1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/compiler.h>
3 #include <linux/rbtree.h>
4 #include <inttypes.h>
5 #include <string.h>
6 #include <ctype.h>
7 #include <stdlib.h>
8 #include "dso.h"
9 #include "map.h"
10 #include "symbol.h"
11 #include <internal/lib.h> // page_size
12 #include "tests.h"
13 #include "debug.h"
14 #include "machine.h"
15 
16 #define UM(x) kallsyms_map->unmap_ip(kallsyms_map, (x))
17 
18 static bool is_ignored_symbol(const char *name, char type)
19 {
20 	/* Symbol names that exactly match to the following are ignored.*/
21 	static const char * const ignored_symbols[] = {
22 		/*
23 		 * Symbols which vary between passes. Passes 1 and 2 must have
24 		 * identical symbol lists. The kallsyms_* symbols below are
25 		 * only added after pass 1, they would be included in pass 2
26 		 * when --all-symbols is specified so exclude them to get a
27 		 * stable symbol list.
28 		 */
29 		"kallsyms_addresses",
30 		"kallsyms_offsets",
31 		"kallsyms_relative_base",
32 		"kallsyms_num_syms",
33 		"kallsyms_names",
34 		"kallsyms_markers",
35 		"kallsyms_token_table",
36 		"kallsyms_token_index",
37 		/* Exclude linker generated symbols which vary between passes */
38 		"_SDA_BASE_",		/* ppc */
39 		"_SDA2_BASE_",		/* ppc */
40 		NULL
41 	};
42 
43 	/* Symbol names that begin with the following are ignored.*/
44 	static const char * const ignored_prefixes[] = {
45 		"$",			/* local symbols for ARM, MIPS, etc. */
46 		".L",			/* local labels, .LBB,.Ltmpxxx,.L__unnamed_xx,.LASANPC, etc. */
47 		"__crc_",		/* modversions */
48 		"__efistub_",		/* arm64 EFI stub namespace */
49 		"__kvm_nvhe_$",		/* arm64 local symbols in non-VHE KVM namespace */
50 		"__kvm_nvhe_.L",	/* arm64 local symbols in non-VHE KVM namespace */
51 		"__AArch64ADRPThunk_",	/* arm64 lld */
52 		"__ARMV5PILongThunk_",	/* arm lld */
53 		"__ARMV7PILongThunk_",
54 		"__ThumbV7PILongThunk_",
55 		"__LA25Thunk_",		/* mips lld */
56 		"__microLA25Thunk_",
57 		NULL
58 	};
59 
60 	/* Symbol names that end with the following are ignored.*/
61 	static const char * const ignored_suffixes[] = {
62 		"_from_arm",		/* arm */
63 		"_from_thumb",		/* arm */
64 		"_veneer",		/* arm */
65 		NULL
66 	};
67 
68 	/* Symbol names that contain the following are ignored.*/
69 	static const char * const ignored_matches[] = {
70 		".long_branch.",	/* ppc stub */
71 		".plt_branch.",		/* ppc stub */
72 		NULL
73 	};
74 
75 	const char * const *p;
76 
77 	for (p = ignored_symbols; *p; p++)
78 		if (!strcmp(name, *p))
79 			return true;
80 
81 	for (p = ignored_prefixes; *p; p++)
82 		if (!strncmp(name, *p, strlen(*p)))
83 			return true;
84 
85 	for (p = ignored_suffixes; *p; p++) {
86 		int l = strlen(name) - strlen(*p);
87 
88 		if (l >= 0 && !strcmp(name + l, *p))
89 			return true;
90 	}
91 
92 	for (p = ignored_matches; *p; p++) {
93 		if (strstr(name, *p))
94 			return true;
95 	}
96 
97 	if (type == 'U' || type == 'u')
98 		return true;
99 	/* exclude debugging symbols */
100 	if (type == 'N' || type == 'n')
101 		return true;
102 
103 	if (toupper(type) == 'A') {
104 		/* Keep these useful absolute symbols */
105 		if (strcmp(name, "__kernel_syscall_via_break") &&
106 		    strcmp(name, "__kernel_syscall_via_epc") &&
107 		    strcmp(name, "__kernel_sigtramp") &&
108 		    strcmp(name, "__gp"))
109 			return true;
110 	}
111 
112 	return false;
113 }
114 
115 static int test__vmlinux_matches_kallsyms(struct test_suite *test __maybe_unused,
116 					int subtest __maybe_unused)
117 {
118 	int err = TEST_FAIL;
119 	struct rb_node *nd;
120 	struct symbol *sym;
121 	struct map *kallsyms_map, *vmlinux_map, *map;
122 	struct machine kallsyms, vmlinux;
123 	struct maps *maps;
124 	u64 mem_start, mem_end;
125 	bool header_printed;
126 
127 	/*
128 	 * Step 1:
129 	 *
130 	 * Init the machines that will hold kernel, modules obtained from
131 	 * both vmlinux + .ko files and from /proc/kallsyms split by modules.
132 	 */
133 	machine__init(&kallsyms, "", HOST_KERNEL_ID);
134 	machine__init(&vmlinux, "", HOST_KERNEL_ID);
135 
136 	maps = machine__kernel_maps(&vmlinux);
137 
138 	/*
139 	 * Step 2:
140 	 *
141 	 * Create the kernel maps for kallsyms and the DSO where we will then
142 	 * load /proc/kallsyms. Also create the modules maps from /proc/modules
143 	 * and find the .ko files that match them in /lib/modules/`uname -r`/.
144 	 */
145 	if (machine__create_kernel_maps(&kallsyms) < 0) {
146 		pr_debug("machine__create_kernel_maps failed");
147 		err = TEST_SKIP;
148 		goto out;
149 	}
150 
151 	/*
152 	 * Step 3:
153 	 *
154 	 * Load and split /proc/kallsyms into multiple maps, one per module.
155 	 * Do not use kcore, as this test was designed before kcore support
156 	 * and has parts that only make sense if using the non-kcore code.
157 	 * XXX: extend it to stress the kcorre code as well, hint: the list
158 	 * of modules extracted from /proc/kcore, in its current form, can't
159 	 * be compacted against the list of modules found in the "vmlinux"
160 	 * code and with the one got from /proc/modules from the "kallsyms" code.
161 	 */
162 	if (machine__load_kallsyms(&kallsyms, "/proc/kallsyms") <= 0) {
163 		pr_debug("machine__load_kallsyms failed");
164 		err = TEST_SKIP;
165 		goto out;
166 	}
167 
168 	/*
169 	 * Step 4:
170 	 *
171 	 * kallsyms will be internally on demand sorted by name so that we can
172 	 * find the reference relocation * symbol, i.e. the symbol we will use
173 	 * to see if the running kernel was relocated by checking if it has the
174 	 * same value in the vmlinux file we load.
175 	 */
176 	kallsyms_map = machine__kernel_map(&kallsyms);
177 
178 	/*
179 	 * Step 5:
180 	 *
181 	 * Now repeat step 2, this time for the vmlinux file we'll auto-locate.
182 	 */
183 	if (machine__create_kernel_maps(&vmlinux) < 0) {
184 		pr_info("machine__create_kernel_maps failed");
185 		goto out;
186 	}
187 
188 	vmlinux_map = machine__kernel_map(&vmlinux);
189 
190 	/*
191 	 * Step 6:
192 	 *
193 	 * Locate a vmlinux file in the vmlinux path that has a buildid that
194 	 * matches the one of the running kernel.
195 	 *
196 	 * While doing that look if we find the ref reloc symbol, if we find it
197 	 * we'll have its ref_reloc_symbol.unrelocated_addr and then
198 	 * maps__reloc_vmlinux will notice and set proper ->[un]map_ip routines
199 	 * to fixup the symbols.
200 	 */
201 	if (machine__load_vmlinux_path(&vmlinux) <= 0) {
202 		pr_info("Couldn't find a vmlinux that matches the kernel running on this machine, skipping test\n");
203 		err = TEST_SKIP;
204 		goto out;
205 	}
206 
207 	err = 0;
208 	/*
209 	 * Step 7:
210 	 *
211 	 * Now look at the symbols in the vmlinux DSO and check if we find all of them
212 	 * in the kallsyms dso. For the ones that are in both, check its names and
213 	 * end addresses too.
214 	 */
215 	map__for_each_symbol(vmlinux_map, sym, nd) {
216 		struct symbol *pair, *first_pair;
217 
218 		sym  = rb_entry(nd, struct symbol, rb_node);
219 
220 		if (sym->start == sym->end)
221 			continue;
222 
223 		mem_start = vmlinux_map->unmap_ip(vmlinux_map, sym->start);
224 		mem_end = vmlinux_map->unmap_ip(vmlinux_map, sym->end);
225 
226 		first_pair = machine__find_kernel_symbol(&kallsyms, mem_start, NULL);
227 		pair = first_pair;
228 
229 		if (pair && UM(pair->start) == mem_start) {
230 next_pair:
231 			if (arch__compare_symbol_names(sym->name, pair->name) == 0) {
232 				/*
233 				 * kallsyms don't have the symbol end, so we
234 				 * set that by using the next symbol start - 1,
235 				 * in some cases we get this up to a page
236 				 * wrong, trace_kmalloc when I was developing
237 				 * this code was one such example, 2106 bytes
238 				 * off the real size. More than that and we
239 				 * _really_ have a problem.
240 				 */
241 				s64 skew = mem_end - UM(pair->end);
242 				if (llabs(skew) >= page_size)
243 					pr_debug("WARN: %#" PRIx64 ": diff end addr for %s v: %#" PRIx64 " k: %#" PRIx64 "\n",
244 						 mem_start, sym->name, mem_end,
245 						 UM(pair->end));
246 
247 				/*
248 				 * Do not count this as a failure, because we
249 				 * could really find a case where it's not
250 				 * possible to get proper function end from
251 				 * kallsyms.
252 				 */
253 				continue;
254 			} else {
255 				pair = machine__find_kernel_symbol_by_name(&kallsyms, sym->name, NULL);
256 				if (pair) {
257 					if (UM(pair->start) == mem_start)
258 						goto next_pair;
259 
260 					pr_debug("WARN: %#" PRIx64 ": diff name v: %s k: %s\n",
261 						 mem_start, sym->name, pair->name);
262 				} else {
263 					pr_debug("WARN: %#" PRIx64 ": diff name v: %s k: %s\n",
264 						 mem_start, sym->name, first_pair->name);
265 				}
266 
267 				continue;
268 			}
269 		} else if (mem_start == kallsyms.vmlinux_map->end) {
270 			/*
271 			 * Ignore aliases to _etext, i.e. to the end of the kernel text area,
272 			 * such as __indirect_thunk_end.
273 			 */
274 			continue;
275 		} else if (is_ignored_symbol(sym->name, sym->type)) {
276 			/*
277 			 * Ignore hidden symbols, see scripts/kallsyms.c for the details
278 			 */
279 			continue;
280 		} else {
281 			pr_debug("ERR : %#" PRIx64 ": %s not on kallsyms\n",
282 				 mem_start, sym->name);
283 		}
284 
285 		err = -1;
286 	}
287 
288 	if (verbose <= 0)
289 		goto out;
290 
291 	header_printed = false;
292 
293 	maps__for_each_entry(maps, map) {
294 		struct map *
295 		/*
296 		 * If it is the kernel, kallsyms is always "[kernel.kallsyms]", while
297 		 * the kernel will have the path for the vmlinux file being used,
298 		 * so use the short name, less descriptive but the same ("[kernel]" in
299 		 * both cases.
300 		 */
301 		pair = maps__find_by_name(kallsyms.kmaps, (map->dso->kernel ?
302 								map->dso->short_name :
303 								map->dso->name));
304 		if (pair) {
305 			pair->priv = 1;
306 		} else {
307 			if (!header_printed) {
308 				pr_info("WARN: Maps only in vmlinux:\n");
309 				header_printed = true;
310 			}
311 			map__fprintf(map, stderr);
312 		}
313 	}
314 
315 	header_printed = false;
316 
317 	maps__for_each_entry(maps, map) {
318 		struct map *pair;
319 
320 		mem_start = vmlinux_map->unmap_ip(vmlinux_map, map->start);
321 		mem_end = vmlinux_map->unmap_ip(vmlinux_map, map->end);
322 
323 		pair = maps__find(kallsyms.kmaps, mem_start);
324 		if (pair == NULL || pair->priv)
325 			continue;
326 
327 		if (pair->start == mem_start) {
328 			if (!header_printed) {
329 				pr_info("WARN: Maps in vmlinux with a different name in kallsyms:\n");
330 				header_printed = true;
331 			}
332 
333 			pr_info("WARN: %" PRIx64 "-%" PRIx64 " %" PRIx64 " %s in kallsyms as",
334 				map->start, map->end, map->pgoff, map->dso->name);
335 			if (mem_end != pair->end)
336 				pr_info(":\nWARN: *%" PRIx64 "-%" PRIx64 " %" PRIx64,
337 					pair->start, pair->end, pair->pgoff);
338 			pr_info(" %s\n", pair->dso->name);
339 			pair->priv = 1;
340 		}
341 	}
342 
343 	header_printed = false;
344 
345 	maps = machine__kernel_maps(&kallsyms);
346 
347 	maps__for_each_entry(maps, map) {
348 		if (!map->priv) {
349 			if (!header_printed) {
350 				pr_info("WARN: Maps only in kallsyms:\n");
351 				header_printed = true;
352 			}
353 			map__fprintf(map, stderr);
354 		}
355 	}
356 out:
357 	machine__exit(&kallsyms);
358 	machine__exit(&vmlinux);
359 	return err;
360 }
361 
362 DEFINE_SUITE("vmlinux symtab matches kallsyms", vmlinux_matches_kallsyms);
363