1/*
2 * Copyright 2014 Advanced Micro Devices, Inc.
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21 * SOFTWARE.
22 *
23 * Authors: Tom Stellard <thomas.stellard@amd.com>
24 *
25 */
26
27#include "radeon_elf_util.h"
28#include "r600_pipe_common.h"
29
30#include "util/u_memory.h"
31
32#include <gelf.h>
33#include <libelf.h>
34#include <stdio.h>
35
36static void parse_symbol_table(Elf_Data *symbol_table_data,
37				const GElf_Shdr *symbol_table_header,
38				struct radeon_shader_binary *binary)
39{
40	GElf_Sym symbol;
41	unsigned i = 0;
42	unsigned symbol_count =
43		symbol_table_header->sh_size / symbol_table_header->sh_entsize;
44
45	/* We are over allocating this list, because symbol_count gives the
46 	 * total number of symbols, and we will only be filling the list
47 	 * with offsets of global symbols.  The memory savings from
48 	 * allocating the correct size of this list will be small, and
49 	 * I don't think it is worth the cost of pre-computing the number
50 	 * of global symbols.
51 	 */
52	binary->global_symbol_offsets = CALLOC(symbol_count, sizeof(uint64_t));
53
54	while (gelf_getsym(symbol_table_data, i++, &symbol)) {
55		unsigned i;
56		if (GELF_ST_BIND(symbol.st_info) != STB_GLOBAL ||
57		    symbol.st_shndx == 0 /* Undefined symbol */) {
58			continue;
59		}
60
61		binary->global_symbol_offsets[binary->global_symbol_count] =
62					symbol.st_value;
63
64		/* Sort the list using bubble sort.  This list will usually
65		 * be small. */
66		for (i = binary->global_symbol_count; i > 0; --i) {
67			uint64_t lhs = binary->global_symbol_offsets[i - 1];
68			uint64_t rhs = binary->global_symbol_offsets[i];
69			if (lhs < rhs) {
70				break;
71			}
72			binary->global_symbol_offsets[i] = lhs;
73			binary->global_symbol_offsets[i - 1] = rhs;
74		}
75		++binary->global_symbol_count;
76	}
77}
78
79static void parse_relocs(Elf *elf, Elf_Data *relocs, Elf_Data *symbols,
80			unsigned symbol_sh_link,
81			struct radeon_shader_binary *binary)
82{
83	unsigned i;
84
85	if (!relocs || !symbols || !binary->reloc_count) {
86		return;
87	}
88	binary->relocs = CALLOC(binary->reloc_count,
89			sizeof(struct radeon_shader_reloc));
90	for (i = 0; i < binary->reloc_count; i++) {
91		GElf_Sym symbol;
92		GElf_Rel rel;
93		char *symbol_name;
94		struct radeon_shader_reloc *reloc = &binary->relocs[i];
95
96		gelf_getrel(relocs, i, &rel);
97		gelf_getsym(symbols, GELF_R_SYM(rel.r_info), &symbol);
98		symbol_name = elf_strptr(elf, symbol_sh_link, symbol.st_name);
99
100		reloc->offset = rel.r_offset;
101		strncpy(reloc->name, symbol_name, sizeof(reloc->name)-1);
102		reloc->name[sizeof(reloc->name)-1] = 0;
103	}
104}
105
106void radeon_elf_read(const char *elf_data, unsigned elf_size,
107		     struct radeon_shader_binary *binary)
108{
109	char *elf_buffer;
110	Elf *elf;
111	Elf_Scn *section = NULL;
112	Elf_Data *symbols = NULL, *relocs = NULL;
113	size_t section_str_index;
114	unsigned symbol_sh_link = 0;
115
116	/* One of the libelf implementations
117	 * (http://www.mr511.de/software/english.htm) requires calling
118	 * elf_version() before elf_memory().
119	 */
120	elf_version(EV_CURRENT);
121	elf_buffer = MALLOC(elf_size);
122	memcpy(elf_buffer, elf_data, elf_size);
123
124	elf = elf_memory(elf_buffer, elf_size);
125
126	elf_getshdrstrndx(elf, &section_str_index);
127
128	while ((section = elf_nextscn(elf, section))) {
129		const char *name;
130		Elf_Data *section_data = NULL;
131		GElf_Shdr section_header;
132		if (gelf_getshdr(section, &section_header) != &section_header) {
133			fprintf(stderr, "Failed to read ELF section header\n");
134			return;
135		}
136		name = elf_strptr(elf, section_str_index, section_header.sh_name);
137		if (!strcmp(name, ".text")) {
138			section_data = elf_getdata(section, section_data);
139			binary->code_size = section_data->d_size;
140			binary->code = MALLOC(binary->code_size * sizeof(unsigned char));
141			memcpy(binary->code, section_data->d_buf, binary->code_size);
142		} else if (!strcmp(name, ".AMDGPU.config")) {
143			section_data = elf_getdata(section, section_data);
144			binary->config_size = section_data->d_size;
145			binary->config = MALLOC(binary->config_size * sizeof(unsigned char));
146			memcpy(binary->config, section_data->d_buf, binary->config_size);
147		} else if (!strcmp(name, ".AMDGPU.disasm")) {
148			/* Always read disassembly if it's available. */
149			section_data = elf_getdata(section, section_data);
150			binary->disasm_string = strndup(section_data->d_buf,
151							section_data->d_size);
152		} else if (!strncmp(name, ".rodata", 7)) {
153			section_data = elf_getdata(section, section_data);
154			binary->rodata_size = section_data->d_size;
155			binary->rodata = MALLOC(binary->rodata_size * sizeof(unsigned char));
156			memcpy(binary->rodata, section_data->d_buf, binary->rodata_size);
157		} else if (!strncmp(name, ".symtab", 7)) {
158			symbols = elf_getdata(section, section_data);
159			symbol_sh_link = section_header.sh_link;
160			parse_symbol_table(symbols, &section_header, binary);
161		} else if (!strcmp(name, ".rel.text")) {
162			relocs = elf_getdata(section, section_data);
163			binary->reloc_count = section_header.sh_size /
164					section_header.sh_entsize;
165		}
166	}
167
168	parse_relocs(elf, relocs, symbols, symbol_sh_link, binary);
169
170	if (elf){
171		elf_end(elf);
172	}
173	FREE(elf_buffer);
174
175	/* Cache the config size per symbol */
176	if (binary->global_symbol_count) {
177		binary->config_size_per_symbol =
178			binary->config_size / binary->global_symbol_count;
179	} else {
180		binary->global_symbol_count = 1;
181		binary->config_size_per_symbol = binary->config_size;
182	}
183}
184
185const unsigned char *radeon_shader_binary_config_start(
186	const struct radeon_shader_binary *binary,
187	uint64_t symbol_offset)
188{
189	unsigned i;
190	for (i = 0; i < binary->global_symbol_count; ++i) {
191		if (binary->global_symbol_offsets[i] == symbol_offset) {
192			unsigned offset = i * binary->config_size_per_symbol;
193			return binary->config + offset;
194		}
195	}
196	return binary->config;
197}
198