0
|
1 /*-
|
|
2 * Copyright (c) 1991 The Regents of the University of California.
|
|
3 * All rights reserved.
|
|
4 *
|
|
5 * Redistribution and use in source and binary forms, with or without
|
|
6 * modification, are permitted provided that the following conditions
|
|
7 * are met:
|
|
8 * 1. Redistributions of source code must retain the above copyright
|
|
9 * notice, this list of conditions and the following disclaimer.
|
|
10 * 2. Redistributions in binary form must reproduce the above copyright
|
|
11 * notice, this list of conditions and the following disclaimer in the
|
|
12 * documentation and/or other materials provided with the distribution.
|
|
13 * 3. [rescinded 22 July 1999]
|
|
14 * 4. Neither the name of the University nor the names of its contributors
|
|
15 * may be used to endorse or promote products derived from this software
|
|
16 * without specific prior written permission.
|
|
17 *
|
|
18 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
|
|
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
|
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
|
|
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
|
|
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
|
|
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
|
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
|
|
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
|
|
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
|
|
28 * SUCH DAMAGE.
|
|
29 */
|
|
30
|
|
31 /* Mangled into a form that works on SPARC Solaris 2 by Mark Eichin
|
|
32 * for Cygnus Support, July 1992.
|
|
33 */
|
|
34
|
|
35 #include "tconfig.h"
|
|
36 #include "tsystem.h"
|
|
37 #include <fcntl.h> /* for creat() */
|
|
38 #include "coretypes.h"
|
|
39 #include "tm.h"
|
|
40
|
|
41 #if 0
|
|
42 #include "sparc/gmon.h"
|
|
43 #else
|
|
44 struct phdr {
|
|
45 char *lpc;
|
|
46 char *hpc;
|
|
47 int ncnt;
|
|
48 };
|
|
49 #define HISTFRACTION 2
|
|
50 #define HISTCOUNTER unsigned short
|
|
51 #define HASHFRACTION 1
|
|
52 #define ARCDENSITY 2
|
|
53 #define MINARCS 50
|
|
54 struct tostruct {
|
|
55 char *selfpc;
|
|
56 long count;
|
|
57 unsigned short link;
|
|
58 };
|
|
59 struct rawarc {
|
|
60 unsigned long raw_frompc;
|
|
61 unsigned long raw_selfpc;
|
|
62 long raw_count;
|
|
63 };
|
|
64 #define ROUNDDOWN(x,y) (((x)/(y))*(y))
|
|
65 #define ROUNDUP(x,y) ((((x)+(y)-1)/(y))*(y))
|
|
66
|
|
67 #endif
|
|
68
|
|
69 /* extern mcount() asm ("mcount"); */
|
|
70 /*extern*/ char *minbrk /* asm ("minbrk") */;
|
|
71
|
|
72 /*
|
|
73 * froms is actually a bunch of unsigned shorts indexing tos
|
|
74 */
|
|
75 static int profiling = 3;
|
|
76 static unsigned short *froms;
|
|
77 static struct tostruct *tos = 0;
|
|
78 static long tolimit = 0;
|
|
79 static char *s_lowpc = 0;
|
|
80 static char *s_highpc = 0;
|
|
81 static unsigned long s_textsize = 0;
|
|
82
|
|
83 static int ssiz;
|
|
84 static char *sbuf;
|
|
85 static int s_scale;
|
|
86 /* see profil(2) where this is describe (incorrectly) */
|
|
87 #define SCALE_1_TO_1 0x10000L
|
|
88
|
|
89 #define MSG "No space for profiling buffer(s)\n"
|
|
90
|
|
91 static void moncontrol (int);
|
|
92 extern void monstartup (char *, char *);
|
|
93 extern void _mcleanup (void);
|
|
94
|
|
95 void monstartup(char *lowpc, char *highpc)
|
|
96 {
|
|
97 int monsize;
|
|
98 char *buffer;
|
|
99 register int o;
|
|
100
|
|
101 /*
|
|
102 * round lowpc and highpc to multiples of the density we're using
|
|
103 * so the rest of the scaling (here and in gprof) stays in ints.
|
|
104 */
|
|
105 lowpc = (char *)
|
|
106 ROUNDDOWN((unsigned long)lowpc, HISTFRACTION*sizeof(HISTCOUNTER));
|
|
107 s_lowpc = lowpc;
|
|
108 highpc = (char *)
|
|
109 ROUNDUP((unsigned long)highpc, HISTFRACTION*sizeof(HISTCOUNTER));
|
|
110 s_highpc = highpc;
|
|
111 s_textsize = highpc - lowpc;
|
|
112 monsize = (s_textsize / HISTFRACTION) + sizeof(struct phdr);
|
|
113 buffer = sbrk( monsize );
|
|
114 if ( buffer == (char *) -1 ) {
|
|
115 write( 2 , MSG , sizeof(MSG) );
|
|
116 return;
|
|
117 }
|
|
118 froms = (unsigned short *) sbrk( s_textsize / HASHFRACTION );
|
|
119 if ( froms == (unsigned short *) -1 ) {
|
|
120 write( 2 , MSG , sizeof(MSG) );
|
|
121 froms = 0;
|
|
122 return;
|
|
123 }
|
|
124 tolimit = s_textsize * ARCDENSITY / 100;
|
|
125 if ( tolimit < MINARCS ) {
|
|
126 tolimit = MINARCS;
|
|
127 } else if ( tolimit > 65534 ) {
|
|
128 tolimit = 65534;
|
|
129 }
|
|
130 tos = (struct tostruct *) sbrk( tolimit * sizeof( struct tostruct ) );
|
|
131 if ( tos == (struct tostruct *) -1 ) {
|
|
132 write( 2 , MSG , sizeof(MSG) );
|
|
133 froms = 0;
|
|
134 tos = 0;
|
|
135 return;
|
|
136 }
|
|
137 minbrk = sbrk(0);
|
|
138 tos[0].link = 0;
|
|
139 sbuf = buffer;
|
|
140 ssiz = monsize;
|
|
141 ( (struct phdr *) buffer ) -> lpc = lowpc;
|
|
142 ( (struct phdr *) buffer ) -> hpc = highpc;
|
|
143 ( (struct phdr *) buffer ) -> ncnt = ssiz;
|
|
144 monsize -= sizeof(struct phdr);
|
|
145 if ( monsize <= 0 )
|
|
146 return;
|
|
147 o = highpc - lowpc;
|
|
148 if( monsize < o )
|
|
149 #ifndef hp300
|
|
150 s_scale = ( (float) monsize / o ) * SCALE_1_TO_1;
|
|
151 #else /* avoid floating point */
|
|
152 {
|
|
153 int quot = o / monsize;
|
|
154
|
|
155 if (quot >= 0x10000)
|
|
156 s_scale = 1;
|
|
157 else if (quot >= 0x100)
|
|
158 s_scale = 0x10000 / quot;
|
|
159 else if (o >= 0x800000)
|
|
160 s_scale = 0x1000000 / (o / (monsize >> 8));
|
|
161 else
|
|
162 s_scale = 0x1000000 / ((o << 8) / monsize);
|
|
163 }
|
|
164 #endif
|
|
165 else
|
|
166 s_scale = SCALE_1_TO_1;
|
|
167 moncontrol(1);
|
|
168 }
|
|
169
|
|
170 void
|
|
171 _mcleanup(void)
|
|
172 {
|
|
173 int fd;
|
|
174 int fromindex;
|
|
175 int endfrom;
|
|
176 char *frompc;
|
|
177 int toindex;
|
|
178 struct rawarc rawarc;
|
|
179 char *profdir;
|
|
180 const char *proffile;
|
|
181 char *progname;
|
|
182 char buf[PATH_MAX];
|
|
183 extern char **___Argv;
|
|
184
|
|
185 moncontrol(0);
|
|
186
|
|
187 if ((profdir = getenv("PROFDIR")) != NULL) {
|
|
188 /* If PROFDIR contains a null value, no profiling output is produced */
|
|
189 if (*profdir == '\0') {
|
|
190 return;
|
|
191 }
|
|
192
|
|
193 progname=strrchr(___Argv[0], '/');
|
|
194 if (progname == NULL)
|
|
195 progname=___Argv[0];
|
|
196 else
|
|
197 progname++;
|
|
198
|
|
199 sprintf(buf, "%s/%ld.%s", profdir, (long) getpid(), progname);
|
|
200 proffile = buf;
|
|
201 } else {
|
|
202 proffile = "gmon.out";
|
|
203 }
|
|
204
|
|
205 fd = creat( proffile, 0666 );
|
|
206 if ( fd < 0 ) {
|
|
207 perror( proffile );
|
|
208 return;
|
|
209 }
|
|
210 # ifdef DEBUG
|
|
211 fprintf( stderr , "[mcleanup] sbuf 0x%x ssiz %d\n" , sbuf , ssiz );
|
|
212 # endif /* DEBUG */
|
|
213 write( fd , sbuf , ssiz );
|
|
214 endfrom = s_textsize / (HASHFRACTION * sizeof(*froms));
|
|
215 for ( fromindex = 0 ; fromindex < endfrom ; fromindex++ ) {
|
|
216 if ( froms[fromindex] == 0 ) {
|
|
217 continue;
|
|
218 }
|
|
219 frompc = s_lowpc + (fromindex * HASHFRACTION * sizeof(*froms));
|
|
220 for (toindex=froms[fromindex]; toindex!=0; toindex=tos[toindex].link) {
|
|
221 # ifdef DEBUG
|
|
222 fprintf( stderr ,
|
|
223 "[mcleanup] frompc 0x%x selfpc 0x%x count %d\n" ,
|
|
224 frompc , tos[toindex].selfpc , tos[toindex].count );
|
|
225 # endif /* DEBUG */
|
|
226 rawarc.raw_frompc = (unsigned long) frompc;
|
|
227 rawarc.raw_selfpc = (unsigned long) tos[toindex].selfpc;
|
|
228 rawarc.raw_count = tos[toindex].count;
|
|
229 write( fd , &rawarc , sizeof rawarc );
|
|
230 }
|
|
231 }
|
|
232 close( fd );
|
|
233 }
|
|
234
|
|
235 /*
|
|
236 * The SPARC stack frame is only held together by the frame pointers
|
|
237 * in the register windows. According to the SVR4 SPARC ABI
|
|
238 * Supplement, Low Level System Information/Operating System
|
|
239 * Interface/Software Trap Types, a type 3 trap will flush all of the
|
|
240 * register windows to the stack, which will make it possible to walk
|
|
241 * the frames and find the return addresses.
|
|
242 * However, it seems awfully expensive to incur a trap (system
|
|
243 * call) for every function call. It turns out that "call" simply puts
|
|
244 * the return address in %o7 expecting the "save" in the procedure to
|
|
245 * shift it into %i7; this means that before the "save" occurs, %o7
|
|
246 * contains the address of the call to mcount, and %i7 still contains
|
|
247 * the caller above that. The asm mcount here simply saves those
|
|
248 * registers in argument registers and branches to internal_mcount,
|
|
249 * simulating a call with arguments.
|
|
250 * Kludges:
|
|
251 * 1) the branch to internal_mcount is hard coded; it should be
|
|
252 * possible to tell asm to use the assembler-name of a symbol.
|
|
253 * 2) in theory, the function calling mcount could have saved %i7
|
|
254 * somewhere and reused the register; in practice, I *think* this will
|
|
255 * break longjmp (and maybe the debugger) but I'm not certain. (I take
|
|
256 * some comfort in the knowledge that it will break the native mcount
|
|
257 * as well.)
|
|
258 * 3) if builtin_return_address worked, this could be portable.
|
|
259 * However, it would really have to be optimized for arguments of 0
|
|
260 * and 1 and do something like what we have here in order to avoid the
|
|
261 * trap per function call performance hit.
|
|
262 * 4) the atexit and monsetup calls prevent this from simply
|
|
263 * being a leaf routine that doesn't do a "save" (and would thus have
|
|
264 * access to %o7 and %i7 directly) but the call to write() at the end
|
|
265 * would have also prevented this.
|
|
266 *
|
|
267 * -- [eichin:19920702.1107EST]
|
|
268 */
|
|
269
|
|
270 static void internal_mcount (char *, unsigned short *) __attribute__ ((used));
|
|
271
|
|
272 /* i7 == last ret, -> frompcindex */
|
|
273 /* o7 == current ret, -> selfpc */
|
|
274 /* Solaris 2 libraries use _mcount. */
|
|
275 asm(".global _mcount; _mcount: mov %i7,%o1; mov %o7,%o0;b,a internal_mcount");
|
|
276 /* This is for compatibility with old versions of gcc which used mcount. */
|
|
277 asm(".global mcount; mcount: mov %i7,%o1; mov %o7,%o0;b,a internal_mcount");
|
|
278
|
|
279 static void internal_mcount(char *selfpc, unsigned short *frompcindex)
|
|
280 {
|
|
281 register struct tostruct *top;
|
|
282 register struct tostruct *prevtop;
|
|
283 register long toindex;
|
|
284 static char already_setup;
|
|
285
|
|
286 /*
|
|
287 * find the return address for mcount,
|
|
288 * and the return address for mcount's caller.
|
|
289 */
|
|
290
|
|
291 if(!already_setup) {
|
|
292 extern char etext[];
|
|
293 extern char _start[];
|
|
294 extern char _init[];
|
|
295 already_setup = 1;
|
|
296 monstartup(_start < _init ? _start : _init, etext);
|
|
297 #ifdef USE_ONEXIT
|
|
298 on_exit(_mcleanup, 0);
|
|
299 #else
|
|
300 atexit(_mcleanup);
|
|
301 #endif
|
|
302 }
|
|
303 /*
|
|
304 * check that we are profiling
|
|
305 * and that we aren't recursively invoked.
|
|
306 */
|
|
307 if (profiling) {
|
|
308 goto out;
|
|
309 }
|
|
310 profiling++;
|
|
311 /*
|
|
312 * check that frompcindex is a reasonable pc value.
|
|
313 * for example: signal catchers get called from the stack,
|
|
314 * not from text space. too bad.
|
|
315 */
|
|
316 frompcindex = (unsigned short *)((long)frompcindex - (long)s_lowpc);
|
|
317 if ((unsigned long)frompcindex > s_textsize) {
|
|
318 goto done;
|
|
319 }
|
|
320 frompcindex =
|
|
321 &froms[((long)frompcindex) / (HASHFRACTION * sizeof(*froms))];
|
|
322 toindex = *frompcindex;
|
|
323 if (toindex == 0) {
|
|
324 /*
|
|
325 * first time traversing this arc
|
|
326 */
|
|
327 toindex = ++tos[0].link;
|
|
328 if (toindex >= tolimit) {
|
|
329 goto overflow;
|
|
330 }
|
|
331 *frompcindex = toindex;
|
|
332 top = &tos[toindex];
|
|
333 top->selfpc = selfpc;
|
|
334 top->count = 1;
|
|
335 top->link = 0;
|
|
336 goto done;
|
|
337 }
|
|
338 top = &tos[toindex];
|
|
339 if (top->selfpc == selfpc) {
|
|
340 /*
|
|
341 * arc at front of chain; usual case.
|
|
342 */
|
|
343 top->count++;
|
|
344 goto done;
|
|
345 }
|
|
346 /*
|
|
347 * have to go looking down chain for it.
|
|
348 * top points to what we are looking at,
|
|
349 * prevtop points to previous top.
|
|
350 * we know it is not at the head of the chain.
|
|
351 */
|
|
352 for (; /* goto done */; ) {
|
|
353 if (top->link == 0) {
|
|
354 /*
|
|
355 * top is end of the chain and none of the chain
|
|
356 * had top->selfpc == selfpc.
|
|
357 * so we allocate a new tostruct
|
|
358 * and link it to the head of the chain.
|
|
359 */
|
|
360 toindex = ++tos[0].link;
|
|
361 if (toindex >= tolimit) {
|
|
362 goto overflow;
|
|
363 }
|
|
364 top = &tos[toindex];
|
|
365 top->selfpc = selfpc;
|
|
366 top->count = 1;
|
|
367 top->link = *frompcindex;
|
|
368 *frompcindex = toindex;
|
|
369 goto done;
|
|
370 }
|
|
371 /*
|
|
372 * otherwise, check the next arc on the chain.
|
|
373 */
|
|
374 prevtop = top;
|
|
375 top = &tos[top->link];
|
|
376 if (top->selfpc == selfpc) {
|
|
377 /*
|
|
378 * there it is.
|
|
379 * increment its count
|
|
380 * move it to the head of the chain.
|
|
381 */
|
|
382 top->count++;
|
|
383 toindex = prevtop->link;
|
|
384 prevtop->link = top->link;
|
|
385 top->link = *frompcindex;
|
|
386 *frompcindex = toindex;
|
|
387 goto done;
|
|
388 }
|
|
389
|
|
390 }
|
|
391 done:
|
|
392 profiling--;
|
|
393 /* and fall through */
|
|
394 out:
|
|
395 return; /* normal return restores saved registers */
|
|
396
|
|
397 overflow:
|
|
398 profiling++; /* halt further profiling */
|
|
399 # define TOLIMIT "mcount: tos overflow\n"
|
|
400 write(2, TOLIMIT, sizeof(TOLIMIT));
|
|
401 goto out;
|
|
402 }
|
|
403
|
|
404 /*
|
|
405 * Control profiling
|
|
406 * profiling is what mcount checks to see if
|
|
407 * all the data structures are ready.
|
|
408 */
|
|
409 static void moncontrol(int mode)
|
|
410 {
|
|
411 if (mode) {
|
|
412 /* start */
|
|
413 profil((unsigned short *)(sbuf + sizeof(struct phdr)),
|
|
414 ssiz - sizeof(struct phdr),
|
|
415 (long)s_lowpc, s_scale);
|
|
416 profiling = 0;
|
|
417 } else {
|
|
418 /* stop */
|
|
419 profil((unsigned short *)0, 0, 0, 0);
|
|
420 profiling = 3;
|
|
421 }
|
|
422 }
|