xref: /freebsd/sys/arm64/linux/linux_sysvec.c (revision d27ba3088424e53eabc0b0186ed122ec43119501)
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 	/* Calculate string base and vector table pointers. */
259 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
260 		execpath_len = strlen(imgp->execpath) + 1;
261 	else
262 		execpath_len = 0;
263 
264 	p = imgp->proc;
265 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
266 	destp = (uintptr_t)arginfo;
267 
268 	if (execpath_len != 0) {
269 		destp -= execpath_len;
270 		destp = rounddown2(destp, sizeof(void *));
271 		imgp->execpathp = (void *)destp;
272 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
273 		if (error != 0)
274 			return (error);
275 	}
276 
277 	/* Prepare the canary for SSP. */
278 	arc4rand(canary, sizeof(canary), 0);
279 	destp -= roundup(sizeof(canary), sizeof(void *));
280 	imgp->canary = (void *)destp;
281 	error = copyout(canary, imgp->canary, sizeof(canary));
282 	if (error != 0)
283 		return (error);
284 
285 	/* Allocate room for the argument and environment strings. */
286 	destp -= ARG_MAX - imgp->args->stringspace;
287 	destp = rounddown2(destp, sizeof(void *));
288 	ustringp = destp;
289 
290 	if (imgp->auxargs) {
291 		/*
292 		 * Allocate room on the stack for the ELF auxargs
293 		 * array.  It has up to LINUX_AT_COUNT entries.
294 		 */
295 		destp -= LINUX_AT_COUNT * sizeof(Elf64_Auxinfo);
296 		destp = rounddown2(destp, sizeof(void *));
297 	}
298 
299 	vectp = (char **)destp;
300 
301 	/*
302 	 * Allocate room for argc and the argv[] and env vectors including the
303 	 * terminating NULL pointers.
304 	 */
305 	vectp -= 1 + imgp->args->argc + 1 + imgp->args->envc + 1;
306 	vectp = (char **)STACKALIGN(vectp);
307 
308 	/* vectp also becomes our initial stack base. */
309 	*stack_base = (uintptr_t)vectp;
310 
311 	stringp = imgp->args->begin_argv;
312 	argc = imgp->args->argc;
313 	envc = imgp->args->envc;
314 
315 	/* Copy out strings - arguments and environment. */
316 	error = copyout(stringp, (void *)ustringp,
317 	    ARG_MAX - imgp->args->stringspace);
318 	if (error != 0)
319 		return (error);
320 
321 	/* Fill in "ps_strings" struct for ps, w, etc. */
322 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
323 	    suword(&arginfo->ps_nargvstr, argc) != 0)
324 		return (EFAULT);
325 
326 	if (suword(vectp++, argc) != 0)
327 		return (EFAULT);
328 
329 	/* Fill in argument portion of vector table. */
330 	for (; argc > 0; --argc) {
331 		if (suword(vectp++, ustringp) != 0)
332 			return (EFAULT);
333 		while (*stringp++ != 0)
334 			ustringp++;
335 		ustringp++;
336 	}
337 
338 	/* A null vector table pointer separates the argp's from the envp's. */
339 	if (suword(vectp++, 0) != 0)
340 		return (EFAULT);
341 
342 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
343 	    suword(&arginfo->ps_nenvstr, envc) != 0)
344 		return (EFAULT);
345 
346 	/* Fill in environment portion of vector table. */
347 	for (; envc > 0; --envc) {
348 		if (suword(vectp++, ustringp) != 0)
349 			return (EFAULT);
350 		while (*stringp++ != 0)
351 			ustringp++;
352 		ustringp++;
353 	}
354 
355 	/* The end of the vector table is a null pointer. */
356 	if (suword(vectp, 0) != 0)
357 		return (EFAULT);
358 
359 	if (imgp->auxargs) {
360 		vectp++;
361 		error = imgp->sysent->sv_copyout_auxargs(imgp,
362 		    (uintptr_t)vectp);
363 		if (error != 0)
364 			return (error);
365 	}
366 
367 	return (0);
368 }
369 
370 /*
371  * Reset registers to default values on exec.
372  */
373 static void
374 linux_exec_setregs(struct thread *td, struct image_params *imgp,
375     uintptr_t stack)
376 {
377 	struct trapframe *regs = td->td_frame;
378 	struct pcb *pcb = td->td_pcb;
379 
380 	/* LINUXTODO: validate */
381 	LIN_SDT_PROBE0(sysvec, linux_exec_setregs, todo);
382 
383 	memset(regs, 0, sizeof(*regs));
384 	/* glibc start.S registers function pointer in x0 with atexit. */
385         regs->tf_sp = stack;
386 #if 0	/* LINUXTODO: See if this is used. */
387 	regs->tf_lr = imgp->entry_addr;
388 #else
389         regs->tf_lr = 0xffffffffffffffff;
390 #endif
391         regs->tf_elr = imgp->entry_addr;
392 
393 	pcb->pcb_tpidr_el0 = 0;
394 	pcb->pcb_tpidrro_el0 = 0;
395 	WRITE_SPECIALREG(tpidrro_el0, 0);
396 	WRITE_SPECIALREG(tpidr_el0, 0);
397 
398 #ifdef VFP
399 	vfp_reset_state(td, pcb);
400 #endif
401 
402 	/*
403 	 * Clear debug register state. It is not applicable to the new process.
404 	 */
405 	bzero(&pcb->pcb_dbg_regs, sizeof(pcb->pcb_dbg_regs));
406 }
407 
408 int
409 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
410 {
411 	struct l_sigframe frame;
412 	struct trapframe *tf;
413 	int error;
414 
415 	tf = td->td_frame;
416 
417 	if (copyin((void *)tf->tf_sp, &frame, sizeof(frame)))
418 		return (EFAULT);
419 
420 	error = set_mcontext(td, &frame.sf_uc.uc_mcontext);
421 	if (error != 0)
422 		return (error);
423 
424 	/* Restore signal mask. */
425 	kern_sigprocmask(td, SIG_SETMASK, &frame.sf_uc.uc_sigmask, NULL, 0);
426 
427 	return (EJUSTRETURN);
428 }
429 
430 static void
431 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
432 {
433 	struct thread *td;
434 	struct proc *p;
435 	struct trapframe *tf;
436 	struct l_sigframe *fp, frame;
437 	struct sigacts *psp;
438 	int onstack, sig;
439 
440 	td = curthread;
441 	p = td->td_proc;
442 	PROC_LOCK_ASSERT(p, MA_OWNED);
443 
444 	sig = ksi->ksi_signo;
445 	psp = p->p_sigacts;
446 	mtx_assert(&psp->ps_mtx, MA_OWNED);
447 
448 	tf = td->td_frame;
449 	onstack = sigonstack(tf->tf_sp);
450 
451 	CTR4(KTR_SIG, "sendsig: td=%p (%s) catcher=%p sig=%d", td, p->p_comm,
452 	    catcher, sig);
453 
454 	/* Allocate and validate space for the signal handler context. */
455 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !onstack &&
456 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
457 		fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
458 		    td->td_sigstk.ss_size);
459 #if defined(COMPAT_43)
460 		td->td_sigstk.ss_flags |= SS_ONSTACK;
461 #endif
462 	} else {
463 		fp = (struct l_sigframe *)td->td_frame->tf_sp;
464 	}
465 
466 	/* Make room, keeping the stack aligned */
467 	fp--;
468 	fp = (struct l_sigframe *)STACKALIGN(fp);
469 
470 	/* Fill in the frame to copy out */
471 	bzero(&frame, sizeof(frame));
472 	get_mcontext(td, &frame.sf_uc.uc_mcontext, 0);
473 
474 	/* Translate the signal. */
475 	sig = bsd_to_linux_signal(sig);
476 
477 	siginfo_to_lsiginfo(&ksi->ksi_info, &frame.sf_si, sig);
478 	frame.sf_uc.uc_sigmask = *mask;
479 	frame.sf_uc.uc_stack = td->td_sigstk;
480 	frame.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) != 0 ?
481 	    (onstack ? SS_ONSTACK : 0) : SS_DISABLE;
482 	mtx_unlock(&psp->ps_mtx);
483 	PROC_UNLOCK(td->td_proc);
484 
485 	/* Copy the sigframe out to the user's stack. */
486 	if (copyout(&frame, fp, sizeof(*fp)) != 0) {
487 		/* Process has trashed its stack. Kill it. */
488 		CTR2(KTR_SIG, "sendsig: sigexit td=%p fp=%p", td, fp);
489 		PROC_LOCK(p);
490 		sigexit(td, SIGILL);
491 	}
492 
493 	tf->tf_x[0]= sig;
494 	tf->tf_x[1] = (register_t)&fp->sf_si;
495 	tf->tf_x[2] = (register_t)&fp->sf_uc;
496 
497 	tf->tf_elr = (register_t)catcher;
498 	tf->tf_sp = (register_t)fp;
499 	tf->tf_lr = (register_t)__kernel_rt_sigreturn;
500 
501 	CTR3(KTR_SIG, "sendsig: return td=%p pc=%#x sp=%#x", td, tf->tf_elr,
502 	    tf->tf_sp);
503 
504 	PROC_LOCK(p);
505 	mtx_lock(&psp->ps_mtx);
506 }
507 
508 struct sysentvec elf_linux_sysvec = {
509 	.sv_size	= LINUX_SYS_MAXSYSCALL,
510 	.sv_table	= linux_sysent,
511 	.sv_transtrap	= linux_translate_traps,
512 	.sv_fixup	= linux_elf_fixup,
513 	.sv_sendsig	= linux_rt_sendsig,
514 	.sv_sigcode	= &_binary_linux_vdso_so_o_start,
515 	.sv_szsigcode	= &linux_szsigcode,
516 	.sv_name	= "Linux ELF64",
517 	.sv_coredump	= elf64_coredump,
518 	.sv_elf_core_osabi = ELFOSABI_NONE,
519 	.sv_elf_core_abi_vendor = LINUX_ABI_VENDOR,
520 	.sv_elf_core_prepare_notes = linux64_prepare_notes,
521 	.sv_imgact_try	= linux_exec_imgact_try,
522 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
523 	.sv_minuser	= VM_MIN_ADDRESS,
524 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
525 	.sv_usrstack	= LINUX_USRSTACK,
526 	.sv_psstrings	= LINUX_PS_STRINGS,
527 	.sv_stackprot	= VM_PROT_READ | VM_PROT_WRITE,
528 	.sv_copyout_auxargs = linux_copyout_auxargs,
529 	.sv_copyout_strings = linux_copyout_strings,
530 	.sv_setregs	= linux_exec_setregs,
531 	.sv_fixlimit	= NULL,
532 	.sv_maxssiz	= NULL,
533 	.sv_flags	= SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN |
534 	    SV_SIG_WAITNDQ | SV_TIMEKEEP,
535 	.sv_set_syscall_retval = linux_set_syscall_retval,
536 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
537 	.sv_syscallnames = NULL,
538 	.sv_shared_page_base = LINUX_SHAREDPAGE,
539 	.sv_shared_page_len = PAGE_SIZE,
540 	.sv_schedtail	= linux_schedtail,
541 	.sv_thread_detach = linux_thread_detach,
542 	.sv_trap	= NULL,
543 	.sv_hwcap	= &elf_hwcap,
544 	.sv_hwcap2	= &elf_hwcap2,
545 	.sv_onexec	= linux_on_exec_vmspace,
546 	.sv_onexit	= linux_on_exit,
547 	.sv_ontdexit	= linux_thread_dtor,
548 	.sv_setid_allowed = &linux_setid_allowed_query,
549 };
550 
551 static int
552 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp)
553 {
554 	int error;
555 
556 	error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base,
557 	    LINUX_VDSOPAGE_SIZE, imgp);
558 	if (error == 0)
559 		linux_on_exec(p, imgp);
560 	return (error);
561 }
562 
563 /*
564  * linux_vdso_install() and linux_exec_sysvec_init() must be called
565  * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY).
566  */
567 static void
568 linux_exec_sysvec_init(void *param)
569 {
570 	l_uintptr_t *ktimekeep_base;
571 	struct sysentvec *sv;
572 	ptrdiff_t tkoff;
573 
574 	sv = param;
575 	/* Fill timekeep_base */
576 	exec_sysvec_init(sv);
577 
578 	tkoff = kern_timekeep_base - linux_vdso_base;
579 	ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
580 	*ktimekeep_base = sv->sv_timekeep_base;
581 }
582 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY,
583     linux_exec_sysvec_init, &elf_linux_sysvec);
584 
585 static void
586 linux_vdso_install(const void *param)
587 {
588 	char *vdso_start = &_binary_linux_vdso_so_o_start;
589 	char *vdso_end = &_binary_linux_vdso_so_o_end;
590 
591 	linux_szsigcode = vdso_end - vdso_start;
592 	MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE);
593 
594 	linux_vdso_base = LINUX_VDSOPAGE;
595 
596 	__elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base);
597 
598 	linux_vdso_obj = __elfN(linux_shared_page_init)
599 	    (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
600 	bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode);
601 
602 	linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base);
603 }
604 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST,
605     linux_vdso_install, NULL);
606 
607 static void
608 linux_vdso_deinstall(const void *param)
609 {
610 
611 	__elfN(linux_shared_page_fini)(linux_vdso_obj,
612 	    linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
613 }
614 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
615     linux_vdso_deinstall, NULL);
616 
617 static void
618 linux_vdso_reloc(char *mapping, Elf_Addr offset)
619 {
620 	Elf_Size rtype, symidx;
621 	const Elf_Rela *rela;
622 	const Elf_Shdr *shdr;
623 	const Elf_Ehdr *ehdr;
624 	Elf_Addr *where;
625 	Elf_Addr addr, addend;
626 	int i, relacnt;
627 
628 	MPASS(offset != 0);
629 
630 	relacnt = 0;
631 	ehdr = (const Elf_Ehdr *)mapping;
632 	shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff);
633 	for (i = 0; i < ehdr->e_shnum; i++)
634 	{
635 		switch (shdr[i].sh_type) {
636 		case SHT_REL:
637 			printf("Linux Aarch64 vDSO: unexpected Rel section\n");
638 			break;
639 		case SHT_RELA:
640 			rela = (const Elf_Rela *)(mapping + shdr[i].sh_offset);
641 			relacnt = shdr[i].sh_size / sizeof(*rela);
642 		}
643 	}
644 
645 	for (i = 0; i < relacnt; i++, rela++) {
646 		where = (Elf_Addr *)(mapping + rela->r_offset);
647 		addend = rela->r_addend;
648 		rtype = ELF_R_TYPE(rela->r_info);
649 		symidx = ELF_R_SYM(rela->r_info);
650 
651 		switch (rtype) {
652 		case R_AARCH64_NONE:	/* none */
653 			break;
654 
655 		case R_AARCH64_RELATIVE:	/* B + A */
656 			addr = (Elf_Addr)(mapping + addend);
657 			if (*where != addr)
658 				*where = addr;
659 			break;
660 		default:
661 			printf("Linux Aarch64 vDSO: unexpected relocation type %ld, "
662 			    "symbol index %ld\n", rtype, symidx);
663 		}
664 	}
665 }
666 
667 static char GNU_ABI_VENDOR[] = "GNU";
668 static int GNU_ABI_LINUX = 0;
669 
670 /* LINUXTODO: deduplicate */
671 static bool
672 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
673 {
674 	const Elf32_Word *desc;
675 	uintptr_t p;
676 
677 	p = (uintptr_t)(note + 1);
678 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
679 
680 	desc = (const Elf32_Word *)p;
681 	if (desc[0] != GNU_ABI_LINUX)
682 		return (false);
683 
684 	*osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
685 	return (true);
686 }
687 
688 static Elf_Brandnote linux64_brandnote = {
689 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
690 	.hdr.n_descsz	= 16,
691 	.hdr.n_type	= 1,
692 	.vendor		= GNU_ABI_VENDOR,
693 	.flags		= BN_TRANSLATE_OSREL,
694 	.trans_osrel	= linux_trans_osrel
695 };
696 
697 static Elf64_Brandinfo linux_glibc2brand = {
698 	.brand		= ELFOSABI_LINUX,
699 	.machine	= EM_AARCH64,
700 	.compat_3_brand	= "Linux",
701 	.emul_path	= linux_emul_path,
702 	.interp_path	= "/lib64/ld-linux-x86-64.so.2",
703 	.sysvec		= &elf_linux_sysvec,
704 	.interp_newpath	= NULL,
705 	.brand_note	= &linux64_brandnote,
706 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
707 };
708 
709 Elf64_Brandinfo *linux_brandlist[] = {
710 	&linux_glibc2brand,
711 	NULL
712 };
713 
714 static int
715 linux64_elf_modevent(module_t mod, int type, void *data)
716 {
717 	Elf64_Brandinfo **brandinfo;
718 	struct linux_ioctl_handler**lihp;
719 	int error;
720 
721 	error = 0;
722 	switch(type) {
723 	case MOD_LOAD:
724 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
725 		    ++brandinfo)
726 			if (elf64_insert_brand_entry(*brandinfo) < 0)
727 				error = EINVAL;
728 		if (error == 0) {
729 			SET_FOREACH(lihp, linux_ioctl_handler_set)
730 				linux_ioctl_register_handler(*lihp);
731 			stclohz = (stathz ? stathz : hz);
732 			if (bootverbose)
733 				printf("Linux arm64 ELF exec handler installed\n");
734 		}
735 		break;
736 	case MOD_UNLOAD:
737 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
738 		    ++brandinfo)
739 			if (elf64_brand_inuse(*brandinfo))
740 				error = EBUSY;
741 		if (error == 0) {
742 			for (brandinfo = &linux_brandlist[0];
743 			    *brandinfo != NULL; ++brandinfo)
744 				if (elf64_remove_brand_entry(*brandinfo) < 0)
745 					error = EINVAL;
746 		}
747 		if (error == 0) {
748 			SET_FOREACH(lihp, linux_ioctl_handler_set)
749 				linux_ioctl_unregister_handler(*lihp);
750 			if (bootverbose)
751 				printf("Linux arm64 ELF exec handler removed\n");
752 		} else
753 			printf("Could not deinstall Linux arm64 ELF interpreter entry\n");
754 		break;
755 	default:
756 		return (EOPNOTSUPP);
757 	}
758 	return (error);
759 }
760 
761 static moduledata_t linux64_elf_mod = {
762 	"linux64elf",
763 	linux64_elf_modevent,
764 	0
765 };
766 
767 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
768 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
769 FEATURE(linux64, "AArch64 Linux 64bit support");
770