1/*-
2 * Copyright (c) 2014 The FreeBSD Foundation
3 * All rights reserved.
4 *
5 * This software was developed by Andrew Turner under
6 * sponsorship from the FreeBSD Foundation.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 *    notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 *    notice, this list of conditions and the following disclaimer in the
15 *    documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 */
29
30#include <machine/asm.h>
31ENTRY(.rtld_start)
32	.cfi_undefined	x30
33	mov	x19, x0		/* Put ps_strings in a callee-saved register */
34
35	sub	sp, sp, #16	/* Make room for obj_main & exit proc */
36	.cfi_adjust_cfa_offset	16
37
38	mov	x1, sp		/* exit_proc */
39	add	x2, x1, #8	/* obj_main */
40	bl	_rtld		/* Call the loader */
41	mov	x8, x0		/* Backup the entry point */
42	ldp	x2, x1, [sp], #16 /* Load cleanup, obj_main */
43	.cfi_adjust_cfa_offset	0
44
45	mov	x0, x19		/* Restore ps_strings */
46	br	x8		/* Jump to the entry point */
47END(.rtld_start)
48
49/*
50 * sp + 0 = &GOT[x + 3]
51 * sp + 8 = RA
52 * x16 = &GOT[2]
53 * x17 = &_rtld_bind_start
54 */
55ENTRY(_rtld_bind_start)
56	mov	x17, sp
57
58	/* Save frame pointer and SP */
59	stp	x29, x30, [sp, #-16]!
60	mov	x29, sp
61	.cfi_def_cfa x29, 16
62	.cfi_offset x30, -8
63	.cfi_offset x29, -16
64
65	/* Save the arguments */
66	stp	x0, x1, [sp, #-16]!
67	stp	x2, x3, [sp, #-16]!
68	stp	x4, x5, [sp, #-16]!
69	stp	x6, x7, [sp, #-16]!
70	stp	x8, xzr, [sp, #-16]!
71
72	/* Save any floating-point arguments */
73	stp	q0, q1, [sp, #-32]!
74	stp	q2, q3, [sp, #-32]!
75	stp	q4, q5, [sp, #-32]!
76	stp	q6, q7, [sp, #-32]!
77
78	/* Calculate reloff */
79	ldr	x2, [x17, #0]	/* Get the address of the entry */
80	sub	x1, x2, x16	/* Find its offset */
81	sub	x1, x1, #8	/* Adjust for x16 not being at offset 0 */
82	/* Each rela item has 3 entriesso we need reloff = 3 * index */
83	lsl	x3, x1, #1	/* x3 = 2 * offset */
84	add	x1, x1, x3	/* x1 = x3 + offset = 3 * offset */
85
86	/* Load obj */
87	ldr	x0, [x16, #-8]
88
89	/* Call into rtld */
90	bl	_rtld_bind
91
92	/* Backup the address to branch to */
93	mov	x16, x0
94
95	/* restore the arguments */
96	ldp	q6, q7, [sp], #32
97	ldp	q4, q5, [sp], #32
98	ldp	q2, q3, [sp], #32
99	ldp	q0, q1, [sp], #32
100	ldp	x8, xzr, [sp], #16
101	ldp	x6, x7, [sp], #16
102	ldp	x4, x5, [sp], #16
103	ldp	x2, x3, [sp], #16
104	ldp	x0, x1, [sp], #16
105
106	/* Restore frame pointer */
107	ldp	x29, xzr, [sp], #16
108
109	 /* Restore link register saved by the plt code */
110	ldp	xzr, x30, [sp], #16
111
112	/* Call into the correct function */
113	br	x16
114END(_rtld_bind_start)
115
116/*
117 * struct rel_tlsdesc {
118 *  uint64_t resolver_fnc;
119 *  uint64_t resolver_arg;
120 *
121 *
122 * uint64_t _rtld_tlsdesc_static(struct rel_tlsdesc *);
123 *
124 * Resolver function for TLS symbols resolved at load time
125 */
126ENTRY(_rtld_tlsdesc_static)
127	ldr	x0, [x0, #8]
128	ret
129END(_rtld_tlsdesc_static)
130
131/*
132 * uint64_t _rtld_tlsdesc_undef(void);
133 *
134 * Resolver function for weak and undefined TLS symbols
135 */
136ENTRY(_rtld_tlsdesc_undef)
137	str	x1, [sp, #-16]!
138	.cfi_adjust_cfa_offset	16
139
140	mrs	x1, tpidr_el0
141	ldr	x0, [x0, #8]
142	sub	x0, x0, x1
143
144	ldr	x1, [sp], #16
145	.cfi_adjust_cfa_offset 	-16
146	ret
147END(_rtld_tlsdesc_undef)
148
149/*
150 * uint64_t _rtld_tlsdesc_dynamic(struct rel_tlsdesc *);
151 *
152 * Resolver function for TLS symbols from dlopen()
153 */
154ENTRY(_rtld_tlsdesc_dynamic)
155	/* Save registers used in fast path */
156	stp	x1,  x2, [sp, #(-2 * 16)]!
157	stp	x3,  x4, [sp, #(1 * 16)]
158	.cfi_adjust_cfa_offset	2 * 16
159	.cfi_rel_offset		x1, 0
160	.cfi_rel_offset		x2, 8
161	.cfi_rel_offset		x3, 16
162	.cfi_rel_offset		x4, 24
163
164	/* Test fastpath - inlined version of tls_get_addr_common(). */
165	ldr	x1, [x0, #8]		/* tlsdesc ptr */
166	mrs	x4, tpidr_el0
167	ldr	x0, [x4]		/* DTV pointer */
168	ldr	x2, [x0]		/* dtv[0] (generation count) */
169	ldr	x3, [x1]		/* tlsdec->dtv_gen */
170	cmp	x2, x3
171	b.ne	1f			/* dtv[0] != tlsdec->dtv_gen */
172
173	ldr	w2, [x1, #8]		/* tlsdec->tls_index */
174	add	w2, w2, #1
175	ldr     x3, [x0, w2, sxtw #3]	/* dtv[tlsdesc->tls_index + 1] */
176	cbz	x3, 1f
177
178	/* Return (dtv[tlsdesc->tls_index + 1] + tlsdesc->tls_offs - tp) */
179	ldr	x2, [x1, #16]		/* tlsdec->tls_offs */
180	add 	x2, x2, x3
181	sub	x0, x2, x4
182	/* Restore registers and return */
183	ldp	 x3,  x4, [sp, #(1 * 16)]
184	ldp	 x1,  x2, [sp], #(2 * 16)
185	.cfi_adjust_cfa_offset 	-2 * 16
186	ret
187
188	/*
189	 * Slow path
190	  * return(
191	 *    tls_get_addr_common(tp, tlsdesc->tls_index, tlsdesc->tls_offs));
192	 *
193	 */
1941:
195	/* Save all integer registers */
196	stp	x29, x30, [sp, #-(8 * 16)]!
197	.cfi_adjust_cfa_offset	8 * 16
198	.cfi_rel_offset		x29, 0
199	.cfi_rel_offset		x30, 8
200
201	mov	x29, sp
202	stp	x5,   x6, [sp, #(1 * 16)]
203	stp	x7,   x8, [sp, #(2 * 16)]
204	stp	x9,  x10, [sp, #(3 * 16)]
205	stp	x11, x12, [sp, #(4 * 16)]
206	stp	x13, x14, [sp, #(5 * 16)]
207	stp	x15, x16, [sp, #(6 * 16)]
208	stp	x17, x18, [sp, #(7 * 16)]
209	.cfi_rel_offset		 x5, 16
210	.cfi_rel_offset		 x6, 24
211	.cfi_rel_offset		 x7, 32
212	.cfi_rel_offset		 x8, 40
213	.cfi_rel_offset		 x9, 48
214	.cfi_rel_offset		x10, 56
215	.cfi_rel_offset		x11, 64
216	.cfi_rel_offset		x12, 72
217	.cfi_rel_offset		x13, 80
218	.cfi_rel_offset		x14, 88
219	.cfi_rel_offset		x15, 96
220	.cfi_rel_offset		x16, 104
221	.cfi_rel_offset		x17, 112
222	.cfi_rel_offset		x18, 120
223
224	/* Find the tls offset */
225	mov	x0, x4			/* tp */
226	mov	x3, x1			/* tlsdesc ptr */
227	ldr	w1, [x3, #8]		/* tlsdec->tls_index */
228	ldr	x2, [x3, #16]		/* tlsdec->tls_offs */
229	bl	tls_get_addr_common
230	mrs	x1, tpidr_el0
231	sub	x0, x0, x1
232
233	/* Restore slow patch registers */
234	ldp	x17, x18, [sp, #(7 * 16)]
235	ldp	x15, x16, [sp, #(6 * 16)]
236	ldp	x13, x14, [sp, #(5 * 16)]
237	ldp	x11, x12, [sp, #(4 * 16)]
238	ldp	x9, x10,  [sp, #(3 * 16)]
239	ldp	x7, x8,   [sp, #(2 * 16)]
240	ldp	x5, x6,   [sp, #(1 * 16)]
241	ldp	x29, x30, [sp], #(8 * 16)
242	.cfi_adjust_cfa_offset 	-8 * 16
243	.cfi_restore		x29
244	.cfi_restore		x30
245
246	/* Restore fast path registers and return */
247	ldp	 x3,  x4, [sp, #16]
248	ldp	 x1,  x2, [sp], #(2 * 16)
249	.cfi_adjust_cfa_offset	-2 * 16
250	ret
251END(_rtld_tlsdesc_dynamic)
252