xref: /freebsd/sys/arm64/linux/linux_sysvec.c (revision 7446514533a40b376eaeb349ea33531ce9c711c2)
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 #define	__ELF_WORD_SIZE	64
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/cdefs.h>
37 #include <sys/elf.h>
38 #include <sys/exec.h>
39 #include <sys/imgact.h>
40 #include <sys/imgact_elf.h>
41 #include <sys/kernel.h>
42 #include <sys/ktr.h>
43 #include <sys/lock.h>
44 #include <sys/module.h>
45 #include <sys/mutex.h>
46 #include <sys/proc.h>
47 #include <sys/stddef.h>
48 #include <sys/signalvar.h>
49 #include <sys/syscallsubr.h>
50 #include <sys/sysctl.h>
51 #include <sys/sysent.h>
52 
53 #include <vm/vm.h>
54 #include <vm/pmap.h>
55 #include <vm/vm_map.h>
56 #include <vm/vm_extern.h>
57 #include <vm/vm_object.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_param.h>
60 
61 #include <arm64/linux/linux.h>
62 #include <arm64/linux/linux_proto.h>
63 #include <compat/linux/linux_dtrace.h>
64 #include <compat/linux/linux_elf.h>
65 #include <compat/linux/linux_emul.h>
66 #include <compat/linux/linux_fork.h>
67 #include <compat/linux/linux_ioctl.h>
68 #include <compat/linux/linux_mib.h>
69 #include <compat/linux/linux_misc.h>
70 #include <compat/linux/linux_signal.h>
71 #include <compat/linux/linux_util.h>
72 #include <compat/linux/linux_vdso.h>
73 
74 #include <arm64/linux/linux_sigframe.h>
75 
76 #include <machine/md_var.h>
77 #include <machine/pcb.h>
78 #ifdef VFP
79 #include <machine/vfp.h>
80 #endif
81 
82 MODULE_VERSION(linux64elf, 1);
83 
84 #define	LINUX_VDSOPAGE_SIZE	PAGE_SIZE * 2
85 #define	LINUX_VDSOPAGE		(VM_MAXUSER_ADDRESS - \
86 				    LINUX_VDSOPAGE_SIZE)
87 #define	LINUX_SHAREDPAGE	(LINUX_VDSOPAGE - PAGE_SIZE)
88 				/*
89 				 * PAGE_SIZE - the size
90 				 * of the native SHAREDPAGE
91 				 */
92 #define	LINUX_USRSTACK		LINUX_SHAREDPAGE
93 #define	LINUX_PS_STRINGS	(LINUX_USRSTACK - \
94 				    sizeof(struct ps_strings))
95 
96 static int linux_szsigcode;
97 static vm_object_t linux_vdso_obj;
98 static char *linux_vdso_mapping;
99 extern char _binary_linux_vdso_so_o_start;
100 extern char _binary_linux_vdso_so_o_end;
101 static vm_offset_t linux_vdso_base;
102 
103 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
104 extern const char *linux_syscallnames[];
105 
106 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
107 
108 static int	linux_copyout_strings(struct image_params *imgp,
109 		    uintptr_t *stack_base);
110 static int	linux_elf_fixup(uintptr_t *stack_base,
111 		    struct image_params *iparams);
112 static bool	linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
113 static void	linux_vdso_install(const void *param);
114 static void	linux_vdso_deinstall(const void *param);
115 static void	linux_vdso_reloc(char *mapping, Elf_Addr offset);
116 static void	linux_set_syscall_retval(struct thread *td, int error);
117 static int	linux_fetch_syscall_args(struct thread *td);
118 static void	linux_exec_setregs(struct thread *td, struct image_params *imgp,
119 		    uintptr_t stack);
120 static void	linux_exec_sysvec_init(void *param);
121 static int	linux_on_exec_vmspace(struct proc *p,
122 		    struct image_params *imgp);
123 
124 /* DTrace init */
125 LIN_SDT_PROVIDER_DECLARE(LINUX_DTRACE);
126 
127 /* DTrace probes */
128 LIN_SDT_PROBE_DEFINE0(sysvec, linux_exec_setregs, todo);
129 LIN_SDT_PROBE_DEFINE0(sysvec, linux_copyout_auxargs, todo);
130 LIN_SDT_PROBE_DEFINE0(sysvec, linux_elf_fixup, todo);
131 
132 LINUX_VDSO_SYM_CHAR(linux_platform);
133 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base);
134 LINUX_VDSO_SYM_INTPTR(linux_vdso_sigcode);
135 
136 static int
137 linux_fetch_syscall_args(struct thread *td)
138 {
139 	struct proc *p;
140 	struct syscall_args *sa;
141 	register_t *ap;
142 
143 	p = td->td_proc;
144 	ap = td->td_frame->tf_x;
145 	sa = &td->td_sa;
146 
147 	sa->code = td->td_frame->tf_x[8];
148 	sa->original_code = sa->code;
149 	/* LINUXTODO: generic syscall? */
150 	if (sa->code >= p->p_sysent->sv_size)
151 		sa->callp = &p->p_sysent->sv_table[0];
152 	else
153 		sa->callp = &p->p_sysent->sv_table[sa->code];
154 
155 	if (sa->callp->sy_narg > nitems(sa->args))
156 		panic("ARM64TODO: Could we have more than %zu args?",
157 		    nitems(sa->args));
158 	memcpy(sa->args, ap, nitems(sa->args) * sizeof(register_t));
159 
160 	td->td_retval[0] = 0;
161 	return (0);
162 }
163 
164 static void
165 linux_set_syscall_retval(struct thread *td, int error)
166 {
167 
168 	td->td_retval[1] = td->td_frame->tf_x[1];
169 	cpu_set_syscall_retval(td, error);
170 
171 	if (__predict_false(error != 0)) {
172 		if (error != ERESTART && error != EJUSTRETURN)
173 			td->td_frame->tf_x[0] = bsd_to_linux_errno(error);
174 	}
175 }
176 
177 static int
178 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
179 {
180 	Elf_Auxargs *args;
181 	Elf_Auxinfo *argarray, *pos;
182 	struct proc *p;
183 	int error, issetugid;
184 
185 	LIN_SDT_PROBE0(sysvec, linux_copyout_auxargs, todo);
186 	p = imgp->proc;
187 
188 	args = (Elf64_Auxargs *)imgp->auxargs;
189 	argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
190 	    M_WAITOK | M_ZERO);
191 
192 	issetugid = p->p_flag & P_SUGID ? 1 : 0;
193 	AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, linux_vdso_base);
194 	AUXARGS_ENTRY(pos, LINUX_AT_MINSIGSTKSZ, LINUX_MINSIGSTKSZ);
195 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, *imgp->sysent->sv_hwcap);
196 	AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
197 	AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
198 	AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
199 	AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
200 	AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
201 	AUXARGS_ENTRY(pos, AT_BASE, args->base);
202 	AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
203 	AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
204 	AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
205 	AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
206 	AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
207 	AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
208 	AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
209 	AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary);
210 	AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, *imgp->sysent->sv_hwcap2);
211 	if (imgp->execpathp != 0)
212 		AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp);
213 	if (args->execfd != -1)
214 		AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
215 	AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
216 	AUXARGS_ENTRY(pos, AT_NULL, 0);
217 
218 	free(imgp->auxargs, M_TEMP);
219 	imgp->auxargs = NULL;
220 	KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
221 
222 	error = copyout(argarray, (void *)base,
223 	    sizeof(*argarray) * LINUX_AT_COUNT);
224 	free(argarray, M_TEMP);
225 	return (error);
226 }
227 
228 static int
229 linux_elf_fixup(uintptr_t *stack_base, struct image_params *imgp)
230 {
231 
232 	LIN_SDT_PROBE0(sysvec, linux_elf_fixup, todo);
233 
234 	return (0);
235 }
236 
237 /*
238  * Copy strings out to the new process address space, constructing new arg
239  * and env vector tables. Return a pointer to the base so that it can be used
240  * as the initial stack pointer.
241  * LINUXTODO: deduplicate against other linuxulator archs
242  */
243 static int
244 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
245 {
246 	char **vectp;
247 	char *stringp;
248 	uintptr_t destp, ustringp;
249 	struct ps_strings *arginfo;
250 	char canary[LINUX_AT_RANDOM_LEN];
251 	size_t execpath_len;
252 	struct proc *p;
253 	int argc, envc, error;
254 
255 	p = imgp->proc;
256 	arginfo = (struct ps_strings *)PROC_PS_STRINGS(p);
257 	destp = (uintptr_t)arginfo;
258 
259 	if (imgp->execpath != NULL && imgp->auxargs != NULL) {
260 		execpath_len = strlen(imgp->execpath) + 1;
261 		destp -= execpath_len;
262 		destp = rounddown2(destp, sizeof(void *));
263 		imgp->execpathp = (void *)destp;
264 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
265 		if (error != 0)
266 			return (error);
267 	}
268 
269 	/* Prepare the canary for SSP. */
270 	arc4rand(canary, sizeof(canary), 0);
271 	destp -= roundup(sizeof(canary), sizeof(void *));
272 	imgp->canary = (void *)destp;
273 	error = copyout(canary, imgp->canary, sizeof(canary));
274 	if (error != 0)
275 		return (error);
276 
277 	/* Allocate room for the argument and environment strings. */
278 	destp -= ARG_MAX - imgp->args->stringspace;
279 	destp = rounddown2(destp, sizeof(void *));
280 	ustringp = destp;
281 
282 	if (imgp->auxargs) {
283 		/*
284 		 * Allocate room on the stack for the ELF auxargs
285 		 * array.  It has up to LINUX_AT_COUNT entries.
286 		 */
287 		destp -= LINUX_AT_COUNT * sizeof(Elf64_Auxinfo);
288 		destp = rounddown2(destp, sizeof(void *));
289 	}
290 
291 	vectp = (char **)destp;
292 
293 	/*
294 	 * Allocate room for argc and the argv[] and env vectors including the
295 	 * terminating NULL pointers.
296 	 */
297 	vectp -= 1 + imgp->args->argc + 1 + imgp->args->envc + 1;
298 	vectp = (char **)STACKALIGN(vectp);
299 
300 	/* vectp also becomes our initial stack base. */
301 	*stack_base = (uintptr_t)vectp;
302 
303 	stringp = imgp->args->begin_argv;
304 	argc = imgp->args->argc;
305 	envc = imgp->args->envc;
306 
307 	/* Copy out strings - arguments and environment. */
308 	error = copyout(stringp, (void *)ustringp,
309 	    ARG_MAX - imgp->args->stringspace);
310 	if (error != 0)
311 		return (error);
312 
313 	/* Fill in "ps_strings" struct for ps, w, etc. */
314 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
315 	    suword(&arginfo->ps_nargvstr, argc) != 0)
316 		return (EFAULT);
317 
318 	if (suword(vectp++, argc) != 0)
319 		return (EFAULT);
320 
321 	/* Fill in argument portion of vector table. */
322 	for (; argc > 0; --argc) {
323 		if (suword(vectp++, ustringp) != 0)
324 			return (EFAULT);
325 		while (*stringp++ != 0)
326 			ustringp++;
327 		ustringp++;
328 	}
329 
330 	/* A null vector table pointer separates the argp's from the envp's. */
331 	if (suword(vectp++, 0) != 0)
332 		return (EFAULT);
333 
334 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
335 	    suword(&arginfo->ps_nenvstr, envc) != 0)
336 		return (EFAULT);
337 
338 	/* Fill in environment portion of vector table. */
339 	for (; envc > 0; --envc) {
340 		if (suword(vectp++, ustringp) != 0)
341 			return (EFAULT);
342 		while (*stringp++ != 0)
343 			ustringp++;
344 		ustringp++;
345 	}
346 
347 	/* The end of the vector table is a null pointer. */
348 	if (suword(vectp, 0) != 0)
349 		return (EFAULT);
350 
351 	if (imgp->auxargs) {
352 		vectp++;
353 		error = imgp->sysent->sv_copyout_auxargs(imgp,
354 		    (uintptr_t)vectp);
355 		if (error != 0)
356 			return (error);
357 	}
358 
359 	return (0);
360 }
361 
362 /*
363  * Reset registers to default values on exec.
364  */
365 static void
366 linux_exec_setregs(struct thread *td, struct image_params *imgp,
367     uintptr_t stack)
368 {
369 	struct trapframe *regs = td->td_frame;
370 	struct pcb *pcb = td->td_pcb;
371 
372 	/* LINUXTODO: validate */
373 	LIN_SDT_PROBE0(sysvec, linux_exec_setregs, todo);
374 
375 	memset(regs, 0, sizeof(*regs));
376 	/* glibc start.S registers function pointer in x0 with atexit. */
377         regs->tf_sp = stack;
378 #if 0	/* LINUXTODO: See if this is used. */
379 	regs->tf_lr = imgp->entry_addr;
380 #else
381         regs->tf_lr = 0xffffffffffffffff;
382 #endif
383         regs->tf_elr = imgp->entry_addr;
384 
385 	pcb->pcb_tpidr_el0 = 0;
386 	pcb->pcb_tpidrro_el0 = 0;
387 	WRITE_SPECIALREG(tpidrro_el0, 0);
388 	WRITE_SPECIALREG(tpidr_el0, 0);
389 
390 #ifdef VFP
391 	vfp_reset_state(td, pcb);
392 #endif
393 
394 	/*
395 	 * Clear debug register state. It is not applicable to the new process.
396 	 */
397 	bzero(&pcb->pcb_dbg_regs, sizeof(pcb->pcb_dbg_regs));
398 }
399 
400 int
401 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
402 {
403 	struct l_sigframe *frame;
404 	ucontext_t uc;
405 	struct trapframe *tf;
406 	int error;
407 
408 	tf = td->td_frame;
409 	frame = (struct l_sigframe *)tf->tf_sp;
410 
411 	if (copyin((void *)&frame->uc, &uc, sizeof(uc)))
412 		return (EFAULT);
413 
414 	error = set_mcontext(td, &uc.uc_mcontext);
415 	if (error != 0)
416 		return (error);
417 
418 	/* Restore signal mask. */
419 	kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0);
420 
421 	return (EJUSTRETURN);
422 }
423 
424 static void
425 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
426 {
427 	struct thread *td;
428 	struct proc *p;
429 	struct trapframe *tf;
430 	struct l_sigframe *fp, *frame;
431 	struct l_fpsimd_context *fpsimd;
432 	struct l_esr_context *esr;
433 	l_stack_t uc_stack;
434 	ucontext_t uc;
435 	uint8_t *scr;
436 	struct sigacts *psp;
437 	int onstack, sig, issiginfo;
438 
439 	td = curthread;
440 	p = td->td_proc;
441 	PROC_LOCK_ASSERT(p, MA_OWNED);
442 
443 	sig = ksi->ksi_signo;
444 	psp = p->p_sigacts;
445 	mtx_assert(&psp->ps_mtx, MA_OWNED);
446 
447 	tf = td->td_frame;
448 	onstack = sigonstack(tf->tf_sp);
449 	issiginfo = SIGISMEMBER(psp->ps_siginfo, sig);
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 	get_mcontext(td, &uc.uc_mcontext, 0);
471 	uc.uc_sigmask = *mask;
472 
473 	uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
474 	uc_stack.ss_size = td->td_sigstk.ss_size;
475 	uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) != 0 ?
476 	    (onstack ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
477 	mtx_unlock(&psp->ps_mtx);
478 	PROC_UNLOCK(td->td_proc);
479 
480 	/* Fill in the frame to copy out */
481 	frame = malloc(sizeof(*frame), M_LINUX, M_WAITOK | M_ZERO);
482 
483 	memcpy(&frame->sf.sf_uc.uc_sc.regs, tf->tf_x, sizeof(tf->tf_x));
484 	frame->sf.sf_uc.uc_sc.regs[30] = tf->tf_lr;
485 	frame->sf.sf_uc.uc_sc.sp = tf->tf_sp;
486 	frame->sf.sf_uc.uc_sc.pc = tf->tf_lr;
487 	frame->sf.sf_uc.uc_sc.pstate = tf->tf_spsr;
488 	frame->sf.sf_uc.uc_sc.fault_address = (register_t)ksi->ksi_addr;
489 
490 	/* Stack frame for unwinding */
491 	frame->fp = tf->tf_x[29];
492 	frame->lr = tf->tf_lr;
493 
494 	/* Translate the signal. */
495 	sig = bsd_to_linux_signal(sig);
496 	siginfo_to_lsiginfo(&ksi->ksi_info, &frame->sf.sf_si, sig);
497 	bsd_to_linux_sigset(mask, &frame->sf.sf_uc.uc_sigmask);
498 
499 	/*
500 	 * Prepare fpsimd & esr. Does not check sizes, as
501 	 * __reserved is big enougth.
502 	 */
503 	scr = (uint8_t *)&frame->sf.sf_uc.uc_sc.__reserved;
504 #ifdef VFP
505 	fpsimd = (struct l_fpsimd_context *) scr;
506 	fpsimd->head.magic = L_FPSIMD_MAGIC;
507 	fpsimd->head.size = sizeof(struct l_fpsimd_context);
508 	fpsimd->fpsr = uc.uc_mcontext.mc_fpregs.fp_sr;
509 	fpsimd->fpcr = uc.uc_mcontext.mc_fpregs.fp_cr;
510 
511 	memcpy(fpsimd->vregs, &uc.uc_mcontext.mc_fpregs.fp_q,
512 	    sizeof(uc.uc_mcontext.mc_fpregs.fp_q));
513 	scr += roundup(sizeof(struct l_fpsimd_context), 16);
514 #endif
515 	if (ksi->ksi_addr != 0) {
516 		esr = (struct l_esr_context *) scr;
517 		esr->head.magic = L_ESR_MAGIC;
518 		esr->head.size = sizeof(struct l_esr_context);
519 		esr->esr = tf->tf_esr;
520 	}
521 
522 	memcpy(&frame->sf.sf_uc.uc_stack, &uc_stack, sizeof(uc_stack));
523 	memcpy(&frame->uc, &uc, sizeof(uc));
524 
525 	/* Copy the sigframe out to the user's stack. */
526 	if (copyout(frame, fp, sizeof(*fp)) != 0) {
527 		/* Process has trashed its stack. Kill it. */
528 		free(frame, M_LINUX);
529 		CTR2(KTR_SIG, "sendsig: sigexit td=%p fp=%p", td, fp);
530 		PROC_LOCK(p);
531 		sigexit(td, SIGILL);
532 	}
533 	free(frame, M_LINUX);
534 
535 	tf->tf_x[0]= sig;
536 	if (issiginfo) {
537 		tf->tf_x[1] = (register_t)&fp->sf.sf_si;
538 		tf->tf_x[2] = (register_t)&fp->sf.sf_uc;
539 	} else {
540 		tf->tf_x[1] = 0;
541 		tf->tf_x[2] = 0;
542 	}
543 	tf->tf_x[8] = (register_t)catcher;
544 	tf->tf_sp = (register_t)fp;
545 	tf->tf_elr = (register_t)linux_vdso_sigcode;
546 
547 	CTR3(KTR_SIG, "sendsig: return td=%p pc=%#x sp=%#x", td, tf->tf_elr,
548 	    tf->tf_sp);
549 
550 	PROC_LOCK(p);
551 	mtx_lock(&psp->ps_mtx);
552 }
553 
554 struct sysentvec elf_linux_sysvec = {
555 	.sv_size	= LINUX_SYS_MAXSYSCALL,
556 	.sv_table	= linux_sysent,
557 	.sv_fixup	= linux_elf_fixup,
558 	.sv_sendsig	= linux_rt_sendsig,
559 	.sv_sigcode	= &_binary_linux_vdso_so_o_start,
560 	.sv_szsigcode	= &linux_szsigcode,
561 	.sv_name	= "Linux ELF64",
562 	.sv_coredump	= elf64_coredump,
563 	.sv_elf_core_osabi = ELFOSABI_NONE,
564 	.sv_elf_core_abi_vendor = LINUX_ABI_VENDOR,
565 	.sv_elf_core_prepare_notes = linux64_prepare_notes,
566 	.sv_imgact_try	= linux_exec_imgact_try,
567 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
568 	.sv_minuser	= VM_MIN_ADDRESS,
569 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
570 	.sv_usrstack	= LINUX_USRSTACK,
571 	.sv_psstrings	= LINUX_PS_STRINGS,
572 	.sv_psstringssz	= sizeof(struct ps_strings),
573 	.sv_stackprot	= VM_PROT_READ | VM_PROT_WRITE,
574 	.sv_copyout_auxargs = linux_copyout_auxargs,
575 	.sv_copyout_strings = linux_copyout_strings,
576 	.sv_setregs	= linux_exec_setregs,
577 	.sv_fixlimit	= NULL,
578 	.sv_maxssiz	= NULL,
579 	.sv_flags	= SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN |
580 	    SV_SIG_WAITNDQ | SV_TIMEKEEP,
581 	.sv_set_syscall_retval = linux_set_syscall_retval,
582 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
583 	.sv_syscallnames = linux_syscallnames,
584 	.sv_shared_page_base = LINUX_SHAREDPAGE,
585 	.sv_shared_page_len = PAGE_SIZE,
586 	.sv_schedtail	= linux_schedtail,
587 	.sv_thread_detach = linux_thread_detach,
588 	.sv_trap	= NULL,
589 	.sv_hwcap	= &elf_hwcap,
590 	.sv_hwcap2	= &elf_hwcap2,
591 	.sv_onexec	= linux_on_exec_vmspace,
592 	.sv_onexit	= linux_on_exit,
593 	.sv_ontdexit	= linux_thread_dtor,
594 	.sv_setid_allowed = &linux_setid_allowed_query,
595 };
596 
597 static int
598 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp)
599 {
600 	int error;
601 
602 	error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base,
603 	    LINUX_VDSOPAGE_SIZE, imgp);
604 	if (error == 0)
605 		linux_on_exec(p, imgp);
606 	return (error);
607 }
608 
609 /*
610  * linux_vdso_install() and linux_exec_sysvec_init() must be called
611  * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY).
612  */
613 static void
614 linux_exec_sysvec_init(void *param)
615 {
616 	l_uintptr_t *ktimekeep_base;
617 	struct sysentvec *sv;
618 	ptrdiff_t tkoff;
619 
620 	sv = param;
621 	/* Fill timekeep_base */
622 	exec_sysvec_init(sv);
623 
624 	tkoff = kern_timekeep_base - linux_vdso_base;
625 	ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
626 	*ktimekeep_base = sv->sv_shared_page_base + sv->sv_timekeep_offset;
627 }
628 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY,
629     linux_exec_sysvec_init, &elf_linux_sysvec);
630 
631 static void
632 linux_vdso_install(const void *param)
633 {
634 	char *vdso_start = &_binary_linux_vdso_so_o_start;
635 	char *vdso_end = &_binary_linux_vdso_so_o_end;
636 
637 	linux_szsigcode = vdso_end - vdso_start;
638 	MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE);
639 
640 	linux_vdso_base = LINUX_VDSOPAGE;
641 
642 	__elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base);
643 
644 	linux_vdso_obj = __elfN(linux_shared_page_init)
645 	    (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
646 	bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode);
647 
648 	linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base);
649 }
650 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST,
651     linux_vdso_install, NULL);
652 
653 static void
654 linux_vdso_deinstall(const void *param)
655 {
656 
657 	__elfN(linux_shared_page_fini)(linux_vdso_obj,
658 	    linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
659 }
660 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
661     linux_vdso_deinstall, NULL);
662 
663 static void
664 linux_vdso_reloc(char *mapping, Elf_Addr offset)
665 {
666 	Elf_Size rtype, symidx;
667 	const Elf_Rela *rela;
668 	const Elf_Shdr *shdr;
669 	const Elf_Ehdr *ehdr;
670 	Elf_Addr *where;
671 	Elf_Addr addr, addend;
672 	int i, relacnt;
673 
674 	MPASS(offset != 0);
675 
676 	relacnt = 0;
677 	ehdr = (const Elf_Ehdr *)mapping;
678 	shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff);
679 	for (i = 0; i < ehdr->e_shnum; i++)
680 	{
681 		switch (shdr[i].sh_type) {
682 		case SHT_REL:
683 			printf("Linux Aarch64 vDSO: unexpected Rel section\n");
684 			break;
685 		case SHT_RELA:
686 			rela = (const Elf_Rela *)(mapping + shdr[i].sh_offset);
687 			relacnt = shdr[i].sh_size / sizeof(*rela);
688 		}
689 	}
690 
691 	for (i = 0; i < relacnt; i++, rela++) {
692 		where = (Elf_Addr *)(mapping + rela->r_offset);
693 		addend = rela->r_addend;
694 		rtype = ELF_R_TYPE(rela->r_info);
695 		symidx = ELF_R_SYM(rela->r_info);
696 
697 		switch (rtype) {
698 		case R_AARCH64_NONE:	/* none */
699 			break;
700 
701 		case R_AARCH64_RELATIVE:	/* B + A */
702 			addr = (Elf_Addr)(mapping + addend);
703 			if (*where != addr)
704 				*where = addr;
705 			break;
706 		default:
707 			printf("Linux Aarch64 vDSO: unexpected relocation type %ld, "
708 			    "symbol index %ld\n", rtype, symidx);
709 		}
710 	}
711 }
712 
713 static char GNU_ABI_VENDOR[] = "GNU";
714 static int GNU_ABI_LINUX = 0;
715 
716 /* LINUXTODO: deduplicate */
717 static bool
718 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
719 {
720 	const Elf32_Word *desc;
721 	uintptr_t p;
722 
723 	p = (uintptr_t)(note + 1);
724 	p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
725 
726 	desc = (const Elf32_Word *)p;
727 	if (desc[0] != GNU_ABI_LINUX)
728 		return (false);
729 
730 	*osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
731 	return (true);
732 }
733 
734 static Elf_Brandnote linux64_brandnote = {
735 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
736 	.hdr.n_descsz	= 16,
737 	.hdr.n_type	= 1,
738 	.vendor		= GNU_ABI_VENDOR,
739 	.flags		= BN_TRANSLATE_OSREL,
740 	.trans_osrel	= linux_trans_osrel
741 };
742 
743 static Elf64_Brandinfo linux_glibc2brand = {
744 	.brand		= ELFOSABI_LINUX,
745 	.machine	= EM_AARCH64,
746 	.compat_3_brand	= "Linux",
747 	.emul_path	= linux_emul_path,
748 	.interp_path	= "/lib64/ld-linux-x86-64.so.2",
749 	.sysvec		= &elf_linux_sysvec,
750 	.interp_newpath	= NULL,
751 	.brand_note	= &linux64_brandnote,
752 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
753 };
754 
755 Elf64_Brandinfo *linux_brandlist[] = {
756 	&linux_glibc2brand,
757 	NULL
758 };
759 
760 static int
761 linux64_elf_modevent(module_t mod, int type, void *data)
762 {
763 	Elf64_Brandinfo **brandinfo;
764 	struct linux_ioctl_handler**lihp;
765 	int error;
766 
767 	error = 0;
768 	switch(type) {
769 	case MOD_LOAD:
770 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
771 		    ++brandinfo)
772 			if (elf64_insert_brand_entry(*brandinfo) < 0)
773 				error = EINVAL;
774 		if (error == 0) {
775 			SET_FOREACH(lihp, linux_ioctl_handler_set)
776 				linux_ioctl_register_handler(*lihp);
777 			stclohz = (stathz ? stathz : hz);
778 			if (bootverbose)
779 				printf("Linux arm64 ELF exec handler installed\n");
780 		}
781 		break;
782 	case MOD_UNLOAD:
783 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
784 		    ++brandinfo)
785 			if (elf64_brand_inuse(*brandinfo))
786 				error = EBUSY;
787 		if (error == 0) {
788 			for (brandinfo = &linux_brandlist[0];
789 			    *brandinfo != NULL; ++brandinfo)
790 				if (elf64_remove_brand_entry(*brandinfo) < 0)
791 					error = EINVAL;
792 		}
793 		if (error == 0) {
794 			SET_FOREACH(lihp, linux_ioctl_handler_set)
795 				linux_ioctl_unregister_handler(*lihp);
796 			if (bootverbose)
797 				printf("Linux arm64 ELF exec handler removed\n");
798 		} else
799 			printf("Could not deinstall Linux arm64 ELF interpreter entry\n");
800 		break;
801 	default:
802 		return (EOPNOTSUPP);
803 	}
804 	return (error);
805 }
806 
807 static moduledata_t linux64_elf_mod = {
808 	"linux64elf",
809 	linux64_elf_modevent,
810 	0
811 };
812 
813 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
814 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
815 FEATURE(linux64, "AArch64 Linux 64bit support");
816