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 $
34 * $DragonFly: src/sys/libkern/mcount.c,v 1.7 2004/04/28 22:05:09 hmp Exp $
37 #include <sys/param.h>
41 #include <sys/systm.h>
44 #include <vm/vm_param.h>
53 * mcount is called on entry to each function compiled with the profiling
54 * switch set. _mcount(), which is declared in a machine-dependent way
55 * with _MCOUNT_DECL, does the actual work and is either inlined into a
56 * C routine or called by an assembly stub. In any case, this magic is
57 * taken care of by the MCOUNT definition in <machine/profile.h>.
59 * _mcount updates data structures that represent traversals of the
60 * program's call graph edges. frompc and selfpc are the return
61 * address and function address that represents the given call graph edge.
63 * Note: the original BSD code used the same variable (frompcindex) for
64 * both frompcindex and frompc. Any reasonable, modern compiler will
65 * perform this optimization.
67 /* _mcount; may be static, inline, etc */
68 _MCOUNT_DECL(uintfptr_t frompc, uintfptr_t selfpc)
75 struct tostruct *top, *prevtop;
83 #ifndef GUPROF /* XXX */
85 * check that we are profiling
86 * and that we aren't recursively invoked.
88 if (p->state != GMON_PROF_ON)
94 p->state = GMON_PROF_BUSY;
96 frompci = frompc - p->lowpc;
100 * When we are called from an exception handler, frompci may be
101 * for a user address. Convert such frompci's to the index of
102 * user() to merge all user counts.
104 if (frompci >= p->textsize) {
105 if (frompci + p->lowpc
106 >= (uintfptr_t)(VM_MAXUSER_ADDRESS + UPAGES * PAGE_SIZE))
108 frompci = (uintfptr_t)user - p->lowpc;
109 if (frompci >= p->textsize)
115 if (p->state == GMON_PROF_HIRES) {
117 * Count the time since cputime() was previously called
118 * against `frompc'. Compensate for overheads.
120 * cputime() sets its prev_count variable to the count when
121 * it is called. This in effect starts a counter for
122 * the next period of execution (normally from now until
123 * the next call to mcount() or mexitcount()). We set
124 * cputime_bias to compensate for our own overhead.
126 * We use the usual sampling counters since they can be
127 * located efficiently. 4-byte counters are usually
128 * necessary. gprof will add up the scattered counts
129 * just like it does for statistical profiling. All
130 * counts are signed so that underflow in the subtractions
131 * doesn't matter much (negative counts are normally
132 * compensated for by larger counts elsewhere). Underflow
133 * shouldn't occur, but may be caused by slightly wrong
134 * calibrations or from not clearing cputime_bias.
136 delta = cputime() - cputime_bias - p->mcount_pre_overhead;
137 cputime_bias = p->mcount_post_overhead;
138 KCOUNT(p, frompci) += delta;
139 *p->cputime_count += p->cputime_overhead;
140 *p->mcount_count += p->mcount_overhead;
146 * When we are called from an exception handler, frompc is faked
147 * to be for where the exception occurred. We've just solidified
148 * the count for there. Now convert frompci to the index of btrap()
149 * for trap handlers and bintr() for interrupt handlers to make
150 * exceptions appear in the call graph as calls from btrap() and
151 * bintr() instead of calls from all over.
153 if ((uintfptr_t)selfpc >= (uintfptr_t)btrap
154 && (uintfptr_t)selfpc < (uintfptr_t)eintr) {
155 if ((uintfptr_t)selfpc >= (uintfptr_t)bintr)
156 frompci = (uintfptr_t)bintr - p->lowpc;
158 frompci = (uintfptr_t)btrap - p->lowpc;
163 * check that frompc is a reasonable pc value.
164 * for example: signal catchers get called from the stack,
165 * not from text space. too bad.
167 if (frompci >= p->textsize)
170 frompcindex = &p->froms[frompci / (p->hashfraction * sizeof(*p->froms))];
171 toindex = *frompcindex;
174 * first time traversing this arc
176 toindex = ++p->tos[0].link;
177 if (toindex >= p->tolimit)
178 /* halt further profiling */
181 *frompcindex = toindex;
182 top = &p->tos[toindex];
183 top->selfpc = selfpc;
188 top = &p->tos[toindex];
189 if (top->selfpc == selfpc) {
191 * arc at front of chain; usual case.
197 * have to go looking down chain for it.
198 * top points to what we are looking at,
199 * prevtop points to previous top.
200 * we know it is not at the head of the chain.
202 for (; /* goto done */; ) {
203 if (top->link == 0) {
205 * top is end of the chain and none of the chain
206 * had top->selfpc == selfpc.
207 * so we allocate a new tostruct
208 * and link it to the head of the chain.
210 toindex = ++p->tos[0].link;
211 if (toindex >= p->tolimit)
214 top = &p->tos[toindex];
215 top->selfpc = selfpc;
217 top->link = *frompcindex;
218 *frompcindex = toindex;
222 * otherwise, check the next arc on the chain.
225 top = &p->tos[top->link];
226 if (top->selfpc == selfpc) {
229 * increment its count
230 * move it to the head of the chain.
233 toindex = prevtop->link;
234 prevtop->link = top->link;
235 top->link = *frompcindex;
236 *frompcindex = toindex;
245 p->state = GMON_PROF_ON;
249 p->state = GMON_PROF_ERROR;
257 * Actual definition of mcount function. Defined in <machine/profile.h>,
258 * which is included by <sys/gmon.h>.
264 mexitcount(uintfptr_t selfpc)
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++)