xref: /freebsd-13-stable/sys/arm64/linux/linux_sysvec.c (revision 72b242ac15af15bea2c6989bb911d2dd6ab11651)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 #define	__ELF_WORD_SIZE	64
31 
32 #include <sys/param.h>
33 #include <sys/elf.h>
34 #include <sys/exec.h>
35 #include <sys/imgact.h>
36 #include <sys/imgact_elf.h>
37 #include <sys/kernel.h>
38 #include <sys/ktr.h>
39 #include <sys/lock.h>
40 #include <sys/module.h>
41 #include <sys/mutex.h>
42 #include <sys/proc.h>
43 #include <sys/stddef.h>
44 #include <sys/syscallsubr.h>
45 #include <sys/sysctl.h>
46 #include <sys/sysent.h>
47 
48 #include <vm/pmap.h>
49 #include <vm/vm.h>
50 #include <vm/vm_map.h>
51 #include <vm/vm_page.h>
52 
53 #include <arm64/linux/linux.h>
54 #include <arm64/linux/linux_proto.h>
55 #include <compat/linux/linux_dtrace.h>
56 #include <compat/linux/linux_elf.h>
57 #include <compat/linux/linux_emul.h>
58 #include <compat/linux/linux_fork.h>
59 #include <compat/linux/linux_ioctl.h>
60 #include <compat/linux/linux_mib.h>
61 #include <compat/linux/linux_misc.h>
62 #include <compat/linux/linux_signal.h>
63 #include <compat/linux/linux_util.h>
64 #include <compat/linux/linux_vdso.h>
65 
66 #include <arm64/linux/linux_sigframe.h>
67 
68 #include <machine/md_var.h>
69 
70 #ifdef VFP
71 #include <machine/vfp.h>
72 #endif
73 
74 MODULE_VERSION(linux64elf, 1);
75 
76 #define	LINUX_VDSOPAGE_SIZE	PAGE_SIZE * 2
77 #define	LINUX_VDSOPAGE		(VM_MAXUSER_ADDRESS - \
78 				    LINUX_VDSOPAGE_SIZE)
79 #define	LINUX_SHAREDPAGE	(LINUX_VDSOPAGE - PAGE_SIZE)
80 				/*
81 				 * PAGE_SIZE - the size
82 				 * of the native SHAREDPAGE
83 				 */
84 #define	LINUX_USRSTACK		LINUX_SHAREDPAGE
85 #define	LINUX_PS_STRINGS	(LINUX_USRSTACK - \
86 				    sizeof(struct ps_strings))
87 
88 static int linux_szsigcode;
89 static vm_object_t linux_vdso_obj;
90 static char *linux_vdso_mapping;
91 extern char _binary_linux_vdso_so_o_start;
92 extern char _binary_linux_vdso_so_o_end;
93 static vm_offset_t linux_vdso_base;
94 
95 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
96 extern const char *linux_syscallnames[];
97 
98 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
99 
100 static void	linux_vdso_install(const void *param);
101 static void	linux_vdso_deinstall(const void *param);
102 static void	linux_vdso_reloc(char *mapping, Elf_Addr offset);
103 static void	linux_set_syscall_retval(struct thread *td, int error);
104 static int	linux_fetch_syscall_args(struct thread *td);
105 static void	linux_exec_setregs(struct thread *td, struct image_params *imgp,
106 		    uintptr_t stack);
107 static void	linux_exec_sysvec_init(void *param);
108 static int	linux_on_exec_vmspace(struct proc *p,
109 		    struct image_params *imgp);
110 
111 /* DTrace init */
112 LIN_SDT_PROVIDER_DECLARE(LINUX_DTRACE);
113 
114 /* DTrace probes */
115 LIN_SDT_PROBE_DEFINE0(sysvec, linux_exec_setregs, todo);
116 
117 LINUX_VDSO_SYM_CHAR(linux_platform);
118 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base);
119 LINUX_VDSO_SYM_INTPTR(linux_vdso_sigcode);
120 
121 static int
linux_fetch_syscall_args(struct thread * td)122 linux_fetch_syscall_args(struct thread *td)
123 {
124 	struct proc *p;
125 	struct syscall_args *sa;
126 	register_t *ap;
127 
128 	p = td->td_proc;
129 	ap = td->td_frame->tf_x;
130 	sa = &td->td_sa;
131 
132 	sa->code = td->td_frame->tf_x[8];
133 	/* LINUXTODO: generic syscall? */
134 	if (sa->code >= p->p_sysent->sv_size)
135 		sa->callp = &nosys_sysent;
136 	else
137 		sa->callp = &p->p_sysent->sv_table[sa->code];
138 
139 	if (sa->callp->sy_narg > MAXARGS)
140 		panic("ARM64TODO: Could we have more than %d args?", MAXARGS);
141 	memcpy(sa->args, ap, MAXARGS * sizeof(register_t));
142 
143 	td->td_retval[0] = 0;
144 	return (0);
145 }
146 
147 static void
linux_set_syscall_retval(struct thread * td,int error)148 linux_set_syscall_retval(struct thread *td, int error)
149 {
150 
151 	td->td_retval[1] = td->td_frame->tf_x[1];
152 	cpu_set_syscall_retval(td, error);
153 
154 	if (__predict_false(error != 0)) {
155 		if (error != ERESTART && error != EJUSTRETURN)
156 			td->td_frame->tf_x[0] = bsd_to_linux_errno(error);
157 	}
158 }
159 
160 void
linux64_arch_copyout_auxargs(struct image_params * imgp,Elf_Auxinfo ** pos)161 linux64_arch_copyout_auxargs(struct image_params *imgp, Elf_Auxinfo **pos)
162 {
163 
164 	AUXARGS_ENTRY((*pos), LINUX_AT_SYSINFO_EHDR, linux_vdso_base);
165 	AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP, *imgp->sysent->sv_hwcap);
166 	AUXARGS_ENTRY((*pos), LINUX_AT_HWCAP2, *imgp->sysent->sv_hwcap2);
167 	AUXARGS_ENTRY((*pos), LINUX_AT_PLATFORM, PTROUT(linux_platform));
168 }
169 
170 /*
171  * Reset registers to default values on exec.
172  */
173 static void
linux_exec_setregs(struct thread * td,struct image_params * imgp,uintptr_t stack)174 linux_exec_setregs(struct thread *td, struct image_params *imgp,
175     uintptr_t stack)
176 {
177 	struct trapframe *regs = td->td_frame;
178 	struct pcb *pcb = td->td_pcb;
179 
180 	/* LINUXTODO: validate */
181 	LIN_SDT_PROBE0(sysvec, linux_exec_setregs, todo);
182 
183 	memset(regs, 0, sizeof(*regs));
184 	/* glibc start.S registers function pointer in x0 with atexit. */
185         regs->tf_sp = stack;
186 #if 0	/* LINUXTODO: See if this is used. */
187 	regs->tf_lr = imgp->entry_addr;
188 #else
189         regs->tf_lr = 0xffffffffffffffff;
190 #endif
191         regs->tf_elr = imgp->entry_addr;
192 
193 	pcb->pcb_tpidr_el0 = 0;
194 	pcb->pcb_tpidrro_el0 = 0;
195 	WRITE_SPECIALREG(tpidrro_el0, 0);
196 	WRITE_SPECIALREG(tpidr_el0, 0);
197 
198 #ifdef VFP
199 	vfp_reset_state(td, pcb);
200 #endif
201 
202 	/*
203 	 * Clear debug register state. It is not applicable to the new process.
204 	 */
205 	bzero(&pcb->pcb_dbg_regs, sizeof(pcb->pcb_dbg_regs));
206 }
207 
208 static bool
linux_parse_sigreturn_ctx(struct thread * td,struct l_sigcontext * sc)209 linux_parse_sigreturn_ctx(struct thread *td, struct l_sigcontext *sc)
210 {
211 	struct l_fpsimd_context *fpsimd;
212 	struct _l_aarch64_ctx *ctx;
213 	int offset;
214 
215 	offset = 0;
216 	while (1) {
217 		/* The offset must be 16 byte aligned */
218 		if ((offset & 15) != 0)
219 			return (false);
220 
221 		/* Check for buffer overflow of the ctx */
222 		if ((offset + sizeof(*ctx)) >
223 		    sizeof(sc->__reserved))
224 			return (false);
225 
226 		ctx = (struct _l_aarch64_ctx *)&sc->__reserved[offset];
227 
228 		/* Check for buffer overflow of the data */
229 		if ((offset + ctx->size) > sizeof(sc->__reserved))
230 			return (false);
231 
232 		switch(ctx->magic) {
233 		case 0:
234 			if (ctx->size != 0)
235 				return (false);
236 			return (true);
237 		case L_ESR_MAGIC:
238 			/* Ignore */
239 			break;
240 #ifdef VFP
241 		case L_FPSIMD_MAGIC:
242 			fpsimd = (struct l_fpsimd_context *)ctx;
243 
244 			/*
245 			 * Discard any vfp state for the current thread, we
246 			 * are about to override it.
247 			 */
248 			critical_enter();
249 			vfp_discard(td);
250 			critical_exit();
251 
252 			td->td_pcb->pcb_fpustate.vfp_fpcr = fpsimd->fpcr;
253 			td->td_pcb->pcb_fpustate.vfp_fpsr = fpsimd->fpsr;
254 			memcpy(td->td_pcb->pcb_fpustate.vfp_regs,
255 			    fpsimd->vregs, sizeof(fpsimd->vregs));
256 
257 			break;
258 #endif
259 		default:
260 			return (false);
261 		}
262 
263 		offset += ctx->size;
264 	}
265 
266 }
267 
268 int
linux_rt_sigreturn(struct thread * td,struct linux_rt_sigreturn_args * args)269 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
270 {
271 	struct l_rt_sigframe *sf;
272 	struct l_sigframe *frame;
273 	struct trapframe *tf;
274 	sigset_t bmask;
275 	int error;
276 
277 	sf = malloc(sizeof(*sf), M_LINUX, M_WAITOK | M_ZERO);
278 
279 	tf = td->td_frame;
280 	frame = (struct l_sigframe *)tf->tf_sp;
281 	error = copyin((void *)&frame->sf, sf, sizeof(*sf));
282 	if (error != 0) {
283 		free(sf, M_LINUX);
284 		return (error);
285 	}
286 
287 	memcpy(tf->tf_x, sf->sf_uc.uc_sc.regs, sizeof(tf->tf_x));
288 	tf->tf_lr = sf->sf_uc.uc_sc.regs[30];
289 	tf->tf_sp = sf->sf_uc.uc_sc.sp;
290 	tf->tf_elr = sf->sf_uc.uc_sc.pc;
291 
292 	if ((sf->sf_uc.uc_sc.pstate & PSR_M_MASK) != PSR_M_EL0t ||
293 	    (sf->sf_uc.uc_sc.pstate & PSR_AARCH32) != 0 ||
294 	    (sf->sf_uc.uc_sc.pstate & PSR_DAIF) !=
295 	    (td->td_frame->tf_spsr & PSR_DAIF))
296 		goto einval;
297 	tf->tf_spsr = sf->sf_uc.uc_sc.pstate;
298 
299 	if (!linux_parse_sigreturn_ctx(td, &sf->sf_uc.uc_sc))
300 		goto einval;
301 
302 	/* Restore signal mask. */
303 	linux_to_bsd_sigset(&sf->sf_uc.uc_sigmask, &bmask);
304 	kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
305 	free(sf, M_LINUX);
306 
307 	return (EJUSTRETURN);
308 einval:
309 	free(sf, M_LINUX);
310 	return (EINVAL);
311 }
312 
313 static void
linux_rt_sendsig(sig_t catcher,ksiginfo_t * ksi,sigset_t * mask)314 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
315 {
316 	struct thread *td;
317 	struct proc *p;
318 	struct trapframe *tf;
319 	struct l_sigframe *fp, *frame;
320 	struct l_fpsimd_context *fpsimd;
321 	struct l_esr_context *esr;
322 	l_stack_t uc_stack;
323 	ucontext_t uc;
324 	uint8_t *scr;
325 	struct sigacts *psp;
326 	int onstack, sig, issiginfo;
327 
328 	td = curthread;
329 	p = td->td_proc;
330 	PROC_LOCK_ASSERT(p, MA_OWNED);
331 
332 	sig = ksi->ksi_signo;
333 	psp = p->p_sigacts;
334 	mtx_assert(&psp->ps_mtx, MA_OWNED);
335 
336 	tf = td->td_frame;
337 	onstack = sigonstack(tf->tf_sp);
338 	issiginfo = SIGISMEMBER(psp->ps_siginfo, sig);
339 
340 	CTR4(KTR_SIG, "sendsig: td=%p (%s) catcher=%p sig=%d", td, p->p_comm,
341 	    catcher, sig);
342 
343 	/* Allocate and validate space for the signal handler context. */
344 	if ((td->td_pflags & TDP_ALTSTACK) != 0 && !onstack &&
345 	    SIGISMEMBER(psp->ps_sigonstack, sig)) {
346 		fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp +
347 		    td->td_sigstk.ss_size);
348 #if defined(COMPAT_43)
349 		td->td_sigstk.ss_flags |= SS_ONSTACK;
350 #endif
351 	} else {
352 		fp = (struct l_sigframe *)td->td_frame->tf_sp;
353 	}
354 
355 	/* Make room, keeping the stack aligned */
356 	fp--;
357 	fp = (struct l_sigframe *)STACKALIGN(fp);
358 
359 	get_mcontext(td, &uc.uc_mcontext, 0);
360 	uc.uc_sigmask = *mask;
361 
362 	uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
363 	uc_stack.ss_size = td->td_sigstk.ss_size;
364 	uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) != 0 ?
365 	    (onstack ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
366 	mtx_unlock(&psp->ps_mtx);
367 	PROC_UNLOCK(td->td_proc);
368 
369 	/* Fill in the frame to copy out */
370 	frame = malloc(sizeof(*frame), M_LINUX, M_WAITOK | M_ZERO);
371 
372 	memcpy(&frame->sf.sf_uc.uc_sc.regs, tf->tf_x, sizeof(tf->tf_x));
373 	frame->sf.sf_uc.uc_sc.regs[30] = tf->tf_lr;
374 	frame->sf.sf_uc.uc_sc.sp = tf->tf_sp;
375 	frame->sf.sf_uc.uc_sc.pc = tf->tf_lr;
376 	frame->sf.sf_uc.uc_sc.pstate = tf->tf_spsr;
377 	frame->sf.sf_uc.uc_sc.fault_address = (register_t)ksi->ksi_addr;
378 
379 	/* Stack frame for unwinding */
380 	frame->fp = tf->tf_x[29];
381 	frame->lr = tf->tf_lr;
382 
383 	/* Translate the signal. */
384 	sig = bsd_to_linux_signal(sig);
385 	siginfo_to_lsiginfo(&ksi->ksi_info, &frame->sf.sf_si, sig);
386 	bsd_to_linux_sigset(mask, &frame->sf.sf_uc.uc_sigmask);
387 
388 	/*
389 	 * Prepare fpsimd & esr. Does not check sizes, as
390 	 * __reserved is big enougth.
391 	 */
392 	scr = (uint8_t *)&frame->sf.sf_uc.uc_sc.__reserved;
393 #ifdef VFP
394 	fpsimd = (struct l_fpsimd_context *) scr;
395 	fpsimd->head.magic = L_FPSIMD_MAGIC;
396 	fpsimd->head.size = sizeof(struct l_fpsimd_context);
397 	fpsimd->fpsr = uc.uc_mcontext.mc_fpregs.fp_sr;
398 	fpsimd->fpcr = uc.uc_mcontext.mc_fpregs.fp_cr;
399 
400 	memcpy(fpsimd->vregs, &uc.uc_mcontext.mc_fpregs.fp_q,
401 	    sizeof(uc.uc_mcontext.mc_fpregs.fp_q));
402 	scr += roundup(sizeof(struct l_fpsimd_context), 16);
403 #endif
404 	if (ksi->ksi_addr != 0) {
405 		esr = (struct l_esr_context *) scr;
406 		esr->head.magic = L_ESR_MAGIC;
407 		esr->head.size = sizeof(struct l_esr_context);
408 		esr->esr = tf->tf_esr;
409 	}
410 
411 	memcpy(&frame->sf.sf_uc.uc_stack, &uc_stack, sizeof(uc_stack));
412 
413 	/* Copy the sigframe out to the user's stack. */
414 	if (copyout(frame, fp, sizeof(*fp)) != 0) {
415 		/* Process has trashed its stack. Kill it. */
416 		free(frame, M_LINUX);
417 		CTR2(KTR_SIG, "sendsig: sigexit td=%p fp=%p", td, fp);
418 		PROC_LOCK(p);
419 		sigexit(td, SIGILL);
420 	}
421 	free(frame, M_LINUX);
422 
423 	tf->tf_x[0]= sig;
424 	if (issiginfo) {
425 		tf->tf_x[1] = (register_t)&fp->sf.sf_si;
426 		tf->tf_x[2] = (register_t)&fp->sf.sf_uc;
427 	} else {
428 		tf->tf_x[1] = 0;
429 		tf->tf_x[2] = 0;
430 	}
431 	tf->tf_x[8] = (register_t)catcher;
432 	tf->tf_sp = (register_t)fp;
433 	tf->tf_elr = (register_t)linux_vdso_sigcode;
434 
435 	CTR3(KTR_SIG, "sendsig: return td=%p pc=%#x sp=%#x", td, tf->tf_elr,
436 	    tf->tf_sp);
437 
438 	PROC_LOCK(p);
439 	mtx_lock(&psp->ps_mtx);
440 }
441 
442 struct sysentvec elf_linux_sysvec = {
443 	.sv_size	= LINUX_SYS_MAXSYSCALL,
444 	.sv_table	= linux_sysent,
445 	.sv_fixup	= __elfN(freebsd_fixup),
446 	.sv_sendsig	= linux_rt_sendsig,
447 	.sv_sigcode	= &_binary_linux_vdso_so_o_start,
448 	.sv_szsigcode	= &linux_szsigcode,
449 	.sv_name	= "Linux ELF64",
450 	.sv_coredump	= elf64_coredump,
451 	.sv_elf_core_osabi = ELFOSABI_NONE,
452 	.sv_elf_core_abi_vendor = LINUX_ABI_VENDOR,
453 	.sv_elf_core_prepare_notes = linux64_prepare_notes,
454 	.sv_imgact_try	= linux_exec_imgact_try,
455 	.sv_minsigstksz	= LINUX_MINSIGSTKSZ,
456 	.sv_minuser	= VM_MIN_ADDRESS,
457 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
458 	.sv_usrstack	= LINUX_USRSTACK,
459 	.sv_psstrings	= LINUX_PS_STRINGS,
460 	.sv_psstringssz	= sizeof(struct ps_strings),
461 	.sv_stackprot	= VM_PROT_READ | VM_PROT_WRITE,
462 	.sv_copyout_auxargs = __linuxN(copyout_auxargs),
463 	.sv_copyout_strings = __linuxN(copyout_strings),
464 	.sv_setregs	= linux_exec_setregs,
465 	.sv_fixlimit	= NULL,
466 	.sv_maxssiz	= NULL,
467 	.sv_flags	= SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN |
468 	    SV_SIG_WAITNDQ | SV_TIMEKEEP,
469 	.sv_set_syscall_retval = linux_set_syscall_retval,
470 	.sv_fetch_syscall_args = linux_fetch_syscall_args,
471 	.sv_syscallnames = linux_syscallnames,
472 	.sv_shared_page_base = LINUX_SHAREDPAGE,
473 	.sv_shared_page_len = PAGE_SIZE,
474 	.sv_schedtail	= linux_schedtail,
475 	.sv_thread_detach = linux_thread_detach,
476 	.sv_trap	= NULL,
477 	.sv_hwcap	= &elf_hwcap,
478 	.sv_hwcap2	= &elf_hwcap2,
479 	.sv_onexec	= linux_on_exec_vmspace,
480 	.sv_onexit	= linux_on_exit,
481 	.sv_ontdexit	= linux_thread_dtor,
482 	.sv_setid_allowed = &linux_setid_allowed_query,
483 };
484 
485 static int
linux_on_exec_vmspace(struct proc * p,struct image_params * imgp)486 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp)
487 {
488 	int error;
489 
490 	error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base,
491 	    LINUX_VDSOPAGE_SIZE, imgp);
492 	if (error == 0)
493 		linux_on_exec(p, imgp);
494 	return (error);
495 }
496 
497 /*
498  * linux_vdso_install() and linux_exec_sysvec_init() must be called
499  * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY).
500  */
501 static void
linux_exec_sysvec_init(void * param)502 linux_exec_sysvec_init(void *param)
503 {
504 	l_uintptr_t *ktimekeep_base;
505 	struct sysentvec *sv;
506 	ptrdiff_t tkoff;
507 
508 	sv = param;
509 	/* Fill timekeep_base */
510 	exec_sysvec_init(sv);
511 
512 	tkoff = kern_timekeep_base - linux_vdso_base;
513 	ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
514 	*ktimekeep_base = sv->sv_timekeep_base;
515 }
516 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY,
517     linux_exec_sysvec_init, &elf_linux_sysvec);
518 
519 static void
linux_vdso_install(const void * param)520 linux_vdso_install(const void *param)
521 {
522 	char *vdso_start = &_binary_linux_vdso_so_o_start;
523 	char *vdso_end = &_binary_linux_vdso_so_o_end;
524 
525 	linux_szsigcode = vdso_end - vdso_start;
526 	MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE);
527 
528 	linux_vdso_base = LINUX_VDSOPAGE;
529 
530 	__elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base);
531 
532 	linux_vdso_obj = __elfN(linux_shared_page_init)
533 	    (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
534 	bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode);
535 
536 	linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base);
537 }
538 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST,
539     linux_vdso_install, NULL);
540 
541 static void
linux_vdso_deinstall(const void * param)542 linux_vdso_deinstall(const void *param)
543 {
544 
545 	__elfN(linux_shared_page_fini)(linux_vdso_obj,
546 	    linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
547 }
548 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
549     linux_vdso_deinstall, NULL);
550 
551 static void
linux_vdso_reloc(char * mapping,Elf_Addr offset)552 linux_vdso_reloc(char *mapping, Elf_Addr offset)
553 {
554 	Elf_Size rtype, symidx;
555 	const Elf_Rela *rela;
556 	const Elf_Shdr *shdr;
557 	const Elf_Ehdr *ehdr;
558 	Elf_Addr *where;
559 	Elf_Addr addr, addend;
560 	int i, relacnt;
561 
562 	MPASS(offset != 0);
563 
564 	relacnt = 0;
565 	ehdr = (const Elf_Ehdr *)mapping;
566 	shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff);
567 	for (i = 0; i < ehdr->e_shnum; i++)
568 	{
569 		switch (shdr[i].sh_type) {
570 		case SHT_REL:
571 			printf("Linux Aarch64 vDSO: unexpected Rel section\n");
572 			break;
573 		case SHT_RELA:
574 			rela = (const Elf_Rela *)(mapping + shdr[i].sh_offset);
575 			relacnt = shdr[i].sh_size / sizeof(*rela);
576 		}
577 	}
578 
579 	for (i = 0; i < relacnt; i++, rela++) {
580 		where = (Elf_Addr *)(mapping + rela->r_offset);
581 		addend = rela->r_addend;
582 		rtype = ELF_R_TYPE(rela->r_info);
583 		symidx = ELF_R_SYM(rela->r_info);
584 
585 		switch (rtype) {
586 		case R_AARCH64_NONE:	/* none */
587 			break;
588 
589 		case R_AARCH64_RELATIVE:	/* B + A */
590 			addr = (Elf_Addr)(mapping + addend);
591 			if (*where != addr)
592 				*where = addr;
593 			break;
594 		default:
595 			printf("Linux Aarch64 vDSO: unexpected relocation type %ld, "
596 			    "symbol index %ld\n", rtype, symidx);
597 		}
598 	}
599 }
600 
601 static Elf_Brandnote linux64_brandnote = {
602 	.hdr.n_namesz	= sizeof(GNU_ABI_VENDOR),
603 	.hdr.n_descsz	= 16,
604 	.hdr.n_type	= 1,
605 	.vendor		= GNU_ABI_VENDOR,
606 	.flags		= BN_TRANSLATE_OSREL,
607 	.trans_osrel	= linux_trans_osrel
608 };
609 
610 static Elf64_Brandinfo linux_glibc2brand = {
611 	.brand		= ELFOSABI_LINUX,
612 	.machine	= EM_AARCH64,
613 	.compat_3_brand	= "Linux",
614 	.emul_path	= linux_emul_path,
615 	.interp_path	= "/lib64/ld-linux-x86-64.so.2",
616 	.sysvec		= &elf_linux_sysvec,
617 	.interp_newpath	= NULL,
618 	.brand_note	= &linux64_brandnote,
619 	.flags		= BI_CAN_EXEC_DYN | BI_BRAND_NOTE
620 };
621 
622 Elf64_Brandinfo *linux_brandlist[] = {
623 	&linux_glibc2brand,
624 	NULL
625 };
626 
627 static int
linux64_elf_modevent(module_t mod,int type,void * data)628 linux64_elf_modevent(module_t mod, int type, void *data)
629 {
630 	Elf64_Brandinfo **brandinfo;
631 	struct linux_ioctl_handler**lihp;
632 	int error;
633 
634 	error = 0;
635 	switch(type) {
636 	case MOD_LOAD:
637 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
638 		    ++brandinfo)
639 			if (elf64_insert_brand_entry(*brandinfo) < 0)
640 				error = EINVAL;
641 		if (error == 0) {
642 			SET_FOREACH(lihp, linux_ioctl_handler_set)
643 				linux_ioctl_register_handler(*lihp);
644 			stclohz = (stathz ? stathz : hz);
645 			if (bootverbose)
646 				printf("Linux arm64 ELF exec handler installed\n");
647 		}
648 		break;
649 	case MOD_UNLOAD:
650 		for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
651 		    ++brandinfo)
652 			if (elf64_brand_inuse(*brandinfo))
653 				error = EBUSY;
654 		if (error == 0) {
655 			for (brandinfo = &linux_brandlist[0];
656 			    *brandinfo != NULL; ++brandinfo)
657 				if (elf64_remove_brand_entry(*brandinfo) < 0)
658 					error = EINVAL;
659 		}
660 		if (error == 0) {
661 			SET_FOREACH(lihp, linux_ioctl_handler_set)
662 				linux_ioctl_unregister_handler(*lihp);
663 			if (bootverbose)
664 				printf("Linux arm64 ELF exec handler removed\n");
665 		} else
666 			printf("Could not deinstall Linux arm64 ELF interpreter entry\n");
667 		break;
668 	default:
669 		return (EOPNOTSUPP);
670 	}
671 	return (error);
672 }
673 
674 static moduledata_t linux64_elf_mod = {
675 	"linux64elf",
676 	linux64_elf_modevent,
677 	0
678 };
679 
680 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
681 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
682 FEATURE(linux64, "AArch64 Linux 64bit support");
683