the given argument.
481 */
482 void
483 cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg,
484 stack_t *stack)
485 {
486
487 /*
488 * Do any extra cleaning that needs to be done.
489 * The thread may have optional components
490 * that are not present in a fresh thread.
491 * This may be a recycled thread so make it look
492 * as though it's newly allocated.
493 */
494 cpu_thread_clean(td);
495
496 #ifdef COMPAT_FREEBSD32
497 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
498 /*
499 * Set the trap frame to point at the beginning of the entry
500 * function.
501 */
502 td->td_frame->tf_rbp = 0;
503 td->td_frame->tf_rsp =
504 (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
505 td->td_frame->tf_rip = (uintptr_t)entry;
506
507 /* Return address sentinel value to stop stack unwinding. */
508 suword32((void *)td->td_frame->tf_rsp, 0);
509
510 /* Pass the argument to the entry point. */
511 suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)),
512 (uint32_t)(uintptr_t)arg);
513
514 return;
515 }
516 #endif
517
518 /*
519 * Set the trap frame to point at the beginning of the uts
520 * function.
521 */
522 td->td_frame->tf_rbp = 0;
523 td->td_frame->tf_rsp =
524 ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f;
525 td->td_frame->tf_rsp -= 8;
526 td->td_frame->tf_rip = (register_t)entry;
527 td->td_frame->tf_ds = _udatasel;
528 td->td_frame->tf_es = _udatasel;
529 td->td_frame->tf_fs = _ufssel;
530 td->td_frame->tf_gs = _ugssel;
531 td->td_frame->tf_flags = TF_HASSEGS;
532
533 /* Return address sentinel value to stop stack unwinding. */
534 suword((void *)td->td_frame->tf_rsp, 0);
535
536 /* Pass the argument to the entry point. */
537 td->td_frame->tf_rdi = (register_t)arg;
538 }
539
540 int
541 cpu_set_user_tls(struct thread *td, void *tls_base)
542 {
543 struct pcb *pcb;
544
545 if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS)
546 return (EINVAL);
547
548 pcb = td->td_pcb;
549 set_pcb_flags(pcb, PCB_FULL_IRET);
550 #ifdef COMPAT_FREEBSD32
551 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
552 pcb->pcb_gsbase = (register_t)tls_base;
553 return (0);
554 }
555 #endif
556 pcb->pcb_fsbase = (register_t)tls_base;
557 return (0);
558 }
559
560 /*
561 * Software interrupt handler for queued VM system processing.
562 */
563 void
564 swi_vm(void *dummy)
565 {
566 if (busdma_swi_pending != 0)
567 busdma_swi();
568 }
569
570 /*
571 * Tell whether this address is in some physical memory region.
572 * Currently used by the kernel coredump code in order to avoid
573 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
574 * or other unpredictable behaviour.
575 */
576
577 int
578 is_physical_memory(vm_paddr_t addr)
579 {
580
581 #ifdef DEV_ISA
582 /* The ISA ``memory hole''. */
583 if (addr >= 0xa0000 && addr < 0x100000)
584 return 0;
585 #endif
586
587 /*
588 * stuff other tests for known memory-mapped devices (PCI?)
589 * here
590 */
591
592 return 1;
593 }
Cache object: 0ce01177d9d5e96aa3dbc86e9ad31ca4
|