xref: /freebsd/sys/amd64/amd64/vm_machdep.c (revision bd9e0f5d) 
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("$FreeBSD$");
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/priv.h>
62 #include <sys/proc.h>
63 #include <sys/procctl.h>
64 #include <sys/smp.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysent.h>
67 #include <sys/unistd.h>
68 #include <sys/vnode.h>
69 #include <sys/vmmeter.h>
70 #include <sys/wait.h>
71 
72 #include <machine/cpu.h>
73 #include <machine/md_var.h>
74 #include <machine/pcb.h>
75 #include <machine/smp.h>
76 #include <machine/specialreg.h>
77 #include <machine/tss.h>
78 
79 #include <vm/vm.h>
80 #include <vm/vm_extern.h>
81 #include <vm/vm_kern.h>
82 #include <vm/vm_page.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_param.h>
85 
86 _Static_assert(OFFSETOF_MONITORBUF == offsetof(struct pcpu, pc_monitorbuf),
87     "OFFSETOF_MONITORBUF does not correspond with offset of pc_monitorbuf.");
88 
89 void
90 set_top_of_stack_td(struct thread *td)
91 {
92 	td->td_md.md_stack_base = td->td_kstack +
93 	    td->td_kstack_pages * PAGE_SIZE;
94 }
95 
96 struct savefpu *
97 get_pcb_user_save_td(struct thread *td)
98 {
99 	KASSERT(((vm_offset_t)td->td_md.md_usr_fpu_save %
100 	    XSAVE_AREA_ALIGN) == 0,
101 	    ("Unaligned pcb_user_save area ptr %p td %p",
102 	    td->td_md.md_usr_fpu_save, td));
103 	return (td->td_md.md_usr_fpu_save);
104 }
105 
106 struct pcb *
107 get_pcb_td(struct thread *td)
108 {
109 
110 	return (&td->td_md.md_pcb);
111 }
112 
113 struct savefpu *
114 get_pcb_user_save_pcb(struct pcb *pcb)
115 {
116 	struct thread *td;
117 
118 	td = __containerof(pcb, struct thread, td_md.md_pcb);
119 	return (get_pcb_user_save_td(td));
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  * Common code shared between cpu_fork() and cpu_copy_thread() for
139  * initializing a thread.
140  */
141 static void
142 copy_thread(struct thread *td1, struct thread *td2)
143 {
144 	struct pcb *pcb2;
145 
146 	pcb2 = td2->td_pcb;
147 
148 	/* Ensure that td1's pcb is up to date for user threads. */
149 	if ((td2->td_pflags & TDP_KTHREAD) == 0) {
150 		MPASS(td1 == curthread);
151 		fpuexit(td1);
152 		update_pcb_bases(td1->td_pcb);
153 	}
154 
155 	/* Copy td1's pcb */
156 	bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
157 
158 	/* Properly initialize pcb_save */
159 	pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
160 
161 	/* Kernel threads start with clean FPU and segment bases. */
162 	if ((td2->td_pflags & TDP_KTHREAD) != 0) {
163 		pcb2->pcb_fsbase = 0;
164 		pcb2->pcb_gsbase = 0;
165 		clear_pcb_flags(pcb2, PCB_FPUINITDONE | PCB_USERFPUINITDONE |
166 		    PCB_KERNFPU | PCB_KERNFPU_THR);
167 	} else {
168 		MPASS((pcb2->pcb_flags & (PCB_KERNFPU | PCB_KERNFPU_THR)) == 0);
169 		bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2),
170 		    cpu_max_ext_state_size);
171 	}
172 
173 	/*
174 	 * Set registers for trampoline to user mode.  Leave space for the
175 	 * return address on stack.  These are the kernel mode register values.
176 	 */
177 	pcb2->pcb_r12 = (register_t)fork_return;	/* fork_trampoline argument */
178 	pcb2->pcb_rbp = 0;
179 	pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *);
180 	pcb2->pcb_rbx = (register_t)td2;		/* fork_trampoline argument */
181 	pcb2->pcb_rip = (register_t)fork_trampoline;
182 	/*-
183 	 * pcb2->pcb_dr*:	cloned above.
184 	 * pcb2->pcb_savefpu:	cloned above.
185 	 * pcb2->pcb_flags:	cloned above.
186 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
187 	 * pcb2->pcb_[fg]sbase:	cloned above
188 	 */
189 
190 	pcb2->pcb_tssp = NULL;
191 
192 	/* Setup to release spin count in fork_exit(). */
193 	td2->td_md.md_spinlock_count = 1;
194 	td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
195 	pmap_thread_init_invl_gen(td2);
196 }
197 
198 /*
199  * Finish a fork operation, with process p2 nearly set up.
200  * Copy and update the pcb, set up the stack so that the child
201  * ready to run and return to user mode.
202  */
203 void
204 cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
205 {
206 	struct proc *p1;
207 	struct pcb *pcb2;
208 	struct mdproc *mdp1, *mdp2;
209 	struct proc_ldt *pldt;
210 
211 	p1 = td1->td_proc;
212 	if ((flags & RFPROC) == 0) {
213 		if ((flags & RFMEM) == 0) {
214 			/* unshare user LDT */
215 			mdp1 = &p1->p_md;
216 			mtx_lock(&dt_lock);
217 			if ((pldt = mdp1->md_ldt) != NULL &&
218 			    pldt->ldt_refcnt > 1 &&
219 			    user_ldt_alloc(p1, 1) == NULL)
220 				panic("could not copy LDT");
221 			mtx_unlock(&dt_lock);
222 		}
223 		return;
224 	}
225 
226 	/* Point the stack and pcb to the actual location */
227 	set_top_of_stack_td(td2);
228 	td2->td_pcb = pcb2 = get_pcb_td(td2);
229 
230 	copy_thread(td1, td2);
231 
232 	/* Reset debug registers in the new process */
233 	x86_clear_dbregs(pcb2);
234 
235 	/* Point mdproc and then copy over p1's contents */
236 	mdp2 = &p2->p_md;
237 	bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
238 
239 	/*
240 	 * Copy the trap frame for the return to user mode as if from a
241 	 * syscall.  This copies most of the user mode register values.
242 	 */
243 	td2->td_frame = (struct trapframe *)td2->td_md.md_stack_base - 1;
244 	bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
245 
246 	/* Set child return values. */
247 	p2->p_sysent->sv_set_fork_retval(td2);
248 
249 	/*
250 	 * If the parent process has the trap bit set (i.e. a debugger
251 	 * had single stepped the process to the system call), we need
252 	 * to clear the trap flag from the new frame.
253 	 */
254 	td2->td_frame->tf_rflags &= ~PSL_T;
255 
256 	/* As on i386, do not copy io permission bitmap. */
257 	pcb2->pcb_tssp = NULL;
258 
259 	/* New segment registers. */
260 	set_pcb_flags_raw(pcb2, PCB_FULL_IRET);
261 
262 	/* Copy the LDT, if necessary. */
263 	mdp1 = &td1->td_proc->p_md;
264 	mdp2 = &p2->p_md;
265 	if (mdp1->md_ldt == NULL) {
266 		mdp2->md_ldt = NULL;
267 		return;
268 	}
269 	mtx_lock(&dt_lock);
270 	if (mdp1->md_ldt != NULL) {
271 		if (flags & RFMEM) {
272 			mdp1->md_ldt->ldt_refcnt++;
273 			mdp2->md_ldt = mdp1->md_ldt;
274 			bcopy(&mdp1->md_ldt_sd, &mdp2->md_ldt_sd, sizeof(struct
275 			    system_segment_descriptor));
276 		} else {
277 			mdp2->md_ldt = NULL;
278 			mdp2->md_ldt = user_ldt_alloc(p2, 0);
279 			if (mdp2->md_ldt == NULL)
280 				panic("could not copy LDT");
281 			amd64_set_ldt_data(td2, 0, max_ldt_segment,
282 			    (struct user_segment_descriptor *)
283 			    mdp1->md_ldt->ldt_base);
284 		}
285 	} else
286 		mdp2->md_ldt = NULL;
287 	mtx_unlock(&dt_lock);
288 
289 	/*
290 	 * Now, cpu_switch() can schedule the new process.
291 	 * pcb_rsp is loaded pointing to the cpu_switch() stack frame
292 	 * containing the return address when exiting cpu_switch.
293 	 * This will normally be to fork_trampoline(), which will have
294 	 * %rbx loaded with the new proc's pointer.  fork_trampoline()
295 	 * will set up a stack to call fork_return(p, frame); to complete
296 	 * the return to user-mode.
297 	 */
298 }
299 
300 void
301 x86_set_fork_retval(struct thread *td)
302 {
303 	struct trapframe *frame = td->td_frame;
304 
305 	frame->tf_rax = 0;		/* Child returns zero */
306 	frame->tf_rflags &= ~PSL_C;	/* success */
307 	frame->tf_rdx = 1;		/* System V emulation */
308 }
309 
310 /*
311  * Intercept the return address from a freshly forked process that has NOT
312  * been scheduled yet.
313  *
314  * This is needed to make kernel threads stay in kernel mode.
315  */
316 void
317 cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg)
318 {
319 	/*
320 	 * Note that the trap frame follows the args, so the function
321 	 * is really called like this:  func(arg, frame);
322 	 */
323 	td->td_pcb->pcb_r12 = (long) func;	/* function */
324 	td->td_pcb->pcb_rbx = (long) arg;	/* first arg */
325 }
326 
327 void
328 cpu_exit(struct thread *td)
329 {
330 
331 	/*
332 	 * If this process has a custom LDT, release it.
333 	 */
334 	if (td->td_proc->p_md.md_ldt != NULL)
335 		user_ldt_free(td);
336 }
337 
338 void
339 cpu_thread_exit(struct thread *td)
340 {
341 	struct pcb *pcb;
342 
343 	critical_enter();
344 	if (td == PCPU_GET(fpcurthread))
345 		fpudrop();
346 	critical_exit();
347 
348 	pcb = td->td_pcb;
349 
350 	/* Disable any hardware breakpoints. */
351 	if (pcb->pcb_flags & PCB_DBREGS) {
352 		reset_dbregs();
353 		clear_pcb_flags(pcb, PCB_DBREGS);
354 	}
355 }
356 
357 void
358 cpu_thread_clean(struct thread *td)
359 {
360 	struct pcb *pcb;
361 
362 	pcb = td->td_pcb;
363 
364 	/*
365 	 * Clean TSS/iomap
366 	 */
367 	if (pcb->pcb_tssp != NULL) {
368 		pmap_pti_remove_kva((vm_offset_t)pcb->pcb_tssp,
369 		    (vm_offset_t)pcb->pcb_tssp + ctob(IOPAGES + 1));
370 		kmem_free((vm_offset_t)pcb->pcb_tssp, ctob(IOPAGES + 1));
371 		pcb->pcb_tssp = NULL;
372 	}
373 }
374 
375 void
376 cpu_thread_swapin(struct thread *td)
377 {
378 }
379 
380 void
381 cpu_thread_swapout(struct thread *td)
382 {
383 }
384 
385 void
386 cpu_thread_alloc(struct thread *td)
387 {
388 	struct pcb *pcb;
389 	struct xstate_hdr *xhdr;
390 
391 	set_top_of_stack_td(td);
392 	td->td_pcb = pcb = get_pcb_td(td);
393 	td->td_frame = (struct trapframe *)td->td_md.md_stack_base - 1;
394 	td->td_md.md_usr_fpu_save = fpu_save_area_alloc();
395 	pcb->pcb_save = get_pcb_user_save_pcb(pcb);
396 	if (use_xsave) {
397 		xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
398 		bzero(xhdr, sizeof(*xhdr));
399 		xhdr->xstate_bv = xsave_mask;
400 	}
401 }
402 
403 void
404 cpu_thread_free(struct thread *td)
405 {
406 	cpu_thread_clean(td);
407 
408 	fpu_save_area_free(td->td_md.md_usr_fpu_save);
409 	td->td_md.md_usr_fpu_save = NULL;
410 }
411 
412 bool
413 cpu_exec_vmspace_reuse(struct proc *p, vm_map_t map)
414 {
415 
416 	return (((curproc->p_md.md_flags & P_MD_KPTI) != 0) ==
417 	    (vm_map_pmap(map)->pm_ucr3 != PMAP_NO_CR3));
418 }
419 
420 static void
421 cpu_procctl_kpti_ctl(struct proc *p, int val)
422 {
423 
424 	if (pti && val == PROC_KPTI_CTL_ENABLE_ON_EXEC)
425 		p->p_md.md_flags |= P_MD_KPTI;
426 	if (val == PROC_KPTI_CTL_DISABLE_ON_EXEC)
427 		p->p_md.md_flags &= ~P_MD_KPTI;
428 }
429 
430 static void
431 cpu_procctl_kpti_status(struct proc *p, int *val)
432 {
433 	*val = (p->p_md.md_flags & P_MD_KPTI) != 0 ?
434 	    PROC_KPTI_CTL_ENABLE_ON_EXEC:
435 	    PROC_KPTI_CTL_DISABLE_ON_EXEC;
436 	if (vmspace_pmap(p->p_vmspace)->pm_ucr3 != PMAP_NO_CR3)
437 		*val |= PROC_KPTI_STATUS_ACTIVE;
438 }
439 
440 static int
441 cpu_procctl_la_ctl(struct proc *p, int val)
442 {
443 	int error;
444 
445 	error = 0;
446 	switch (val) {
447 	case PROC_LA_CTL_LA48_ON_EXEC:
448 		p->p_md.md_flags |= P_MD_LA48;
449 		p->p_md.md_flags &= ~P_MD_LA57;
450 		break;
451 	case PROC_LA_CTL_LA57_ON_EXEC:
452 		if (la57) {
453 			p->p_md.md_flags &= ~P_MD_LA48;
454 			p->p_md.md_flags |= P_MD_LA57;
455 		} else {
456 			error = ENOTSUP;
457 		}
458 		break;
459 	case PROC_LA_CTL_DEFAULT_ON_EXEC:
460 		p->p_md.md_flags &= ~(P_MD_LA48 | P_MD_LA57);
461 		break;
462 	}
463 	return (error);
464 }
465 
466 static void
467 cpu_procctl_la_status(struct proc *p, int *val)
468 {
469 	int res;
470 
471 	if ((p->p_md.md_flags & P_MD_LA48) != 0)
472 		res = PROC_LA_CTL_LA48_ON_EXEC;
473 	else if ((p->p_md.md_flags & P_MD_LA57) != 0)
474 		res = PROC_LA_CTL_LA57_ON_EXEC;
475 	else
476 		res = PROC_LA_CTL_DEFAULT_ON_EXEC;
477 	if (p->p_sysent->sv_maxuser == VM_MAXUSER_ADDRESS_LA48)
478 		res |= PROC_LA_STATUS_LA48;
479 	else
480 		res |= PROC_LA_STATUS_LA57;
481 	*val = res;
482 }
483 
484 int
485 cpu_procctl(struct thread *td, int idtype, id_t id, int com, void *data)
486 {
487 	struct proc *p;
488 	int error, val;
489 
490 	switch (com) {
491 	case PROC_KPTI_CTL:
492 	case PROC_KPTI_STATUS:
493 	case PROC_LA_CTL:
494 	case PROC_LA_STATUS:
495 		if (idtype != P_PID) {
496 			error = EINVAL;
497 			break;
498 		}
499 		if (com == PROC_KPTI_CTL) {
500 			/* sad but true and not a joke */
501 			error = priv_check(td, PRIV_IO);
502 			if (error != 0)
503 				break;
504 		}
505 		if (com == PROC_KPTI_CTL || com == PROC_LA_CTL) {
506 			error = copyin(data, &val, sizeof(val));
507 			if (error != 0)
508 				break;
509 		}
510 		if (com == PROC_KPTI_CTL &&
511 		    val != PROC_KPTI_CTL_ENABLE_ON_EXEC &&
512 		    val != PROC_KPTI_CTL_DISABLE_ON_EXEC) {
513 			error = EINVAL;
514 			break;
515 		}
516 		if (com == PROC_LA_CTL &&
517 		    val != PROC_LA_CTL_LA48_ON_EXEC &&
518 		    val != PROC_LA_CTL_LA57_ON_EXEC &&
519 		    val != PROC_LA_CTL_DEFAULT_ON_EXEC) {
520 			error = EINVAL;
521 			break;
522 		}
523 		error = pget(id, PGET_CANSEE | PGET_NOTWEXIT | PGET_NOTID, &p);
524 		if (error != 0)
525 			break;
526 		switch (com) {
527 		case PROC_KPTI_CTL:
528 			cpu_procctl_kpti_ctl(p, val);
529 			break;
530 		case PROC_KPTI_STATUS:
531 			cpu_procctl_kpti_status(p, &val);
532 			break;
533 		case PROC_LA_CTL:
534 			error = cpu_procctl_la_ctl(p, val);
535 			break;
536 		case PROC_LA_STATUS:
537 			cpu_procctl_la_status(p, &val);
538 			break;
539 		}
540 		PROC_UNLOCK(p);
541 		if (com == PROC_KPTI_STATUS || com == PROC_LA_STATUS)
542 			error = copyout(&val, data, sizeof(val));
543 		break;
544 	default:
545 		error = EINVAL;
546 		break;
547 	}
548 	return (error);
549 }
550 
551 void
552 cpu_set_syscall_retval(struct thread *td, int error)
553 {
554 	struct trapframe *frame;
555 
556 	frame = td->td_frame;
557 	if (__predict_true(error == 0)) {
558 		frame->tf_rax = td->td_retval[0];
559 		frame->tf_rdx = td->td_retval[1];
560 		frame->tf_rflags &= ~PSL_C;
561 		return;
562 	}
563 
564 	switch (error) {
565 	case ERESTART:
566 		/*
567 		 * Reconstruct pc, we know that 'syscall' is 2 bytes,
568 		 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
569 		 * We saved this in tf_err.
570 		 * %r10 (which was holding the value of %rcx) is restored
571 		 * for the next iteration.
572 		 * %r10 restore is only required for freebsd/amd64 processes,
573 		 * but shall be innocent for any ia32 ABI.
574 		 *
575 		 * Require full context restore to get the arguments
576 		 * in the registers reloaded at return to usermode.
577 		 */
578 		frame->tf_rip -= frame->tf_err;
579 		frame->tf_r10 = frame->tf_rcx;
580 		set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
581 		break;
582 
583 	case EJUSTRETURN:
584 		break;
585 
586 	default:
587 		frame->tf_rax = error;
588 		frame->tf_rflags |= PSL_C;
589 		break;
590 	}
591 }
592 
593 /*
594  * Initialize machine state, mostly pcb and trap frame for a new
595  * thread, about to return to userspace.  Put enough state in the new
596  * thread's PCB to get it to go back to the fork_return(), which
597  * finalizes the thread state and handles peculiarities of the first
598  * return to userspace for the new thread.
599  */
600 void
601 cpu_copy_thread(struct thread *td, struct thread *td0)
602 {
603 	copy_thread(td0, td);
604 
605 	/*
606 	 * Copy user general-purpose registers.
607 	 *
608 	 * Some of these registers are rewritten by cpu_set_upcall()
609 	 * and linux_set_upcall().
610 	 */
611 	bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
612 
613 	/* If the current thread has the trap bit set (i.e. a debugger had
614 	 * single stepped the process to the system call), we need to clear
615 	 * the trap flag from the new frame. Otherwise, the new thread will
616 	 * receive a (likely unexpected) SIGTRAP when it executes the first
617 	 * instruction after returning to userland.
618 	 */
619 	td->td_frame->tf_rflags &= ~PSL_T;
620 
621 	set_pcb_flags_raw(td->td_pcb, PCB_FULL_IRET);
622 }
623 
624 /*
625  * Set that machine state for performing an upcall that starts
626  * the entry function with the given argument.
627  */
628 void
629 cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg,
630     stack_t *stack)
631 {
632 
633 	/*
634 	 * Do any extra cleaning that needs to be done.
635 	 * The thread may have optional components
636 	 * that are not present in a fresh thread.
637 	 * This may be a recycled thread so make it look
638 	 * as though it's newly allocated.
639 	 */
640 	cpu_thread_clean(td);
641 
642 #ifdef COMPAT_FREEBSD32
643 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
644 		/*
645 		 * Set the trap frame to point at the beginning of the entry
646 		 * function.
647 		 */
648 		td->td_frame->tf_rbp = 0;
649 		td->td_frame->tf_rsp =
650 		   (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
651 		td->td_frame->tf_rip = (uintptr_t)entry;
652 
653 		/* Return address sentinel value to stop stack unwinding. */
654 		suword32((void *)td->td_frame->tf_rsp, 0);
655 
656 		/* Pass the argument to the entry point. */
657 		suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)),
658 		    (uint32_t)(uintptr_t)arg);
659 
660 		return;
661 	}
662 #endif
663 
664 	/*
665 	 * Set the trap frame to point at the beginning of the uts
666 	 * function.
667 	 */
668 	td->td_frame->tf_rbp = 0;
669 	td->td_frame->tf_rsp =
670 	    ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f;
671 	td->td_frame->tf_rsp -= 8;
672 	td->td_frame->tf_rip = (register_t)entry;
673 	td->td_frame->tf_ds = _udatasel;
674 	td->td_frame->tf_es = _udatasel;
675 	td->td_frame->tf_fs = _ufssel;
676 	td->td_frame->tf_gs = _ugssel;
677 	td->td_frame->tf_flags = TF_HASSEGS;
678 
679 	/* Return address sentinel value to stop stack unwinding. */
680 	suword((void *)td->td_frame->tf_rsp, 0);
681 
682 	/* Pass the argument to the entry point. */
683 	td->td_frame->tf_rdi = (register_t)arg;
684 }
685 
686 int
687 cpu_set_user_tls(struct thread *td, void *tls_base)
688 {
689 	struct pcb *pcb;
690 
691 	if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS)
692 		return (EINVAL);
693 
694 	pcb = td->td_pcb;
695 	set_pcb_flags(pcb, PCB_FULL_IRET);
696 #ifdef COMPAT_FREEBSD32
697 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
698 		pcb->pcb_gsbase = (register_t)tls_base;
699 		return (0);
700 	}
701 #endif
702 	pcb->pcb_fsbase = (register_t)tls_base;
703 	return (0);
704 }
705 
706 /*
707  * Software interrupt handler for queued VM system processing.
708  */
709 void
710 swi_vm(void *dummy)
711 {
712 	if (busdma_swi_pending != 0)
713 		busdma_swi();
714 }
715 
716 /*
717  * Tell whether this address is in some physical memory region.
718  * Currently used by the kernel coredump code in order to avoid
719  * dumping the ``ISA memory hole'' which could cause indefinite hangs,
720  * or other unpredictable behaviour.
721  */
722 
723 int
724 is_physical_memory(vm_paddr_t addr)
725 {
726 
727 #ifdef DEV_ISA
728 	/* The ISA ``memory hole''. */
729 	if (addr >= 0xa0000 && addr < 0x100000)
730 		return 0;
731 #endif
732 
733 	/*
734 	 * stuff other tests for known memory-mapped devices (PCI?)
735 	 * here
736 	 */
737 
738 	return 1;
739 }
740