2 * Copyright (c) 1983, 1992, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * $FreeBSD: src/sys/libkern/mcount.c,v 1.16 1999/12/29 04:54:41 peter Exp $
36 #include <sys/param.h>
40 #include <sys/systm.h>
43 #include <vm/vm_param.h>
45 void bintr __P((void));
46 void btrap __P((void));
47 void eintr __P((void));
48 void user __P((void));
52 * mcount is called on entry to each function compiled with the profiling
53 * switch set. _mcount(), which is declared in a machine-dependent way
54 * with _MCOUNT_DECL, does the actual work and is either inlined into a
55 * C routine or called by an assembly stub. In any case, this magic is
56 * taken care of by the MCOUNT definition in <machine/profile.h>.
58 * _mcount updates data structures that represent traversals of the
59 * program's call graph edges. frompc and selfpc are the return
60 * address and function address that represents the given call graph edge.
62 * Note: the original BSD code used the same variable (frompcindex) for
63 * both frompcindex and frompc. Any reasonable, modern compiler will
64 * perform this optimization.
66 _MCOUNT_DECL(frompc, selfpc) /* _mcount; may be static, inline, etc */
67 register uintfptr_t frompc, selfpc;
72 register fptrdiff_t frompci;
73 register u_short *frompcindex;
74 register struct tostruct *top, *prevtop;
75 register struct gmonparam *p;
76 register long toindex;
82 #ifndef GUPROF /* XXX */
84 * check that we are profiling
85 * and that we aren't recursively invoked.
87 if (p->state != GMON_PROF_ON)
93 p->state = GMON_PROF_BUSY;
95 frompci = frompc - p->lowpc;
99 * When we are called from an exception handler, frompci may be
100 * for a user address. Convert such frompci's to the index of
101 * user() to merge all user counts.
103 if (frompci >= p->textsize) {
104 if (frompci + p->lowpc
105 >= (uintfptr_t)(VM_MAXUSER_ADDRESS + UPAGES * PAGE_SIZE))
107 frompci = (uintfptr_t)user - p->lowpc;
108 if (frompci >= p->textsize)
114 if (p->state == GMON_PROF_HIRES) {
116 * Count the time since cputime() was previously called
117 * against `frompc'. Compensate for overheads.
119 * cputime() sets its prev_count variable to the count when
120 * it is called. This in effect starts a counter for
121 * the next period of execution (normally from now until
122 * the next call to mcount() or mexitcount()). We set
123 * cputime_bias to compensate for our own overhead.
125 * We use the usual sampling counters since they can be
126 * located efficiently. 4-byte counters are usually
127 * necessary. gprof will add up the scattered counts
128 * just like it does for statistical profiling. All
129 * counts are signed so that underflow in the subtractions
130 * doesn't matter much (negative counts are normally
131 * compensated for by larger counts elsewhere). Underflow
132 * shouldn't occur, but may be caused by slightly wrong
133 * calibrations or from not clearing cputime_bias.
135 delta = cputime() - cputime_bias - p->mcount_pre_overhead;
136 cputime_bias = p->mcount_post_overhead;
137 KCOUNT(p, frompci) += delta;
138 *p->cputime_count += p->cputime_overhead;
139 *p->mcount_count += p->mcount_overhead;
145 * When we are called from an exception handler, frompc is faked
146 * to be for where the exception occurred. We've just solidified
147 * the count for there. Now convert frompci to the index of btrap()
148 * for trap handlers and bintr() for interrupt handlers to make
149 * exceptions appear in the call graph as calls from btrap() and
150 * bintr() instead of calls from all over.
152 if ((uintfptr_t)selfpc >= (uintfptr_t)btrap
153 && (uintfptr_t)selfpc < (uintfptr_t)eintr) {
154 if ((uintfptr_t)selfpc >= (uintfptr_t)bintr)
155 frompci = (uintfptr_t)bintr - p->lowpc;
157 frompci = (uintfptr_t)btrap - p->lowpc;
162 * check that frompc is a reasonable pc value.
163 * for example: signal catchers get called from the stack,
164 * not from text space. too bad.
166 if (frompci >= p->textsize)
169 frompcindex = &p->froms[frompci / (p->hashfraction * sizeof(*p->froms))];
170 toindex = *frompcindex;
173 * first time traversing this arc
175 toindex = ++p->tos[0].link;
176 if (toindex >= p->tolimit)
177 /* halt further profiling */
180 *frompcindex = toindex;
181 top = &p->tos[toindex];
182 top->selfpc = selfpc;
187 top = &p->tos[toindex];
188 if (top->selfpc == selfpc) {
190 * arc at front of chain; usual case.
196 * have to go looking down chain for it.
197 * top points to what we are looking at,
198 * prevtop points to previous top.
199 * we know it is not at the head of the chain.
201 for (; /* goto done */; ) {
202 if (top->link == 0) {
204 * top is end of the chain and none of the chain
205 * had top->selfpc == selfpc.
206 * so we allocate a new tostruct
207 * and link it to the head of the chain.
209 toindex = ++p->tos[0].link;
210 if (toindex >= p->tolimit)
213 top = &p->tos[toindex];
214 top->selfpc = selfpc;
216 top->link = *frompcindex;
217 *frompcindex = toindex;
221 * otherwise, check the next arc on the chain.
224 top = &p->tos[top->link];
225 if (top->selfpc == selfpc) {
228 * increment its count
229 * move it to the head of the chain.
232 toindex = prevtop->link;
233 prevtop->link = top->link;
234 top->link = *frompcindex;
235 *frompcindex = toindex;
244 p->state = GMON_PROF_ON;
248 p->state = GMON_PROF_ERROR;
256 * Actual definition of mcount function. Defined in <machine/profile.h>,
257 * which is included by <sys/gmon.h>.
267 uintfptr_t selfpcdiff;
270 selfpcdiff = selfpc - (uintfptr_t)p->lowpc;
271 if (selfpcdiff < p->textsize) {
275 * Count the time since cputime() was previously called
276 * against `selfpc'. Compensate for overheads.
278 delta = cputime() - cputime_bias - p->mexitcount_pre_overhead;
279 cputime_bias = p->mexitcount_post_overhead;
280 KCOUNT(p, selfpcdiff) += delta;
281 *p->cputime_count += p->cputime_overhead;
282 *p->mexitcount_count += p->mexitcount_overhead;
291 for (i = 0; i < CALIB_SCALE; i++)
305 for (i = 0; i < CALIB_SCALE; i++)