xref: /linux/scripts/basic/fixdep.c (revision 903a7d37d9ea03cfed21040467d3d345d1e6fc76)
1 /*
2  * "Optimize" a list of dependencies as spit out by gcc -MD
3  * for the kernel build
4  * ===========================================================================
5  *
6  * Author       Kai Germaschewski
7  * Copyright    2002 by Kai Germaschewski  <kai.germaschewski@gmx.de>
8  *
9  * This software may be used and distributed according to the terms
10  * of the GNU General Public License, incorporated herein by reference.
11  *
12  *
13  * Introduction:
14  *
15  * gcc produces a very nice and correct list of dependencies which
16  * tells make when to remake a file.
17  *
18  * To use this list as-is however has the drawback that virtually
19  * every file in the kernel includes autoconf.h.
20  *
21  * If the user re-runs make *config, autoconf.h will be
22  * regenerated.  make notices that and will rebuild every file which
23  * includes autoconf.h, i.e. basically all files. This is extremely
24  * annoying if the user just changed CONFIG_HIS_DRIVER from n to m.
25  *
26  * So we play the same trick that "mkdep" played before. We replace
27  * the dependency on autoconf.h by a dependency on every config
28  * option which is mentioned in any of the listed prerequisites.
29  *
30  * kconfig populates a tree in include/config/ with an empty file
31  * for each config symbol and when the configuration is updated
32  * the files representing changed config options are touched
33  * which then let make pick up the changes and the files that use
34  * the config symbols are rebuilt.
35  *
36  * So if the user changes his CONFIG_HIS_DRIVER option, only the objects
37  * which depend on "include/config/HIS_DRIVER" will be rebuilt,
38  * so most likely only his driver ;-)
39  *
40  * The idea above dates, by the way, back to Michael E Chastain, AFAIK.
41  *
42  * So to get dependencies right, there are two issues:
43  * o if any of the files the compiler read changed, we need to rebuild
44  * o if the command line given to the compile the file changed, we
45  *   better rebuild as well.
46  *
47  * The former is handled by using the -MD output, the later by saving
48  * the command line used to compile the old object and comparing it
49  * to the one we would now use.
50  *
51  * Again, also this idea is pretty old and has been discussed on
52  * kbuild-devel a long time ago. I don't have a sensibly working
53  * internet connection right now, so I rather don't mention names
54  * without double checking.
55  *
56  * This code here has been based partially based on mkdep.c, which
57  * says the following about its history:
58  *
59  *   Copyright abandoned, Michael Chastain, <mailto:mec@shout.net>.
60  *   This is a C version of syncdep.pl by Werner Almesberger.
61  *
62  *
63  * It is invoked as
64  *
65  *   fixdep <depfile> <target> <cmdline>
66  *
67  * and will read the dependency file <depfile>
68  *
69  * The transformed dependency snipped is written to stdout.
70  *
71  * It first generates a line
72  *
73  *   savedcmd_<target> = <cmdline>
74  *
75  * and then basically copies the .<target>.d file to stdout, in the
76  * process filtering out the dependency on autoconf.h and adding
77  * dependencies on include/config/MY_OPTION for every
78  * CONFIG_MY_OPTION encountered in any of the prerequisites.
79  *
80  * We don't even try to really parse the header files, but
81  * merely grep, i.e. if CONFIG_FOO is mentioned in a comment, it will
82  * be picked up as well. It's not a problem with respect to
83  * correctness, since that can only give too many dependencies, thus
84  * we cannot miss a rebuild. Since people tend to not mention totally
85  * unrelated CONFIG_ options all over the place, it's not an
86  * efficiency problem either.
87  *
88  * (Note: it'd be easy to port over the complete mkdep state machine,
89  *  but I don't think the added complexity is worth it)
90  */
91 
92 #include <sys/types.h>
93 #include <sys/stat.h>
94 #include <unistd.h>
95 #include <fcntl.h>
96 #include <string.h>
97 #include <stdbool.h>
98 #include <stdlib.h>
99 #include <stdio.h>
100 #include <ctype.h>
101 
102 #include <xalloc.h>
103 
104 static void usage(void)
105 {
106 	fprintf(stderr, "Usage: fixdep <depfile> <target> <cmdline>\n");
107 	exit(1);
108 }
109 
110 struct item {
111 	struct item	*next;
112 	unsigned int	len;
113 	unsigned int	hash;
114 	char		name[];
115 };
116 
117 #define HASHSZ 256
118 static struct item *config_hashtab[HASHSZ], *file_hashtab[HASHSZ];
119 
120 static unsigned int strhash(const char *str, unsigned int sz)
121 {
122 	/* fnv32 hash */
123 	unsigned int i, hash = 2166136261U;
124 
125 	for (i = 0; i < sz; i++)
126 		hash = (hash ^ str[i]) * 0x01000193;
127 	return hash;
128 }
129 
130 /*
131  * Add a new value to the configuration string.
132  */
133 static void add_to_hashtable(const char *name, int len, unsigned int hash,
134 			     struct item *hashtab[])
135 {
136 	struct item *aux;
137 
138 	aux = xmalloc(sizeof(*aux) + len);
139 	memcpy(aux->name, name, len);
140 	aux->len = len;
141 	aux->hash = hash;
142 	aux->next = hashtab[hash % HASHSZ];
143 	hashtab[hash % HASHSZ] = aux;
144 }
145 
146 /*
147  * Lookup a string in the hash table. If found, just return true.
148  * If not, add it to the hashtable and return false.
149  */
150 static bool in_hashtable(const char *name, int len, struct item *hashtab[])
151 {
152 	struct item *aux;
153 	unsigned int hash = strhash(name, len);
154 
155 	for (aux = hashtab[hash % HASHSZ]; aux; aux = aux->next) {
156 		if (aux->hash == hash && aux->len == len &&
157 		    memcmp(aux->name, name, len) == 0)
158 			return true;
159 	}
160 
161 	add_to_hashtable(name, len, hash, hashtab);
162 
163 	return false;
164 }
165 
166 /*
167  * Record the use of a CONFIG_* word.
168  */
169 static void use_config(const char *m, int slen)
170 {
171 	if (in_hashtable(m, slen, config_hashtab))
172 		return;
173 
174 	/* Print out a dependency path from a symbol name. */
175 	printf("    $(wildcard include/config/%.*s) \\\n", slen, m);
176 }
177 
178 /* test if s ends in sub */
179 static int str_ends_with(const char *s, int slen, const char *sub)
180 {
181 	int sublen = strlen(sub);
182 
183 	if (sublen > slen)
184 		return 0;
185 
186 	return !memcmp(s + slen - sublen, sub, sublen);
187 }
188 
189 static void parse_config_file(const char *p)
190 {
191 	const char *q, *r;
192 	const char *start = p;
193 
194 	while ((p = strstr(p, "CONFIG_"))) {
195 		if (p > start && (isalnum(p[-1]) || p[-1] == '_')) {
196 			p += 7;
197 			continue;
198 		}
199 		p += 7;
200 		q = p;
201 		while (isalnum(*q) || *q == '_')
202 			q++;
203 		if (str_ends_with(p, q - p, "_MODULE"))
204 			r = q - 7;
205 		else
206 			r = q;
207 		if (r > p)
208 			use_config(p, r - p);
209 		p = q;
210 	}
211 }
212 
213 static void *read_file(const char *filename)
214 {
215 	struct stat st;
216 	int fd;
217 	char *buf;
218 
219 	fd = open(filename, O_RDONLY);
220 	if (fd < 0) {
221 		fprintf(stderr, "fixdep: error opening file: ");
222 		perror(filename);
223 		exit(2);
224 	}
225 	if (fstat(fd, &st) < 0) {
226 		fprintf(stderr, "fixdep: error fstat'ing file: ");
227 		perror(filename);
228 		exit(2);
229 	}
230 	buf = xmalloc(st.st_size + 1);
231 	if (read(fd, buf, st.st_size) != st.st_size) {
232 		perror("fixdep: read");
233 		exit(2);
234 	}
235 	buf[st.st_size] = '\0';
236 	close(fd);
237 
238 	return buf;
239 }
240 
241 /* Ignore certain dependencies */
242 static int is_ignored_file(const char *s, int len)
243 {
244 	return str_ends_with(s, len, "include/generated/autoconf.h");
245 }
246 
247 /* Do not parse these files */
248 static int is_no_parse_file(const char *s, int len)
249 {
250 	/* rustc may list binary files in dep-info */
251 	return str_ends_with(s, len, ".rlib") ||
252 	       str_ends_with(s, len, ".rmeta") ||
253 	       str_ends_with(s, len, ".so");
254 }
255 
256 /*
257  * Important: The below generated source_foo.o and deps_foo.o variable
258  * assignments are parsed not only by make, but also by the rather simple
259  * parser in scripts/mod/sumversion.c.
260  */
261 static void parse_dep_file(char *p, const char *target)
262 {
263 	bool saw_any_target = false;
264 	bool is_target = true;
265 	bool is_source = false;
266 	bool need_parse;
267 	char *q, saved_c;
268 
269 	while (*p) {
270 		/* handle some special characters first. */
271 		switch (*p) {
272 		case '#':
273 			/*
274 			 * skip comments.
275 			 * rustc may emit comments to dep-info.
276 			 */
277 			p++;
278 			while (*p != '\0' && *p != '\n') {
279 				/*
280 				 * escaped newlines continue the comment across
281 				 * multiple lines.
282 				 */
283 				if (*p == '\\')
284 					p++;
285 				p++;
286 			}
287 			continue;
288 		case ' ':
289 		case '\t':
290 			/* skip whitespaces */
291 			p++;
292 			continue;
293 		case '\\':
294 			/*
295 			 * backslash/newline combinations continue the
296 			 * statement. Skip it just like a whitespace.
297 			 */
298 			if (*(p + 1) == '\n') {
299 				p += 2;
300 				continue;
301 			}
302 			break;
303 		case '\n':
304 			/*
305 			 * Makefiles use a line-based syntax, where the newline
306 			 * is the end of a statement. After seeing a newline,
307 			 * we expect the next token is a target.
308 			 */
309 			p++;
310 			is_target = true;
311 			continue;
312 		case ':':
313 			/*
314 			 * assume the first dependency after a colon as the
315 			 * source file.
316 			 */
317 			p++;
318 			is_target = false;
319 			is_source = true;
320 			continue;
321 		}
322 
323 		/* find the end of the token */
324 		q = p;
325 		while (*q != ' ' && *q != '\t' && *q != '\n' && *q != '#' && *q != ':') {
326 			if (*q == '\\') {
327 				/*
328 				 * backslash/newline combinations work like as
329 				 * a whitespace, so this is the end of token.
330 				 */
331 				if (*(q + 1) == '\n')
332 					break;
333 
334 				/* escaped special characters */
335 				if (*(q + 1) == '#' || *(q + 1) == ':') {
336 					memmove(p + 1, p, q - p);
337 					p++;
338 				}
339 
340 				q++;
341 			}
342 
343 			if (*q == '\0')
344 				break;
345 			q++;
346 		}
347 
348 		/* Just discard the target */
349 		if (is_target) {
350 			p = q;
351 			continue;
352 		}
353 
354 		saved_c = *q;
355 		*q = '\0';
356 		need_parse = false;
357 
358 		/*
359 		 * Do not list the source file as dependency, so that kbuild is
360 		 * not confused if a .c file is rewritten into .S or vice versa.
361 		 * Storing it in source_* is needed for modpost to compute
362 		 * srcversions.
363 		 */
364 		if (is_source) {
365 			/*
366 			 * The DT build rule concatenates multiple dep files.
367 			 * When processing them, only process the first source
368 			 * name, which will be the original one, and ignore any
369 			 * other source names, which will be intermediate
370 			 * temporary files.
371 			 *
372 			 * rustc emits the same dependency list for each
373 			 * emission type. It is enough to list the source name
374 			 * just once.
375 			 */
376 			if (!saw_any_target) {
377 				saw_any_target = true;
378 				printf("source_%s := %s\n\n", target, p);
379 				printf("deps_%s := \\\n", target);
380 				need_parse = true;
381 			}
382 		} else if (!is_ignored_file(p, q - p) &&
383 			   !in_hashtable(p, q - p, file_hashtab)) {
384 			printf("  %s \\\n", p);
385 			need_parse = true;
386 		}
387 
388 		if (need_parse && !is_no_parse_file(p, q - p)) {
389 			void *buf;
390 
391 			buf = read_file(p);
392 			parse_config_file(buf);
393 			free(buf);
394 		}
395 
396 		is_source = false;
397 		*q = saved_c;
398 		p = q;
399 	}
400 
401 	if (!saw_any_target) {
402 		fprintf(stderr, "fixdep: parse error; no targets found\n");
403 		exit(1);
404 	}
405 
406 	printf("\n%s: $(deps_%s)\n\n", target, target);
407 	printf("$(deps_%s):\n", target);
408 }
409 
410 int main(int argc, char *argv[])
411 {
412 	const char *depfile, *target, *cmdline;
413 	void *buf;
414 
415 	if (argc != 4)
416 		usage();
417 
418 	depfile = argv[1];
419 	target = argv[2];
420 	cmdline = argv[3];
421 
422 	printf("savedcmd_%s := %s\n\n", target, cmdline);
423 
424 	buf = read_file(depfile);
425 	parse_dep_file(buf, target);
426 	free(buf);
427 
428 	fflush(stdout);
429 
430 	/*
431 	 * In the intended usage, the stdout is redirected to .*.cmd files.
432 	 * Call ferror() to catch errors such as "No space left on device".
433 	 */
434 	if (ferror(stdout)) {
435 		fprintf(stderr, "fixdep: not all data was written to the output\n");
436 		exit(1);
437 	}
438 
439 	return 0;
440 }
441