xref: /freebsd/sys/arm64/linux/linux_sysvec.c (revision 7be9a3b45356747f9fcb6d69a722c1c95f8060bf)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1994-1996 Søren Schmidt
5  * Copyright (c) 2018 Turing Robotic Industries Inc.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/cdefs.h>
35 #include <sys/elf.h>
36 #include <sys/exec.h>
37 #include <sys/imgact.h>
38 #include <sys/imgact_elf.h>
39 #include <sys/kernel.h>
40 #include <sys/ktr.h>
41 #include <sys/lock.h>
42 #include <sys/module.h>
43 #include <sys/mutex.h>
44 #include <sys/proc.h>
45 #include <sys/stddef.h>
46 #include <sys/signalvar.h>
47 #include <sys/syscallsubr.h>
48 #include <sys/sysctl.h>
49 #include <sys/sysent.h>
50 
51 #include <vm/vm.h>
52 #include <vm/pmap.h>
53 #include <vm/vm_map.h>
54 #include <vm/vm_extern.h>
55 #include <vm/vm_object.h>
56 #include <vm/vm_page.h>
57 #include <vm/vm_param.h>
58 
59 #include <arm64/linux/linux.h>
60 #include <arm64/linux/linux_proto.h>
61 #include <compat/linux/linux_dtrace.h>
62 #include <compat/linux/linux_emul.h>
63 #include <compat/linux/linux_fork.h>
64 #include <compat/linux/linux_ioctl.h>
65 #include <compat/linux/linux_mib.h>
66 #include <compat/linux/linux_misc.h>
67 #include <compat/linux/linux_signal.h>
68 #include <compat/linux/linux_util.h>
69 #include <compat/linux/linux_vdso.h>
70 
71 #include <machine/md_var.h>
72 
73 #ifdef VFP
74 #include <machine/vfp.h>
75 #endif
76 
77 MODULE_VERSION(linux64elf, 1);
78 
79 #define	LINUX_VDSOPAGE_SIZE	PAGE_SIZE * 2
80 #define	LINUX_VDSOPAGE		(VM_MAXUSER_ADDRESS - \
81 				    LINUX_VDSOPAGE_SIZE)
82 #define	LINUX_SHAREDPAGE	(LINUX_VDSOPAGE - PAGE_SIZE)
83 				/*
84 				 * PAGE_SIZE - the size
85 				 * of the native SHAREDPAGE
86 				 */
87 #define	LINUX_USRSTACK		LINUX_SHAREDPAGE
88 #define	LINUX_PS_STRINGS	(LINUX_USRSTACK - \
89 				    sizeof(struct ps_strings))
90 
91 static int linux_szsigcode;
92 static vm_object_t linux_vdso_obj;
93 static char *linux_vdso_mapping;
94 extern char _binary_linux_vdso_so_o_start;
95 extern char _binary_linux_vdso_so_o_end;
96 static vm_offset_t linux_vdso_base;
97 
98 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
99 
100 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
101 
102 static int	linux_copyout_strings(struct image_params *imgp,
103 		    uintptr_t *stack_base);
104 static int	linux_elf_fixup(uintptr_t *stack_base,
105 		    struct image_params *iparams);
106 static bool	linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
107 static void	linux_vdso_install(const void *param);
108 static void	linux_vdso_deinstall(const void *param);
109 static void	linux_vdso_reloc(char *mapping, Elf_Addr offset);
110 static void	linux_set_syscall_retval(struct thread *td, int error);
111 static int	linux_fetch_syscall_args(struct thread *td);
112 static void	linux_exec_setregs(struct thread *td, struct image_params *imgp,
113 		    uintptr_t stack);
114 static void	linux_exec_sysvec_init(void *param);
115 static int	linux_on_exec_vmspace(struct proc *p,
116 		    struct image_params *imgp);
117 
118 /* DTrace init */
119 LIN_SDT_PROVIDER_DECLARE(LINUX_DTRACE);
120 
121 /* DTrace probes */
122 LIN_SDT_PROBE_DEFINE2(sysvec, linux_translate_traps, todo, "int", "int");
123 LIN_SDT_PROBE_DEFINE0(sysvec, linux_exec_setregs, todo);
124 LIN_SDT_PROBE_DEFINE0(sysvec, linux_copyout_auxargs, todo);
125 LIN_SDT_PROBE_DEFINE0(sysvec, linux_elf_fixup, todo);
126 
127 LINUX_VDSO_SYM_CHAR(linux_platform);
128 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base);
129 LINUX_VDSO_SYM_INTPTR(__kernel_rt_sigreturn);
130 
131 /* LINUXTODO: do we have traps to translate? */
132 static int
133 linux_translate_traps(int signal, int trap_code)
134 {
135 
136 	LIN_SDT_PROBE2(sysvec, linux_translate_traps, todo, signal, trap_code);
137 	return (signal);
138 }
139 
140 static int
141 linux_fetch_syscall_args(struct thread *td)
142 {
143 	struct proc *p;
144 	struct syscall_args *sa;
145 	register_t *ap;
146 
147 	p = td->td_proc;
148 	ap = td->td_frame->tf_x;
149 	sa = &td->td_sa;
150 
151 	sa->code = td->td_frame->tf_x[8];
152 	sa->original_code = sa->code;
153 	/* LINUXTODO: generic syscall? */
154 	if (sa->code >= p->p_sysent->sv_size)
155 		sa->callp = &p->p_sysent->sv_table[0];
156 	else
157 		sa->callp = &p->p_sysent->sv_table[sa->code];
158 
159 	if (sa->callp->sy_narg > nitems(sa->args))
160 		panic("ARM64TODO: Could we have more than %zu args?",
161 		    nitems(sa->args));
162 	memcpy(sa->args, ap, nitems(sa->args) * sizeof(register_t));
163 
164 	td->td_retval[0] = 0;
165 	return (0);
166 }
167 
168 static void
169 linux_set_syscall_retval(struct thread *td, int error)
170 {
171 
172 	td->td_retval[1] = td->td_frame->tf_x[1];
173 	cpu_set_syscall_retval(td, error);
174 
175 	if (__predict_false(error != 0)) {
176 		if (error != ERESTART && error != EJUSTRETURN)
177 			td->td_frame->tf_x[0] = bsd_to_linux_errno(error);
178 	}
179 }
180 
181 static int
182 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
183 {
184 	Elf_Auxargs *args;
185 	Elf_Auxinfo *argarray, *pos;
186 	struct proc *p;
187 	int error, issetugid;
188 
189 	LIN_SDT_PROBE0(sysvec, linux_copyout_auxargs, todo);
190 	p = imgp->proc;
191 
192 	args = (Elf64_Auxargs *)imgp->auxargs;
193 	argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
194 	    M_WAITOK | M_ZERO);
195 
196 	issetugid = p->p_flag & P_SUGID ? 1 : 0;
197 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, linux_vdso_base);
198 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, *imgp->sysent->sv_hwcap);
199 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
200 	AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
201 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
202 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
203 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
204 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
205 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
206 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
207 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
208 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
209 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
210 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
211 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
212 	AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary);
213 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, *imgp->sysent->sv_hwcap2);
214 	if (imgp->execpathp != 0)
215 		AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp);
216 	if (args->execfd != -1)
217 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
218 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
219 	AUXARGS_ENTRY(pos, AT_NULL, 0);
220 
221 	free(imgp->auxargs, M_TEMP);
222 	imgp->auxargs = NULL;
223 	KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
224 
225 	error = copyout(argarray, (void *)base,
226 	    sizeof(*argarray) * LINUX_AT_COUNT);
227 	free(argarray, M_TEMP);
228 	return (error);
229 }
230 
231 static int
232 linux_elf_fixup(uintptr_t *stack_base, struct image_params *imgp)
233 {
234 
235 	LIN_SDT_PROBE0(sysvec, linux_elf_fixup, todo);
236 
237 	return (0);
238 }
239 
240 /*
241  * Copy strings out to the new process address space, constructing new arg
242  * and env vector tables. Return a pointer to the base so that it can be used
243  * as the initial stack pointer.
244  * LINUXTODO: deduplicate against other linuxulator archs
245  */
246 static int
247 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
248 {
249 	char **vectp;
250 	char *stringp;
251 	uintptr_t destp, ustringp;
252 	struct ps_strings *arginfo;
253 	char canary[LINUX_AT_RANDOM_LEN];
254 	size_t execpath_len;
255 	struct proc *p;
256 	int argc, envc, error;
257 
258 	p = imgp->proc;
259 	arginfo = (struct ps_strings *)PROC_PS_STRINGS(p);
260 	destp = (uintptr_t)arginfo;
261 
262 	if (imgp->execpath != NULL && imgp->auxargs != NULL) {
263 		execpath_len = strlen(imgp->execpath) + 1;
264 		destp -= execpath_len;
265 		destp = rounddown2(destp, sizeof(void *));
266 		imgp->execpathp = (void *)destp;
267 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
268 		if (error != 0)
269 			return (error);
270 	}
271 
272 	/* Prepare the canary for SSP. */
273 	arc4rand(canary, sizeof(canary), 0);
274 	destp -= roundup(sizeof(canary), sizeof(void *));
275 	imgp->canary = (void *)destp;
276 	error = copyout(canary, imgp->canary, sizeof(canary));
277 	if (error != 0)
278 		return (error);
279 
280 	/* Allocate room for the argument and environment strings. */
281 	destp -= ARG_MAX - imgp->args->stringspace;
282 	destp = rounddown2(destp, sizeof(void *));
283 	ustringp = destp;
284 
285 	if (imgp->auxargs) {
286 		/*
287 		 * Allocate room on the stack for the ELF auxargs
288 		 * array.  It has up to LINUX_AT_COUNT entries.
289 		 */
290 		destp -= LINUX_AT_COUNT * sizeof(Elf64_Auxinfo);
291 		destp = rounddown2(destp, sizeof(void *));
292 	}
293 
294 	vectp = (char **)destp;
295 
296 	/*
297 	 * Allocate room for argc and the argv[] and env vectors including the
298 	 * terminating NULL pointers.
299 	 */
300 	vectp -= 1 + imgp->args->argc + 1 + imgp->args->envc + 1;
301 	vectp = (char **)STACKALIGN(vectp);
302 
303 	/* vectp also becomes our initial stack base. */
304 	*stack_base = (uintptr_t)vectp;
305 
306 	stringp = imgp->args->begin_argv;
307 	argc = imgp->args->argc;
308 	envc = imgp->args->envc;
309 
310 	/* Copy out strings - arguments and environment. */
311 	error = copyout(stringp, (void *)ustringp,
312 	    ARG_MAX - imgp->args->stringspace);
313 	if (error != 0)
314 		return (error);
315 
316 	/* Fill in "ps_strings" struct for ps, w, etc. */
317 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
318 	    suword(&arginfo->ps_nargvstr, argc) != 0)
319 		return (EFAULT);
320 
321 	if (suword(vectp++, argc) != 0)
322 		return (EFAULT);
323 
324 	/* Fill in argument portion of vector table. */
325 	for (; argc > 0; --argc) {
326 		if (suword(vectp++, ustringp) != 0)
327 			return (EFAULT);
328 		while (*stringp++ != 0)
329 			ustringp++;
330 		ustringp++;
331 	}
332 
333 	/* A null vector table pointer separates the argp's from the envp's. */
334 	if (suword(vectp++, 0) != 0)
335 		return (EFAULT);
336 
337 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
338 	    suword(&arginfo->ps_nenvstr, envc) != 0)
339 		return (EFAULT);
340 
341 	/* Fill in environment portion of vector table. */
342 	for (; envc > 0; --envc) {
343 		if (suword(vectp++, ustringp) != 0)
344 			return (EFAULT);
345 		while (*stringp++ != 0)
346 			ustringp++;
347 		ustringp++;
348 	}
349 
350 	/* The end of the vector table is a null pointer. */
351 	if (suword(vectp, 0) != 0)
352 		return (EFAULT);
353 
354 	if (imgp->auxargs) {
355 		vectp++;
356 		error = imgp->sysent->sv_copyout_auxargs(imgp,
357 		    (uintptr_t)vectp);
358 		if (error != 0)
359 			return (error);
360 	}
361 
362 	return (0);
363 }
364 
365 /*
366  * Reset registers to default values on exec.
367  */
368 static void
369 linux_exec_setregs(struct thread *td, struct image_params *imgp,
370     uintptr_t stack)
371 {
372 	struct trapframe *regs = td->td_frame;
373 	struct pcb *pcb = td->td_pcb;
374 
375 	/* LINUXTODO: validate */
376 	LIN_SDT_PROBE0(sysvec, linux_exec_setregs, todo);
377 
378 	memset(regs, 0, sizeof(*regs));
379 	/* glibc start.S registers function pointer in x0 with atexit. */
380         regs->tf_sp = stack;
381 #if 0	/* LINUXTODO: See if this is used. */
382 	regs->tf_lr = imgp->entry_addr;
383 #else
384         regs->tf_lr = 0xffffffffffffffff;
385 #endif
386         regs->tf_elr = imgp->entry_addr;
387 
388 	pcb->pcb_tpidr_el0 = 0;
389 	pcb->pcb_tpidrro_el0 = 0;
390 	WRITE_SPECIALREG(tpidrro_el0, 0);
391 	WRITE_SPECIALREG(tpidr_el0, 0);
392 
393 #ifdef VFP
394 	vfp_reset_state(td, pcb);
395 #endif
396 
397 	/*
398 	 * Clear debug register state. It is not applicable to the new process.
399 	 */
400 	bzero(&pcb->pcb_dbg_regs, sizeof(pcb->pcb_dbg_regs));
401 }
402 
403 int
404 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
405 {
406 	struct l_sigframe frame;
407 	struct trapframe *tf;
408 	int error;
409 
410 	tf = td->td_frame;
411 
412 	if (copyin((void *)tf->tf_sp, &frame, sizeof(frame)))
413 		return (EFAULT);
414 
415 	error = set_mcontext(td, &frame.sf_uc.uc_mcontext);
416 	if (error != 0)
417 		return (error);
418 
419 	/* Restore signal mask. */
420 	kern_sigprocmask(td, SIG_SETMASK, &frame.sf_uc.uc_sigmask, NULL, 0);
421 
422 	return (EJUSTRETURN);
423 }
424 
425 static void
426 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
427 {
428 	struct thread *td;
429 	struct proc *p;
430 	struct trapframe *tf;
431 	struct l_sigframe *fp, frame;
432 	struct sigacts *psp;
433 	int onstack, sig;
434 
435 	td = curthread;
436 	p = td->td_proc;
437 	PROC_LOCK_ASSERT(p, MA_OWNED);
438 
439 	sig = ksi->ksi_signo;
440 	psp = p->p_sigacts;
441 	mtx_assert(&psp->ps_mtx, MA_OWNED);
442 
443 	tf = td->td_frame;
444 	onstack = sigonstack(tf->tf_sp);
445 
446 	CTR4(KTR_SIG, "sendsig: td=%p (%s) catcher=%p sig=%d", td, p->p_comm,
447 	    catcher, sig);
448 
449 	/* Allocate and validate space for the signal handler context. */
450 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !onstack &&
451 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
452 		fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
453 		    td->td_sigstk.ss_size);
454 #if defined(COMPAT_43)
455 		td->td_sigstk.ss_flags |= SS_ONSTACK;
456 #endif
457 	} else {
458 		fp = (struct l_sigframe *)td->td_frame->tf_sp;
459 	}
460 
461 	/* Make room, keeping the stack aligned */
462 	fp--;
463 	fp = (struct l_sigframe *)STACKALIGN(fp);
464 
465 	/* Fill in the frame to copy out */
466 	bzero(&frame, sizeof(frame));
467 	get_mcontext(td, &frame.sf_uc.uc_mcontext, 0);
468 
469 	/* Translate the signal. */
470 	sig = bsd_to_linux_signal(sig);
471 
472 	siginfo_to_lsiginfo(&ksi->ksi_info, &frame.sf_si, sig);
473 	frame.sf_uc.uc_sigmask = *mask;
474 	frame.sf_uc.uc_stack = td->td_sigstk;
475 	frame.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) != 0 ?
476 	    (onstack ? SS_ONSTACK : 0) : SS_DISABLE;
477 	mtx_unlock(&psp->ps_mtx);
478 	PROC_UNLOCK(td->td_proc);
479 
480 	/* Copy the sigframe out to the user's stack. */
481 	if (copyout(&frame, fp, sizeof(*fp)) != 0) {
482 		/* Process has trashed its stack. Kill it. */
483 		CTR2(KTR_SIG, "sendsig: sigexit td=%p fp=%p", td, fp);
484 		PROC_LOCK(p);
485 		sigexit(td, SIGILL);
486 	}
487 
488 	tf->tf_x[0]= sig;
489 	tf->tf_x[1] = (register_t)&fp->sf_si;
490 	tf->tf_x[2] = (register_t)&fp->sf_uc;
491 
492 	tf->tf_elr = (register_t)catcher;
493 	tf->tf_sp = (register_t)fp;
494 	tf->tf_lr = (register_t)__kernel_rt_sigreturn;
495 
496 	CTR3(KTR_SIG, "sendsig: return td=%p pc=%#x sp=%#x", td, tf->tf_elr,
497 	    tf->tf_sp);
498 
499 	PROC_LOCK(p);
500 	mtx_lock(&psp->ps_mtx);
501 }
502 
503 struct sysentvec elf_linux_sysvec = {
504 	.sv_size	= LINUX_SYS_MAXSYSCALL,
505 	.sv_table	= linux_sysent,
506 	.sv_transtrap	= linux_translate_traps,
507 	.sv_fixup	= linux_elf_fixup,
508 	.sv_sendsig	= linux_rt_sendsig,
509 	.sv_sigcode	= &_binary_linux_vdso_so_o_start,
510 	.sv_szsigcode	= &linux_szsigcode,
511 	.sv_name	= "Linux ELF64",
512 	.sv_coredump	= elf64_coredump,
513 	.sv_elf_core_osabi = ELFOSABI_NONE,
514 	.sv_elf_core_abi_vendor = LINUX_ABI_VENDOR,
515 	.sv_elf_core_prepare_notes = linux64_prepare_notes,
516 	.sv_imgact_try	= linux_exec_imgact_try,
517 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
518 	.sv_minuser	= VM_MIN_ADDRESS,
519 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
520 	.sv_usrstack	= LINUX_USRSTACK,
521 	.sv_psstrings	= LINUX_PS_STRINGS,
522 	.sv_psstringssz	= sizeof(struct ps_strings),
523 	.sv_stackprot	= VM_PROT_READ | VM_PROT_WRITE,
524 	.sv_copyout_auxargs = linux_copyout_auxargs,
525 	.sv_copyout_strings = linux_copyout_strings,
526 	.sv_setregs	= linux_exec_setregs,
527 	.sv_fixlimit	= NULL,
528 	.sv_maxssiz	= NULL,
529 	.sv_flags	= SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN |
530 	    SV_SIG_WAITNDQ | SV_TIMEKEEP,
531 	.sv_set_syscall_retval = linux_set_syscall_retval,
532 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
533 	.sv_syscallnames = NULL,
534 	.sv_shared_page_base = LINUX_SHAREDPAGE,
535 	.sv_shared_page_len = PAGE_SIZE,
536 	.sv_schedtail	= linux_schedtail,
537 	.sv_thread_detach = linux_thread_detach,
538 	.sv_trap	= NULL,
539 	.sv_hwcap	= &elf_hwcap,
540 	.sv_hwcap2	= &elf_hwcap2,
541 	.sv_onexec	= linux_on_exec_vmspace,
542 	.sv_onexit	= linux_on_exit,
543 	.sv_ontdexit	= linux_thread_dtor,
544 	.sv_setid_allowed = &linux_setid_allowed_query,
545 };
546 
547 static int
548 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp)
549 {
550 	int error;
551 
552 	error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base,
553 	    LINUX_VDSOPAGE_SIZE, imgp);
554 	if (error == 0)
555 		linux_on_exec(p, imgp);
556 	return (error);
557 }
558 
559 /*
560  * linux_vdso_install() and linux_exec_sysvec_init() must be called
561  * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY).
562  */
563 static void
564 linux_exec_sysvec_init(void *param)
565 {
566 	l_uintptr_t *ktimekeep_base;
567 	struct sysentvec *sv;
568 	ptrdiff_t tkoff;
569 
570 	sv = param;
571 	/* Fill timekeep_base */
572 	exec_sysvec_init(sv);
573 
574 	tkoff = kern_timekeep_base - linux_vdso_base;
575 	ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
576 	*ktimekeep_base = sv->sv_timekeep_base;
577 }
578 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY,
579     linux_exec_sysvec_init, &elf_linux_sysvec);
580 
581 static void
582 linux_vdso_install(const void *param)
583 {
584 	char *vdso_start = &_binary_linux_vdso_so_o_start;
585 	char *vdso_end = &_binary_linux_vdso_so_o_end;
586 
587 	linux_szsigcode = vdso_end - vdso_start;
588 	MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE);
589 
590 	linux_vdso_base = LINUX_VDSOPAGE;
591 
592 	__elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base);
593 
594 	linux_vdso_obj = __elfN(linux_shared_page_init)
595 	    (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
596 	bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode);
597 
598 	linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base);
599 }
600 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST,
601     linux_vdso_install, NULL);
602 
603 static void
604 linux_vdso_deinstall(const void *param)
605 {
606 
607 	__elfN(linux_shared_page_fini)(linux_vdso_obj,
608 	    linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
609 }
610 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
611     linux_vdso_deinstall, NULL);
612 
613 static void
614 linux_vdso_reloc(char *mapping, Elf_Addr offset)
615 {
616 	Elf_Size rtype, symidx;
617 	const Elf_Rela *rela;
618 	const Elf_Shdr *shdr;
619 	const Elf_Ehdr *ehdr;
620 	Elf_Addr *where;
621 	Elf_Addr addr, addend;
622 	int i, relacnt;
623 
624 	MPASS(offset != 0);
625 
626 	relacnt = 0;
627 	ehdr = (const Elf_Ehdr *)mapping;
628 	shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff);
629 	for (i = 0; i < ehdr->e_shnum; i++)
630 	{
631 		switch (shdr[i].sh_type) {
632 		case SHT_REL:
633 			printf("Linux Aarch64 vDSO: unexpected Rel section\n");
634 			break;
635 		case SHT_RELA:
636 			rela = (const Elf_Rela *)(mapping + shdr[i].sh_offset);
637 			relacnt = shdr[i].sh_size / sizeof(*rela);
638 		}
639 	}
640 
641 	for (i = 0; i < relacnt; i++, rela++) {
642 		where = (Elf_Addr *)(mapping + rela->r_offset);
643 		addend = rela->r_addend;
644 		rtype = ELF_R_TYPE(rela->r_info);
645 		symidx = ELF_R_SYM(rela->r_info);
646 
647 		switch (rtype) {
648 		case R_AARCH64_NONE:	/* none */
649 			break;
650 
651 		case R_AARCH64_RELATIVE:	/* B + A */
652 			addr = (Elf_Addr)(mapping + addend);
653 			if (*where != addr)
654 				*where = addr;
655 			break;
656 		default:
657 			printf("Linux Aarch64 vDSO: unexpected relocation type %ld, "
658 			    "symbol index %ld\n", rtype, symidx);
659 		}
660 	}
661 }
662 
663 static char GNU_ABI_VENDOR[] = "GNU";
664 static int GNU_ABI_LINUX = 0;
665 
666 /* LINUXTODO: deduplicate */
667 static bool
668 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
669 {
670 	const Elf32_Word *desc;
671 	uintptr_t p;
672 
673 	p = (uintptr_t)(note + 1);
674 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
675 
676 	desc = (const Elf32_Word *)p;
677 	if (desc[0] != GNU_ABI_LINUX)
678 		return (false);
679 
680 	*osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
681 	return (true);
682 }
683 
684 static Elf_Brandnote linux64_brandnote = {
685 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
686 	.hdr.n_descsz	= 16,
687 	.hdr.n_type	= 1,
688 	.vendor		= GNU_ABI_VENDOR,
689 	.flags		= BN_TRANSLATE_OSREL,
690 	.trans_osrel	= linux_trans_osrel
691 };
692 
693 static Elf64_Brandinfo linux_glibc2brand = {
694 	.brand		= ELFOSABI_LINUX,
695 	.machine	= EM_AARCH64,
696 	.compat_3_brand	= "Linux",
697 	.emul_path	= linux_emul_path,
698 	.interp_path	= "/lib64/ld-linux-x86-64.so.2",
699 	.sysvec		= &elf_linux_sysvec,
700 	.interp_newpath	= NULL,
701 	.brand_note	= &linux64_brandnote,
702 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
703 };
704 
705 Elf64_Brandinfo *linux_brandlist[] = {
706 	&linux_glibc2brand,
707 	NULL
708 };
709 
710 static int
711 linux64_elf_modevent(module_t mod, int type, void *data)
712 {
713 	Elf64_Brandinfo **brandinfo;
714 	struct linux_ioctl_handler**lihp;
715 	int error;
716 
717 	error = 0;
718 	switch(type) {
719 	case MOD_LOAD:
720 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
721 		    ++brandinfo)
722 			if (elf64_insert_brand_entry(*brandinfo) < 0)
723 				error = EINVAL;
724 		if (error == 0) {
725 			SET_FOREACH(lihp, linux_ioctl_handler_set)
726 				linux_ioctl_register_handler(*lihp);
727 			stclohz = (stathz ? stathz : hz);
728 			if (bootverbose)
729 				printf("Linux arm64 ELF exec handler installed\n");
730 		}
731 		break;
732 	case MOD_UNLOAD:
733 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
734 		    ++brandinfo)
735 			if (elf64_brand_inuse(*brandinfo))
736 				error = EBUSY;
737 		if (error == 0) {
738 			for (brandinfo = &linux_brandlist[0];
739 			    *brandinfo != NULL; ++brandinfo)
740 				if (elf64_remove_brand_entry(*brandinfo) < 0)
741 					error = EINVAL;
742 		}
743 		if (error == 0) {
744 			SET_FOREACH(lihp, linux_ioctl_handler_set)
745 				linux_ioctl_unregister_handler(*lihp);
746 			if (bootverbose)
747 				printf("Linux arm64 ELF exec handler removed\n");
748 		} else
749 			printf("Could not deinstall Linux arm64 ELF interpreter entry\n");
750 		break;
751 	default:
752 		return (EOPNOTSUPP);
753 	}
754 	return (error);
755 }
756 
757 static moduledata_t linux64_elf_mod = {
758 	"linux64elf",
759 	linux64_elf_modevent,
760 	0
761 };
762 
763 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
764 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
765 FEATURE(linux64, "AArch64 Linux 64bit support");
766