]> git.saurik.com Git - apple/xnu.git/blob - bsd/kern/subr_prof.c
ed0455eeb49e95c000e675f7d37067cb14503df9
[apple/xnu.git] / bsd / kern / subr_prof.c
1 /*
2 * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
3 *
4 * @APPLE_LICENSE_HEADER_START@
5 *
6 * The contents of this file constitute Original Code as defined in and
7 * are subject to the Apple Public Source License Version 1.1 (the
8 * "License"). You may not use this file except in compliance with the
9 * License. Please obtain a copy of the License at
10 * http://www.apple.com/publicsource and read it before using this file.
11 *
12 * This Original Code and all software distributed under the License are
13 * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
14 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
15 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the
17 * License for the specific language governing rights and limitations
18 * under the License.
19 *
20 * @APPLE_LICENSE_HEADER_END@
21 */
22 /* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */
23 /*-
24 * Copyright (c) 1982, 1986, 1993
25 * The Regents of the University of California. All rights reserved.
26 *
27 * Redistribution and use in source and binary forms, with or without
28 * modification, are permitted provided that the following conditions
29 * are met:
30 * 1. Redistributions of source code must retain the above copyright
31 * notice, this list of conditions and the following disclaimer.
32 * 2. Redistributions in binary form must reproduce the above copyright
33 * notice, this list of conditions and the following disclaimer in the
34 * documentation and/or other materials provided with the distribution.
35 * 3. All advertising materials mentioning features or use of this software
36 * must display the following acknowledgement:
37 * This product includes software developed by the University of
38 * California, Berkeley and its contributors.
39 * 4. Neither the name of the University nor the names of its contributors
40 * may be used to endorse or promote products derived from this software
41 * without specific prior written permission.
42 *
43 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
44 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
45 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
46 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
47 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
48 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
49 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
50 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
51 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
52 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
53 * SUCH DAMAGE.
54 *
55 * @(#)subr_prof.c 8.3 (Berkeley) 9/23/93
56 */
57
58 #include <sys/param.h>
59 #include <sys/systm.h>
60 #include <sys/kernel.h>
61 #include <sys/proc.h>
62 #include <sys/user.h>
63 #include <machine/spl.h>
64
65 #include <sys/mount.h>
66
67 #include <kern/cpu_number.h>
68
69 #ifdef GPROF
70 #include <sys/malloc.h>
71 #include <sys/gmon.h>
72 #include <kern/mach_header.h>
73 #include <machine/profile.h>
74
75 /*
76 * Froms is actually a bunch of unsigned shorts indexing tos
77 */
78 struct gmonparam _gmonparam = { GMON_PROF_OFF };
79
80 kmstartup()
81 {
82 char *cp;
83 u_long fromssize, tossize;
84 struct segment_command *sgp;
85 struct gmonparam *p = &_gmonparam;
86
87 sgp = getsegbyname("__TEXT");
88 p->lowpc = (u_long)sgp->vmaddr;
89 p->highpc = (u_long)(sgp->vmaddr + sgp->vmsize);
90
91 /*
92 * Round lowpc and highpc to multiples of the density we're using
93 * so the rest of the scaling (here and in gprof) stays in ints.
94 */
95 p->lowpc = ROUNDDOWN(p->lowpc, HISTFRACTION * sizeof(HISTCOUNTER));
96 p->highpc = ROUNDUP(p->highpc, HISTFRACTION * sizeof(HISTCOUNTER));
97 p->textsize = p->highpc - p->lowpc;
98 printf("Profiling kernel, textsize=%d [0x%08x..0x%08x]\n",
99 p->textsize, p->lowpc, p->highpc);
100 p->kcountsize = p->textsize / HISTFRACTION;
101 p->hashfraction = HASHFRACTION;
102 p->fromssize = p->textsize / HASHFRACTION;
103 p->tolimit = p->textsize * ARCDENSITY / 100;
104 if (p->tolimit < MINARCS)
105 p->tolimit = MINARCS;
106 else if (p->tolimit > MAXARCS)
107 p->tolimit = MAXARCS;
108 p->tossize = p->tolimit * sizeof(struct tostruct);
109 /* Why not use MALLOC with M_GPROF ? */
110 cp = (char *)kalloc(p->kcountsize + p->fromssize + p->tossize);
111 if (cp == 0) {
112 printf("No memory for profiling.\n");
113 return;
114 }
115 bzero(cp, p->kcountsize + p->tossize + p->fromssize);
116 p->tos = (struct tostruct *)cp;
117 cp += p->tossize;
118 p->kcount = (u_short *)cp;
119 cp += p->kcountsize;
120 p->froms = (u_short *)cp;
121 }
122
123 /*
124 * Return kernel profiling information.
125 */
126 sysctl_doprof(name, namelen, oldp, oldlenp, newp, newlen, p)
127 int *name;
128 u_int namelen;
129 void *oldp;
130 size_t *oldlenp;
131 void *newp;
132 size_t newlen;
133 {
134 struct gmonparam *gp = &_gmonparam;
135 int error;
136
137 /* all sysctl names at this level are terminal */
138 if (namelen != 1)
139 return (ENOTDIR); /* overloaded */
140
141 switch (name[0]) {
142 case GPROF_STATE:
143 error = sysctl_int(oldp, oldlenp, newp, newlen, &gp->state);
144 if (error)
145 return (error);
146 if (gp->state == GMON_PROF_OFF)
147 stopprofclock(kernproc);
148 else
149 startprofclock(kernproc);
150 return (0);
151 case GPROF_COUNT:
152 return (sysctl_struct(oldp, oldlenp, newp, newlen,
153 gp->kcount, gp->kcountsize));
154 case GPROF_FROMS:
155 return (sysctl_struct(oldp, oldlenp, newp, newlen,
156 gp->froms, gp->fromssize));
157 case GPROF_TOS:
158 return (sysctl_struct(oldp, oldlenp, newp, newlen,
159 gp->tos, gp->tossize));
160 case GPROF_GMONPARAM:
161 return (sysctl_rdstruct(oldp, oldlenp, newp, gp, sizeof *gp));
162 default:
163 return (EOPNOTSUPP);
164 }
165 /* NOTREACHED */
166 }
167
168
169 /*
170 * mcount() called with interrupts disabled.
171 */
172 void
173 mcount(
174 register u_long frompc,
175 register u_long selfpc
176 )
177 {
178 unsigned short *frompcindex;
179 register struct tostruct *top, *prevtop;
180 struct gmonparam *p = &_gmonparam;
181 register long toindex;
182 MCOUNT_INIT;
183
184 /*
185 * check that we are profiling
186 * and that we aren't recursively invoked.
187 */
188 if (p->state != GMON_PROF_ON)
189 return;
190
191 MCOUNT_ENTER;
192
193 /*
194 * check that frompcindex is a reasonable pc value.
195 * for example: signal catchers get called from the stack,
196 * not from text space. too bad.
197 */
198 frompc -= p->lowpc;
199 if (frompc > p->textsize)
200 goto done;
201
202 frompcindex = &p->froms[frompc / (p->hashfraction * sizeof(*p->froms))];
203 toindex = *frompcindex;
204 if (toindex == 0) {
205 /*
206 * first time traversing this arc
207 */
208 toindex = ++p->tos[0].link;
209 if (toindex >= p->tolimit) {
210 /* halt further profiling */
211 goto overflow;
212 }
213 *frompcindex = toindex;
214 top = &p->tos[toindex];
215 top->selfpc = selfpc;
216 top->count = 1;
217 top->link = 0;
218 goto done;
219 }
220 top = &p->tos[toindex];
221 if (top->selfpc == selfpc) {
222 /*
223 * arc at front of chain; usual case.
224 */
225 top->count++;
226 goto done;
227 }
228 /*
229 * have to go looking down chain for it.
230 * top points to what we are looking at,
231 * prevtop points to previous top.
232 * we know it is not at the head of the chain.
233 */
234 for (; /* goto done */; ) {
235 if (top->link == 0) {
236 /*
237 * top is end of the chain and none of the chain
238 * had top->selfpc == selfpc.
239 * so we allocate a new tostruct
240 * and link it to the head of the chain.
241 */
242 toindex = ++p->tos[0].link;
243 if (toindex >= p->tolimit) {
244 goto overflow;
245 }
246 top = &p->tos[toindex];
247 top->selfpc = selfpc;
248 top->count = 1;
249 top->link = *frompcindex;
250 *frompcindex = toindex;
251 goto done;
252 }
253 /*
254 * otherwise, check the next arc on the chain.
255 */
256 prevtop = top;
257 top = &p->tos[top->link];
258 if (top->selfpc == selfpc) {
259 /*
260 * there it is.
261 * increment its count
262 * move it to the head of the chain.
263 */
264 top->count++;
265 toindex = prevtop->link;
266 prevtop->link = top->link;
267 top->link = *frompcindex;
268 *frompcindex = toindex;
269 goto done;
270 }
271
272 }
273 done:
274 MCOUNT_EXIT;
275 return;
276
277 overflow:
278 p->state = GMON_PROF_ERROR;
279 MCOUNT_EXIT;
280 printf("mcount: tos overflow\n");
281 return;
282 }
283
284 #endif /* GPROF */
285
286 #if NCPUS > 1
287 #define PROFILE_LOCK(x) simple_lock(x)
288 #define PROFILE_UNLOCK(x) simple_unlock(x)
289 #else
290 #define PROFILE_LOCK(x)
291 #define PROFILE_UNLOCK(x)
292 #endif
293
294 struct profil_args {
295 short *bufbase;
296 u_int bufsize;
297 u_int pcoffset;
298 u_int pcscale;
299 };
300 int
301 profil(p, uap, retval)
302 struct proc *p;
303 register struct profil_args *uap;
304 register_t *retval;
305 {
306 register struct uprof *upp = &p->p_stats->p_prof;
307 struct uprof *upc, *nupc;
308 int s;
309
310 if (uap->pcscale > (1 << 16))
311 return (EINVAL);
312 if (uap->pcscale == 0) {
313 stopprofclock(p);
314 return (0);
315 }
316
317 /* Block profile interrupts while changing state. */
318 s = splstatclock();
319 PROFILE_LOCK(&upp->pr_lock);
320 upp->pr_base = (caddr_t)uap->bufbase;
321 upp->pr_size = uap->bufsize;
322 upp->pr_off = uap->pcoffset;
323 upp->pr_scale = uap->pcscale;
324
325 /* remove buffers previously allocated with add_profil() */
326 for (upc = upp->pr_next; upc; upc = nupc) {
327 nupc = upc->pr_next;
328 kfree(upc, sizeof (struct uprof));
329 }
330
331 upp->pr_next = 0;
332 PROFILE_UNLOCK(&upp->pr_lock);
333 startprofclock(p);
334 splx(s);
335 return(0);
336 }
337
338 struct add_profile_args {
339 short *bufbase;
340 u_int bufsize;
341 u_int pcoffset;
342 u_int pcscale;
343 };
344 int
345 add_profil(p, uap, retval)
346 struct proc *p;
347 register struct add_profile_args *uap;
348 register_t *retval;
349 {
350 struct uprof *upp = &p->p_stats->p_prof, *upc;
351 int s;
352
353 if (upp->pr_scale == 0)
354 return (0);
355 s = splstatclock();
356 upc = (struct uprof *) kalloc(sizeof (struct uprof));
357 upc->pr_base = (caddr_t)uap->bufbase;
358 upc->pr_size = uap->bufsize;
359 upc->pr_off = uap->pcoffset;
360 upc->pr_scale = uap->pcscale;
361 PROFILE_LOCK(&upp->pr_lock);
362 upc->pr_next = upp->pr_next;
363 upp->pr_next = upc;
364 PROFILE_UNLOCK(&upp->pr_lock);
365 splx(s);
366 return(0);
367 }
368
369 /*
370 * Scale is a fixed-point number with the binary point 16 bits
371 * into the value, and is <= 1.0. pc is at most 32 bits, so the
372 * intermediate result is at most 48 bits.
373 */
374 #define PC_TO_INDEX(pc, prof) \
375 ((int)(((u_quad_t)((pc) - (prof)->pr_off) * \
376 (u_quad_t)((prof)->pr_scale)) >> 16) & ~1)
377
378 /*
379 * Collect user-level profiling statistics; called on a profiling tick,
380 * when a process is running in user-mode. We use
381 * an AST that will vector us to trap() with a context in which copyin
382 * and copyout will work. Trap will then call addupc_task().
383 *
384 * Note that we may (rarely) not get around to the AST soon enough, and
385 * lose profile ticks when the next tick overwrites this one, but in this
386 * case the system is overloaded and the profile is probably already
387 * inaccurate.
388 *
389 * We can afford to take faults here. If the
390 * update fails, we simply turn off profiling.
391 */
392 void
393 addupc_task(p, pc, ticks)
394 register struct proc *p;
395 register u_long pc;
396 u_int ticks;
397 {
398 register struct uprof *prof;
399 register short *cell;
400 register u_int off;
401 u_short count;
402
403 /* Testing P_PROFIL may be unnecessary, but is certainly safe. */
404 if ((p->p_flag & P_PROFIL) == 0 || ticks == 0)
405 return;
406
407 for (prof = &p->p_stats->p_prof; prof; prof = prof->pr_next) {
408 off = PC_TO_INDEX(pc,prof);
409 cell = (short *)(prof->pr_base + off);
410 if (cell >= (short *)prof->pr_base &&
411 cell < (short*)(prof->pr_size + (int) prof->pr_base)) {
412 if (copyin((caddr_t)cell, (caddr_t) &count, sizeof(count)) == 0) {
413 count += ticks;
414 if(copyout((caddr_t) &count, (caddr_t)cell, sizeof(count)) == 0)
415 return;
416 }
417 p->p_stats->p_prof.pr_scale = 0;
418 stopprofclock(p);
419 break;
420 }
421 }
422 }