xref: /freebsd/sys/amd64/amd64/vm_machdep.c (revision 290b0d12) 
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 	    roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN);
95 }
96 
97 struct savefpu *
98 get_pcb_user_save_td(struct thread *td)
99 {
100 	vm_offset_t p;
101 
102 	p = td->td_md.md_stack_base;
103 	KASSERT((p % XSAVE_AREA_ALIGN) == 0,
104 	    ("Unaligned pcb_user_save area ptr %#lx td %p", p, td));
105 	return ((struct savefpu *)p);
106 }
107 
108 struct pcb *
109 get_pcb_td(struct thread *td)
110 {
111 
112 	return (&td->td_md.md_pcb);
113 }
114 
115 struct savefpu *
116 get_pcb_user_save_pcb(struct pcb *pcb)
117 {
118 	struct thread *td;
119 
120 	td = __containerof(pcb, struct thread, td_md.md_pcb);
121 	return (get_pcb_user_save_td(td));
122 }
123 
124 void *
125 alloc_fpusave(int flags)
126 {
127 	void *res;
128 	struct savefpu_ymm *sf;
129 
130 	res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags);
131 	if (use_xsave) {
132 		sf = (struct savefpu_ymm *)res;
133 		bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd));
134 		sf->sv_xstate.sx_hd.xstate_bv = xsave_mask;
135 	}
136 	return (res);
137 }
138 
139 /*
140  * Common code shared between cpu_fork() and cpu_copy_thread() for
141  * initializing a thread.
142  */
143 static void
144 copy_thread(struct thread *td1, struct thread *td2)
145 {
146 	struct pcb *pcb2;
147 
148 	pcb2 = td2->td_pcb;
149 
150 	/* Ensure that td1's pcb is up to date for user threads. */
151 	if ((td2->td_pflags & TDP_KTHREAD) == 0) {
152 		MPASS(td1 == curthread);
153 		fpuexit(td1);
154 		update_pcb_bases(td1->td_pcb);
155 	}
156 
157 	/* Copy td1's pcb */
158 	bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
159 
160 	/* Properly initialize pcb_save */
161 	pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
162 
163 	/* Kernel threads start with clean FPU and segment bases. */
164 	if ((td2->td_pflags & TDP_KTHREAD) != 0) {
165 		pcb2->pcb_fsbase = 0;
166 		pcb2->pcb_gsbase = 0;
167 		clear_pcb_flags(pcb2, PCB_FPUINITDONE | PCB_USERFPUINITDONE |
168 		    PCB_KERNFPU | PCB_KERNFPU_THR);
169 	} else {
170 		MPASS((pcb2->pcb_flags & (PCB_KERNFPU | PCB_KERNFPU_THR)) == 0);
171 		bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2),
172 		    cpu_max_ext_state_size);
173 	}
174 
175 	/*
176 	 * Set registers for trampoline to user mode.  Leave space for the
177 	 * return address on stack.  These are the kernel mode register values.
178 	 */
179 	pcb2->pcb_r12 = (register_t)fork_return;	/* fork_trampoline argument */
180 	pcb2->pcb_rbp = 0;
181 	pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *);
182 	pcb2->pcb_rbx = (register_t)td2;		/* fork_trampoline argument */
183 	pcb2->pcb_rip = (register_t)fork_trampoline;
184 	/*-
185 	 * pcb2->pcb_dr*:	cloned above.
186 	 * pcb2->pcb_savefpu:	cloned above.
187 	 * pcb2->pcb_flags:	cloned above.
188 	 * pcb2->pcb_onfault:	cloned above (always NULL here?).
189 	 * pcb2->pcb_[fg]sbase:	cloned above
190 	 */
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 	td2->td_frame->tf_rax = 0;		/* Child returns zero */
247 	td2->td_frame->tf_rflags &= ~PSL_C;	/* success */
248 	td2->td_frame->tf_rdx = 1;
249 
250 	/*
251 	 * If the parent process has the trap bit set (i.e. a debugger
252 	 * had single stepped the process to the system call), we need
253 	 * to clear the trap flag from the new frame.
254 	 */
255 	td2->td_frame->tf_rflags &= ~PSL_T;
256 
257 	/* As on i386, do not copy io permission bitmap. */
258 	pcb2->pcb_tssp = NULL;
259 
260 	/* New segment registers. */
261 	set_pcb_flags_raw(pcb2, PCB_FULL_IRET);
262 
263 	/* Copy the LDT, if necessary. */
264 	mdp1 = &td1->td_proc->p_md;
265 	mdp2 = &p2->p_md;
266 	if (mdp1->md_ldt == NULL) {
267 		mdp2->md_ldt = NULL;
268 		return;
269 	}
270 	mtx_lock(&dt_lock);
271 	if (mdp1->md_ldt != NULL) {
272 		if (flags & RFMEM) {
273 			mdp1->md_ldt->ldt_refcnt++;
274 			mdp2->md_ldt = mdp1->md_ldt;
275 			bcopy(&mdp1->md_ldt_sd, &mdp2->md_ldt_sd, sizeof(struct
276 			    system_segment_descriptor));
277 		} else {
278 			mdp2->md_ldt = NULL;
279 			mdp2->md_ldt = user_ldt_alloc(p2, 0);
280 			if (mdp2->md_ldt == NULL)
281 				panic("could not copy LDT");
282 			amd64_set_ldt_data(td2, 0, max_ldt_segment,
283 			    (struct user_segment_descriptor *)
284 			    mdp1->md_ldt->ldt_base);
285 		}
286 	} else
287 		mdp2->md_ldt = NULL;
288 	mtx_unlock(&dt_lock);
289 
290 	/*
291 	 * Now, cpu_switch() can schedule the new process.
292 	 * pcb_rsp is loaded pointing to the cpu_switch() stack frame
293 	 * containing the return address when exiting cpu_switch.
294 	 * This will normally be to fork_trampoline(), which will have
295 	 * %rbx loaded with the new proc's pointer.  fork_trampoline()
296 	 * will set up a stack to call fork_return(p, frame); to complete
297 	 * the return to user-mode.
298 	 */
299 }
300 
301 /*
302  * Intercept the return address from a freshly forked process that has NOT
303  * been scheduled yet.
304  *
305  * This is needed to make kernel threads stay in kernel mode.
306  */
307 void
308 cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg)
309 {
310 	/*
311 	 * Note that the trap frame follows the args, so the function
312 	 * is really called like this:  func(arg, frame);
313 	 */
314 	td->td_pcb->pcb_r12 = (long) func;	/* function */
315 	td->td_pcb->pcb_rbx = (long) arg;	/* first arg */
316 }
317 
318 void
319 cpu_exit(struct thread *td)
320 {
321 
322 	/*
323 	 * If this process has a custom LDT, release it.
324 	 */
325 	if (td->td_proc->p_md.md_ldt != NULL)
326 		user_ldt_free(td);
327 }
328 
329 void
330 cpu_thread_exit(struct thread *td)
331 {
332 	struct pcb *pcb;
333 
334 	critical_enter();
335 	if (td == PCPU_GET(fpcurthread))
336 		fpudrop();
337 	critical_exit();
338 
339 	pcb = td->td_pcb;
340 
341 	/* Disable any hardware breakpoints. */
342 	if (pcb->pcb_flags & PCB_DBREGS) {
343 		reset_dbregs();
344 		clear_pcb_flags(pcb, PCB_DBREGS);
345 	}
346 }
347 
348 void
349 cpu_thread_clean(struct thread *td)
350 {
351 	struct pcb *pcb;
352 
353 	pcb = td->td_pcb;
354 
355 	/*
356 	 * Clean TSS/iomap
357 	 */
358 	if (pcb->pcb_tssp != NULL) {
359 		pmap_pti_remove_kva((vm_offset_t)pcb->pcb_tssp,
360 		    (vm_offset_t)pcb->pcb_tssp + ctob(IOPAGES + 1));
361 		kmem_free((vm_offset_t)pcb->pcb_tssp, ctob(IOPAGES + 1));
362 		pcb->pcb_tssp = NULL;
363 	}
364 }
365 
366 void
367 cpu_thread_swapin(struct thread *td)
368 {
369 }
370 
371 void
372 cpu_thread_swapout(struct thread *td)
373 {
374 }
375 
376 void
377 cpu_thread_alloc(struct thread *td)
378 {
379 	struct pcb *pcb;
380 	struct xstate_hdr *xhdr;
381 
382 	set_top_of_stack_td(td);
383 	td->td_pcb = pcb = get_pcb_td(td);
384 	td->td_frame = (struct trapframe *)td->td_md.md_stack_base - 1;
385 	pcb->pcb_save = get_pcb_user_save_pcb(pcb);
386 	if (use_xsave) {
387 		xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
388 		bzero(xhdr, sizeof(*xhdr));
389 		xhdr->xstate_bv = xsave_mask;
390 	}
391 }
392 
393 void
394 cpu_thread_free(struct thread *td)
395 {
396 
397 	cpu_thread_clean(td);
398 }
399 
400 bool
401 cpu_exec_vmspace_reuse(struct proc *p, vm_map_t map)
402 {
403 
404 	return (((curproc->p_md.md_flags & P_MD_KPTI) != 0) ==
405 	    (vm_map_pmap(map)->pm_ucr3 != PMAP_NO_CR3));
406 }
407 
408 static void
409 cpu_procctl_kpti_ctl(struct proc *p, int val)
410 {
411 
412 	if (pti && val == PROC_KPTI_CTL_ENABLE_ON_EXEC)
413 		p->p_md.md_flags |= P_MD_KPTI;
414 	if (val == PROC_KPTI_CTL_DISABLE_ON_EXEC)
415 		p->p_md.md_flags &= ~P_MD_KPTI;
416 }
417 
418 static void
419 cpu_procctl_kpti_status(struct proc *p, int *val)
420 {
421 	*val = (p->p_md.md_flags & P_MD_KPTI) != 0 ?
422 	    PROC_KPTI_CTL_ENABLE_ON_EXEC:
423 	    PROC_KPTI_CTL_DISABLE_ON_EXEC;
424 	if (vmspace_pmap(p->p_vmspace)->pm_ucr3 != PMAP_NO_CR3)
425 		*val |= PROC_KPTI_STATUS_ACTIVE;
426 }
427 
428 static int
429 cpu_procctl_la_ctl(struct proc *p, int val)
430 {
431 	int error;
432 
433 	error = 0;
434 	switch (val) {
435 	case PROC_LA_CTL_LA48_ON_EXEC:
436 		p->p_md.md_flags |= P_MD_LA48;
437 		p->p_md.md_flags &= ~P_MD_LA57;
438 		break;
439 	case PROC_LA_CTL_LA57_ON_EXEC:
440 		if (la57) {
441 			p->p_md.md_flags &= ~P_MD_LA48;
442 			p->p_md.md_flags |= P_MD_LA57;
443 		} else {
444 			error = ENOTSUP;
445 		}
446 		break;
447 	case PROC_LA_CTL_DEFAULT_ON_EXEC:
448 		p->p_md.md_flags &= ~(P_MD_LA48 | P_MD_LA57);
449 		break;
450 	}
451 	return (error);
452 }
453 
454 static void
455 cpu_procctl_la_status(struct proc *p, int *val)
456 {
457 	int res;
458 
459 	if ((p->p_md.md_flags & P_MD_LA48) != 0)
460 		res = PROC_LA_CTL_LA48_ON_EXEC;
461 	else if ((p->p_md.md_flags & P_MD_LA57) != 0)
462 		res = PROC_LA_CTL_LA57_ON_EXEC;
463 	else
464 		res = PROC_LA_CTL_DEFAULT_ON_EXEC;
465 	if (p->p_sysent->sv_maxuser == VM_MAXUSER_ADDRESS_LA48)
466 		res |= PROC_LA_STATUS_LA48;
467 	else
468 		res |= PROC_LA_STATUS_LA57;
469 	*val = res;
470 }
471 
472 int
473 cpu_procctl(struct thread *td, int idtype, id_t id, int com, void *data)
474 {
475 	struct proc *p;
476 	int error, val;
477 
478 	switch (com) {
479 	case PROC_KPTI_CTL:
480 	case PROC_KPTI_STATUS:
481 	case PROC_LA_CTL:
482 	case PROC_LA_STATUS:
483 		if (idtype != P_PID) {
484 			error = EINVAL;
485 			break;
486 		}
487 		if (com == PROC_KPTI_CTL) {
488 			/* sad but true and not a joke */
489 			error = priv_check(td, PRIV_IO);
490 			if (error != 0)
491 				break;
492 		}
493 		if (com == PROC_KPTI_CTL || com == PROC_LA_CTL) {
494 			error = copyin(data, &val, sizeof(val));
495 			if (error != 0)
496 				break;
497 		}
498 		if (com == PROC_KPTI_CTL &&
499 		    val != PROC_KPTI_CTL_ENABLE_ON_EXEC &&
500 		    val != PROC_KPTI_CTL_DISABLE_ON_EXEC) {
501 			error = EINVAL;
502 			break;
503 		}
504 		if (com == PROC_LA_CTL &&
505 		    val != PROC_LA_CTL_LA48_ON_EXEC &&
506 		    val != PROC_LA_CTL_LA57_ON_EXEC &&
507 		    val != PROC_LA_CTL_DEFAULT_ON_EXEC) {
508 			error = EINVAL;
509 			break;
510 		}
511 		error = pget(id, PGET_CANSEE | PGET_NOTWEXIT | PGET_NOTID, &p);
512 		if (error != 0)
513 			break;
514 		switch (com) {
515 		case PROC_KPTI_CTL:
516 			cpu_procctl_kpti_ctl(p, val);
517 			break;
518 		case PROC_KPTI_STATUS:
519 			cpu_procctl_kpti_status(p, &val);
520 			break;
521 		case PROC_LA_CTL:
522 			error = cpu_procctl_la_ctl(p, val);
523 			break;
524 		case PROC_LA_STATUS:
525 			cpu_procctl_la_status(p, &val);
526 			break;
527 		}
528 		PROC_UNLOCK(p);
529 		if (com == PROC_KPTI_STATUS || com == PROC_LA_STATUS)
530 			error = copyout(&val, data, sizeof(val));
531 		break;
532 	default:
533 		error = EINVAL;
534 		break;
535 	}
536 	return (error);
537 }
538 
539 void
540 cpu_set_syscall_retval(struct thread *td, int error)
541 {
542 	struct trapframe *frame;
543 
544 	frame = td->td_frame;
545 	if (__predict_true(error == 0)) {
546 		frame->tf_rax = td->td_retval[0];
547 		frame->tf_rdx = td->td_retval[1];
548 		frame->tf_rflags &= ~PSL_C;
549 		return;
550 	}
551 
552 	switch (error) {
553 	case ERESTART:
554 		/*
555 		 * Reconstruct pc, we know that 'syscall' is 2 bytes,
556 		 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
557 		 * We saved this in tf_err.
558 		 * %r10 (which was holding the value of %rcx) is restored
559 		 * for the next iteration.
560 		 * %r10 restore is only required for freebsd/amd64 processes,
561 		 * but shall be innocent for any ia32 ABI.
562 		 *
563 		 * Require full context restore to get the arguments
564 		 * in the registers reloaded at return to usermode.
565 		 */
566 		frame->tf_rip -= frame->tf_err;
567 		frame->tf_r10 = frame->tf_rcx;
568 		set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
569 		break;
570 
571 	case EJUSTRETURN:
572 		break;
573 
574 	default:
575 		frame->tf_rax = error;
576 		frame->tf_rflags |= PSL_C;
577 		break;
578 	}
579 }
580 
581 /*
582  * Initialize machine state, mostly pcb and trap frame for a new
583  * thread, about to return to userspace.  Put enough state in the new
584  * thread's PCB to get it to go back to the fork_return(), which
585  * finalizes the thread state and handles peculiarities of the first
586  * return to userspace for the new thread.
587  */
588 void
589 cpu_copy_thread(struct thread *td, struct thread *td0)
590 {
591 	copy_thread(td0, td);
592 
593 	/*
594 	 * Copy user general-purpose registers.
595 	 *
596 	 * Some of these registers are rewritten by cpu_set_upcall()
597 	 * and linux_set_upcall().
598 	 */
599 	bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
600 
601 	/* If the current thread has the trap bit set (i.e. a debugger had
602 	 * single stepped the process to the system call), we need to clear
603 	 * the trap flag from the new frame. Otherwise, the new thread will
604 	 * receive a (likely unexpected) SIGTRAP when it executes the first
605 	 * instruction after returning to userland.
606 	 */
607 	td->td_frame->tf_rflags &= ~PSL_T;
608 
609 	set_pcb_flags_raw(td->td_pcb, PCB_FULL_IRET);
610 }
611 
612 /*
613  * Set that machine state for performing an upcall that starts
614  * the entry function with the given argument.
615  */
616 void
617 cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg,
618     stack_t *stack)
619 {
620 
621 	/*
622 	 * Do any extra cleaning that needs to be done.
623 	 * The thread may have optional components
624 	 * that are not present in a fresh thread.
625 	 * This may be a recycled thread so make it look
626 	 * as though it's newly allocated.
627 	 */
628 	cpu_thread_clean(td);
629 
630 #ifdef COMPAT_FREEBSD32
631 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
632 		/*
633 		 * Set the trap frame to point at the beginning of the entry
634 		 * function.
635 		 */
636 		td->td_frame->tf_rbp = 0;
637 		td->td_frame->tf_rsp =
638 		   (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
639 		td->td_frame->tf_rip = (uintptr_t)entry;
640 
641 		/* Return address sentinel value to stop stack unwinding. */
642 		suword32((void *)td->td_frame->tf_rsp, 0);
643 
644 		/* Pass the argument to the entry point. */
645 		suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)),
646 		    (uint32_t)(uintptr_t)arg);
647 
648 		return;
649 	}
650 #endif
651 
652 	/*
653 	 * Set the trap frame to point at the beginning of the uts
654 	 * function.
655 	 */
656 	td->td_frame->tf_rbp = 0;
657 	td->td_frame->tf_rsp =
658 	    ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f;
659 	td->td_frame->tf_rsp -= 8;
660 	td->td_frame->tf_rip = (register_t)entry;
661 	td->td_frame->tf_ds = _udatasel;
662 	td->td_frame->tf_es = _udatasel;
663 	td->td_frame->tf_fs = _ufssel;
664 	td->td_frame->tf_gs = _ugssel;
665 	td->td_frame->tf_flags = TF_HASSEGS;
666 
667 	/* Return address sentinel value to stop stack unwinding. */
668 	suword((void *)td->td_frame->tf_rsp, 0);
669 
670 	/* Pass the argument to the entry point. */
671 	td->td_frame->tf_rdi = (register_t)arg;
672 }
673 
674 int
675 cpu_set_user_tls(struct thread *td, void *tls_base)
676 {
677 	struct pcb *pcb;
678 
679 	if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS)
680 		return (EINVAL);
681 
682 	pcb = td->td_pcb;
683 	set_pcb_flags(pcb, PCB_FULL_IRET);
684 #ifdef COMPAT_FREEBSD32
685 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
686 		pcb->pcb_gsbase = (register_t)tls_base;
687 		return (0);
688 	}
689 #endif
690 	pcb->pcb_fsbase = (register_t)tls_base;
691 	return (0);
692 }
693 
694 /*
695  * Software interrupt handler for queued VM system processing.
696  */
697 void
698 swi_vm(void *dummy)
699 {
700 	if (busdma_swi_pending != 0)
701 		busdma_swi();
702 }
703 
704 /*
705  * Tell whether this address is in some physical memory region.
706  * Currently used by the kernel coredump code in order to avoid
707  * dumping the ``ISA memory hole'' which could cause indefinite hangs,
708  * or other unpredictable behaviour.
709  */
710 
711 int
712 is_physical_memory(vm_paddr_t addr)
713 {
714 
715 #ifdef DEV_ISA
716 	/* The ISA ``memory hole''. */
717 	if (addr >= 0xa0000 && addr < 0x100000)
718 		return 0;
719 #endif
720 
721 	/*
722 	 * stuff other tests for known memory-mapped devices (PCI?)
723 	 * here
724 	 */
725 
726 	return 1;
727 }
728