xref: /freebsd/sys/amd64/amd64/vm_machdep.c (revision de8374df) 
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 	pcb2->pcb_tssp = NULL;
193 
194 	/* Setup to release spin count in fork_exit(). */
195 	td2->td_md.md_spinlock_count = 1;
196 	td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
197 	pmap_thread_init_invl_gen(td2);
198 }
199 
200 /*
201  * Finish a fork operation, with process p2 nearly set up.
202  * Copy and update the pcb, set up the stack so that the child
203  * ready to run and return to user mode.
204  */
205 void
206 cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
207 {
208 	struct proc *p1;
209 	struct pcb *pcb2;
210 	struct mdproc *mdp1, *mdp2;
211 	struct proc_ldt *pldt;
212 
213 	p1 = td1->td_proc;
214 	if ((flags & RFPROC) == 0) {
215 		if ((flags & RFMEM) == 0) {
216 			/* unshare user LDT */
217 			mdp1 = &p1->p_md;
218 			mtx_lock(&dt_lock);
219 			if ((pldt = mdp1->md_ldt) != NULL &&
220 			    pldt->ldt_refcnt > 1 &&
221 			    user_ldt_alloc(p1, 1) == NULL)
222 				panic("could not copy LDT");
223 			mtx_unlock(&dt_lock);
224 		}
225 		return;
226 	}
227 
228 	/* Point the stack and pcb to the actual location */
229 	set_top_of_stack_td(td2);
230 	td2->td_pcb = pcb2 = get_pcb_td(td2);
231 
232 	copy_thread(td1, td2);
233 
234 	/* Reset debug registers in the new process */
235 	x86_clear_dbregs(pcb2);
236 
237 	/* Point mdproc and then copy over p1's contents */
238 	mdp2 = &p2->p_md;
239 	bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
240 
241 	/*
242 	 * Copy the trap frame for the return to user mode as if from a
243 	 * syscall.  This copies most of the user mode register values.
244 	 */
245 	td2->td_frame = (struct trapframe *)td2->td_md.md_stack_base - 1;
246 	bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
247 
248 	/* Set child return values. */
249 	p2->p_sysent->sv_set_fork_retval(td2);
250 
251 	/*
252 	 * If the parent process has the trap bit set (i.e. a debugger
253 	 * had single stepped the process to the system call), we need
254 	 * to clear the trap flag from the new frame.
255 	 */
256 	td2->td_frame->tf_rflags &= ~PSL_T;
257 
258 	/* As on i386, do not copy io permission bitmap. */
259 	pcb2->pcb_tssp = NULL;
260 
261 	/* New segment registers. */
262 	set_pcb_flags_raw(pcb2, PCB_FULL_IRET);
263 
264 	/* Copy the LDT, if necessary. */
265 	mdp1 = &td1->td_proc->p_md;
266 	mdp2 = &p2->p_md;
267 	if (mdp1->md_ldt == NULL) {
268 		mdp2->md_ldt = NULL;
269 		return;
270 	}
271 	mtx_lock(&dt_lock);
272 	if (mdp1->md_ldt != NULL) {
273 		if (flags & RFMEM) {
274 			mdp1->md_ldt->ldt_refcnt++;
275 			mdp2->md_ldt = mdp1->md_ldt;
276 			bcopy(&mdp1->md_ldt_sd, &mdp2->md_ldt_sd, sizeof(struct
277 			    system_segment_descriptor));
278 		} else {
279 			mdp2->md_ldt = NULL;
280 			mdp2->md_ldt = user_ldt_alloc(p2, 0);
281 			if (mdp2->md_ldt == NULL)
282 				panic("could not copy LDT");
283 			amd64_set_ldt_data(td2, 0, max_ldt_segment,
284 			    (struct user_segment_descriptor *)
285 			    mdp1->md_ldt->ldt_base);
286 		}
287 	} else
288 		mdp2->md_ldt = NULL;
289 	mtx_unlock(&dt_lock);
290 
291 	/*
292 	 * Now, cpu_switch() can schedule the new process.
293 	 * pcb_rsp is loaded pointing to the cpu_switch() stack frame
294 	 * containing the return address when exiting cpu_switch.
295 	 * This will normally be to fork_trampoline(), which will have
296 	 * %rbx loaded with the new proc's pointer.  fork_trampoline()
297 	 * will set up a stack to call fork_return(p, frame); to complete
298 	 * the return to user-mode.
299 	 */
300 }
301 
302 void
303 x86_set_fork_retval(struct thread *td)
304 {
305 	struct trapframe *frame = td->td_frame;
306 
307 	frame->tf_rax = 0;		/* Child returns zero */
308 	frame->tf_rflags &= ~PSL_C;	/* success */
309 	frame->tf_rdx = 1;		/* System V emulation */
310 }
311 
312 /*
313  * Intercept the return address from a freshly forked process that has NOT
314  * been scheduled yet.
315  *
316  * This is needed to make kernel threads stay in kernel mode.
317  */
318 void
319 cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg)
320 {
321 	/*
322 	 * Note that the trap frame follows the args, so the function
323 	 * is really called like this:  func(arg, frame);
324 	 */
325 	td->td_pcb->pcb_r12 = (long) func;	/* function */
326 	td->td_pcb->pcb_rbx = (long) arg;	/* first arg */
327 }
328 
329 void
330 cpu_exit(struct thread *td)
331 {
332 
333 	/*
334 	 * If this process has a custom LDT, release it.
335 	 */
336 	if (td->td_proc->p_md.md_ldt != NULL)
337 		user_ldt_free(td);
338 }
339 
340 void
341 cpu_thread_exit(struct thread *td)
342 {
343 	struct pcb *pcb;
344 
345 	critical_enter();
346 	if (td == PCPU_GET(fpcurthread))
347 		fpudrop();
348 	critical_exit();
349 
350 	pcb = td->td_pcb;
351 
352 	/* Disable any hardware breakpoints. */
353 	if (pcb->pcb_flags & PCB_DBREGS) {
354 		reset_dbregs();
355 		clear_pcb_flags(pcb, PCB_DBREGS);
356 	}
357 }
358 
359 void
360 cpu_thread_clean(struct thread *td)
361 {
362 	struct pcb *pcb;
363 
364 	pcb = td->td_pcb;
365 
366 	/*
367 	 * Clean TSS/iomap
368 	 */
369 	if (pcb->pcb_tssp != NULL) {
370 		pmap_pti_remove_kva((vm_offset_t)pcb->pcb_tssp,
371 		    (vm_offset_t)pcb->pcb_tssp + ctob(IOPAGES + 1));
372 		kmem_free((vm_offset_t)pcb->pcb_tssp, ctob(IOPAGES + 1));
373 		pcb->pcb_tssp = NULL;
374 	}
375 }
376 
377 void
378 cpu_thread_swapin(struct thread *td)
379 {
380 }
381 
382 void
383 cpu_thread_swapout(struct thread *td)
384 {
385 }
386 
387 void
388 cpu_thread_alloc(struct thread *td)
389 {
390 	struct pcb *pcb;
391 	struct xstate_hdr *xhdr;
392 
393 	set_top_of_stack_td(td);
394 	td->td_pcb = pcb = get_pcb_td(td);
395 	td->td_frame = (struct trapframe *)td->td_md.md_stack_base - 1;
396 	pcb->pcb_save = get_pcb_user_save_pcb(pcb);
397 	if (use_xsave) {
398 		xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
399 		bzero(xhdr, sizeof(*xhdr));
400 		xhdr->xstate_bv = xsave_mask;
401 	}
402 }
403 
404 void
405 cpu_thread_free(struct thread *td)
406 {
407 
408 	cpu_thread_clean(td);
409 }
410 
411 bool
412 cpu_exec_vmspace_reuse(struct proc *p, vm_map_t map)
413 {
414 
415 	return (((curproc->p_md.md_flags & P_MD_KPTI) != 0) ==
416 	    (vm_map_pmap(map)->pm_ucr3 != PMAP_NO_CR3));
417 }
418 
419 static void
420 cpu_procctl_kpti_ctl(struct proc *p, int val)
421 {
422 
423 	if (pti && val == PROC_KPTI_CTL_ENABLE_ON_EXEC)
424 		p->p_md.md_flags |= P_MD_KPTI;
425 	if (val == PROC_KPTI_CTL_DISABLE_ON_EXEC)
426 		p->p_md.md_flags &= ~P_MD_KPTI;
427 }
428 
429 static void
430 cpu_procctl_kpti_status(struct proc *p, int *val)
431 {
432 	*val = (p->p_md.md_flags & P_MD_KPTI) != 0 ?
433 	    PROC_KPTI_CTL_ENABLE_ON_EXEC:
434 	    PROC_KPTI_CTL_DISABLE_ON_EXEC;
435 	if (vmspace_pmap(p->p_vmspace)->pm_ucr3 != PMAP_NO_CR3)
436 		*val |= PROC_KPTI_STATUS_ACTIVE;
437 }
438 
439 static int
440 cpu_procctl_la_ctl(struct proc *p, int val)
441 {
442 	int error;
443 
444 	error = 0;
445 	switch (val) {
446 	case PROC_LA_CTL_LA48_ON_EXEC:
447 		p->p_md.md_flags |= P_MD_LA48;
448 		p->p_md.md_flags &= ~P_MD_LA57;
449 		break;
450 	case PROC_LA_CTL_LA57_ON_EXEC:
451 		if (la57) {
452 			p->p_md.md_flags &= ~P_MD_LA48;
453 			p->p_md.md_flags |= P_MD_LA57;
454 		} else {
455 			error = ENOTSUP;
456 		}
457 		break;
458 	case PROC_LA_CTL_DEFAULT_ON_EXEC:
459 		p->p_md.md_flags &= ~(P_MD_LA48 | P_MD_LA57);
460 		break;
461 	}
462 	return (error);
463 }
464 
465 static void
466 cpu_procctl_la_status(struct proc *p, int *val)
467 {
468 	int res;
469 
470 	if ((p->p_md.md_flags & P_MD_LA48) != 0)
471 		res = PROC_LA_CTL_LA48_ON_EXEC;
472 	else if ((p->p_md.md_flags & P_MD_LA57) != 0)
473 		res = PROC_LA_CTL_LA57_ON_EXEC;
474 	else
475 		res = PROC_LA_CTL_DEFAULT_ON_EXEC;
476 	if (p->p_sysent->sv_maxuser == VM_MAXUSER_ADDRESS_LA48)
477 		res |= PROC_LA_STATUS_LA48;
478 	else
479 		res |= PROC_LA_STATUS_LA57;
480 	*val = res;
481 }
482 
483 int
484 cpu_procctl(struct thread *td, int idtype, id_t id, int com, void *data)
485 {
486 	struct proc *p;
487 	int error, val;
488 
489 	switch (com) {
490 	case PROC_KPTI_CTL:
491 	case PROC_KPTI_STATUS:
492 	case PROC_LA_CTL:
493 	case PROC_LA_STATUS:
494 		if (idtype != P_PID) {
495 			error = EINVAL;
496 			break;
497 		}
498 		if (com == PROC_KPTI_CTL) {
499 			/* sad but true and not a joke */
500 			error = priv_check(td, PRIV_IO);
501 			if (error != 0)
502 				break;
503 		}
504 		if (com == PROC_KPTI_CTL || com == PROC_LA_CTL) {
505 			error = copyin(data, &val, sizeof(val));
506 			if (error != 0)
507 				break;
508 		}
509 		if (com == PROC_KPTI_CTL &&
510 		    val != PROC_KPTI_CTL_ENABLE_ON_EXEC &&
511 		    val != PROC_KPTI_CTL_DISABLE_ON_EXEC) {
512 			error = EINVAL;
513 			break;
514 		}
515 		if (com == PROC_LA_CTL &&
516 		    val != PROC_LA_CTL_LA48_ON_EXEC &&
517 		    val != PROC_LA_CTL_LA57_ON_EXEC &&
518 		    val != PROC_LA_CTL_DEFAULT_ON_EXEC) {
519 			error = EINVAL;
520 			break;
521 		}
522 		error = pget(id, PGET_CANSEE | PGET_NOTWEXIT | PGET_NOTID, &p);
523 		if (error != 0)
524 			break;
525 		switch (com) {
526 		case PROC_KPTI_CTL:
527 			cpu_procctl_kpti_ctl(p, val);
528 			break;
529 		case PROC_KPTI_STATUS:
530 			cpu_procctl_kpti_status(p, &val);
531 			break;
532 		case PROC_LA_CTL:
533 			error = cpu_procctl_la_ctl(p, val);
534 			break;
535 		case PROC_LA_STATUS:
536 			cpu_procctl_la_status(p, &val);
537 			break;
538 		}
539 		PROC_UNLOCK(p);
540 		if (com == PROC_KPTI_STATUS || com == PROC_LA_STATUS)
541 			error = copyout(&val, data, sizeof(val));
542 		break;
543 	default:
544 		error = EINVAL;
545 		break;
546 	}
547 	return (error);
548 }
549 
550 void
551 cpu_set_syscall_retval(struct thread *td, int error)
552 {
553 	struct trapframe *frame;
554 
555 	frame = td->td_frame;
556 	if (__predict_true(error == 0)) {
557 		frame->tf_rax = td->td_retval[0];
558 		frame->tf_rdx = td->td_retval[1];
559 		frame->tf_rflags &= ~PSL_C;
560 		return;
561 	}
562 
563 	switch (error) {
564 	case ERESTART:
565 		/*
566 		 * Reconstruct pc, we know that 'syscall' is 2 bytes,
567 		 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
568 		 * We saved this in tf_err.
569 		 * %r10 (which was holding the value of %rcx) is restored
570 		 * for the next iteration.
571 		 * %r10 restore is only required for freebsd/amd64 processes,
572 		 * but shall be innocent for any ia32 ABI.
573 		 *
574 		 * Require full context restore to get the arguments
575 		 * in the registers reloaded at return to usermode.
576 		 */
577 		frame->tf_rip -= frame->tf_err;
578 		frame->tf_r10 = frame->tf_rcx;
579 		set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
580 		break;
581 
582 	case EJUSTRETURN:
583 		break;
584 
585 	default:
586 		frame->tf_rax = error;
587 		frame->tf_rflags |= PSL_C;
588 		break;
589 	}
590 }
591 
592 /*
593  * Initialize machine state, mostly pcb and trap frame for a new
594  * thread, about to return to userspace.  Put enough state in the new
595  * thread's PCB to get it to go back to the fork_return(), which
596  * finalizes the thread state and handles peculiarities of the first
597  * return to userspace for the new thread.
598  */
599 void
600 cpu_copy_thread(struct thread *td, struct thread *td0)
601 {
602 	copy_thread(td0, td);
603 
604 	/*
605 	 * Copy user general-purpose registers.
606 	 *
607 	 * Some of these registers are rewritten by cpu_set_upcall()
608 	 * and linux_set_upcall().
609 	 */
610 	bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
611 
612 	/* If the current thread has the trap bit set (i.e. a debugger had
613 	 * single stepped the process to the system call), we need to clear
614 	 * the trap flag from the new frame. Otherwise, the new thread will
615 	 * receive a (likely unexpected) SIGTRAP when it executes the first
616 	 * instruction after returning to userland.
617 	 */
618 	td->td_frame->tf_rflags &= ~PSL_T;
619 
620 	set_pcb_flags_raw(td->td_pcb, PCB_FULL_IRET);
621 }
622 
623 /*
624  * Set that machine state for performing an upcall that starts
625  * the entry function with the given argument.
626  */
627 void
628 cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg,
629     stack_t *stack)
630 {
631 
632 	/*
633 	 * Do any extra cleaning that needs to be done.
634 	 * The thread may have optional components
635 	 * that are not present in a fresh thread.
636 	 * This may be a recycled thread so make it look
637 	 * as though it's newly allocated.
638 	 */
639 	cpu_thread_clean(td);
640 
641 #ifdef COMPAT_FREEBSD32
642 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
643 		/*
644 		 * Set the trap frame to point at the beginning of the entry
645 		 * function.
646 		 */
647 		td->td_frame->tf_rbp = 0;
648 		td->td_frame->tf_rsp =
649 		   (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
650 		td->td_frame->tf_rip = (uintptr_t)entry;
651 
652 		/* Return address sentinel value to stop stack unwinding. */
653 		suword32((void *)td->td_frame->tf_rsp, 0);
654 
655 		/* Pass the argument to the entry point. */
656 		suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)),
657 		    (uint32_t)(uintptr_t)arg);
658 
659 		return;
660 	}
661 #endif
662 
663 	/*
664 	 * Set the trap frame to point at the beginning of the uts
665 	 * function.
666 	 */
667 	td->td_frame->tf_rbp = 0;
668 	td->td_frame->tf_rsp =
669 	    ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f;
670 	td->td_frame->tf_rsp -= 8;
671 	td->td_frame->tf_rip = (register_t)entry;
672 	td->td_frame->tf_ds = _udatasel;
673 	td->td_frame->tf_es = _udatasel;
674 	td->td_frame->tf_fs = _ufssel;
675 	td->td_frame->tf_gs = _ugssel;
676 	td->td_frame->tf_flags = TF_HASSEGS;
677 
678 	/* Return address sentinel value to stop stack unwinding. */
679 	suword((void *)td->td_frame->tf_rsp, 0);
680 
681 	/* Pass the argument to the entry point. */
682 	td->td_frame->tf_rdi = (register_t)arg;
683 }
684 
685 int
686 cpu_set_user_tls(struct thread *td, void *tls_base)
687 {
688 	struct pcb *pcb;
689 
690 	if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS)
691 		return (EINVAL);
692 
693 	pcb = td->td_pcb;
694 	set_pcb_flags(pcb, PCB_FULL_IRET);
695 #ifdef COMPAT_FREEBSD32
696 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
697 		pcb->pcb_gsbase = (register_t)tls_base;
698 		return (0);
699 	}
700 #endif
701 	pcb->pcb_fsbase = (register_t)tls_base;
702 	return (0);
703 }
704 
705 /*
706  * Software interrupt handler for queued VM system processing.
707  */
708 void
709 swi_vm(void *dummy)
710 {
711 	if (busdma_swi_pending != 0)
712 		busdma_swi();
713 }
714 
715 /*
716  * Tell whether this address is in some physical memory region.
717  * Currently used by the kernel coredump code in order to avoid
718  * dumping the ``ISA memory hole'' which could cause indefinite hangs,
719  * or other unpredictable behaviour.
720  */
721 
722 int
723 is_physical_memory(vm_paddr_t addr)
724 {
725 
726 #ifdef DEV_ISA
727 	/* The ISA ``memory hole''. */
728 	if (addr >= 0xa0000 && addr < 0x100000)
729 		return 0;
730 #endif
731 
732 	/*
733 	 * stuff other tests for known memory-mapped devices (PCI?)
734 	 * here
735 	 */
736 
737 	return 1;
738 }
739