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 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 #if !defined(_KERNEL) && defined(LIBC_SCCS) && !defined(lint)
31 static char sccsid[] = "@(#)mcount.c 8.1 (Berkeley) 6/4/93";
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
36 #include <sys/param.h>
39 #include <sys/systm.h>
41 #include <vm/vm_param.h>
48 #include <machine/atomic.h>
51 * mcount is called on entry to each function compiled with the profiling
52 * switch set. _mcount(), which is declared in a machine-dependent way
53 * with _MCOUNT_DECL, does the actual work and is either inlined into a
54 * C routine or called by an assembly stub. In any case, this magic is
55 * taken care of by the MCOUNT definition in <machine/profile.h>.
57 * _mcount updates data structures that represent traversals of the
58 * program's call graph edges. frompc and selfpc are the return
59 * address and function address that represents the given call graph edge.
61 * Note: the original BSD code used the same variable (frompcindex) for
62 * both frompcindex and frompc. Any reasonable, modern compiler will
63 * perform this optimization.
65 /* _mcount; may be static, inline, etc */
66 _MCOUNT_DECL(uintfptr_t frompc, uintfptr_t selfpc)
73 struct tostruct *top, *prevtop;
81 #ifndef GUPROF /* XXX */
83 * check that we are profiling
84 * and that we aren't recursively invoked.
86 if (p->state != GMON_PROF_ON)
92 if (!atomic_cmpset_acq_int(&p->state, GMON_PROF_ON, 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)
115 goto skip_guprof_stuff;
117 * Look at the clock and add the count of clock cycles since the
118 * clock was last looked at to a counter for frompc. This
119 * solidifies the count for the function containing frompc and
120 * effectively starts another clock for the current function.
121 * The count for the new clock will be solidified when another
122 * function call is made or the function returns.
124 * We use the usual sampling counters since they can be located
125 * efficiently. 4-byte counters are usually necessary.
127 * There are many complications for subtracting the profiling
128 * overheads from the counts for normal functions and adding
129 * them to the counts for mcount(), mexitcount() and cputime().
130 * We attempt to handle fractional cycles, but the overheads
131 * are usually underestimated because they are calibrated for
132 * a simpler than usual setup.
134 delta = cputime() - p->mcount_overhead;
135 p->cputime_overhead_resid += p->cputime_overhead_frac;
136 p->mcount_overhead_resid += p->mcount_overhead_frac;
138 *p->mcount_count += delta + p->mcount_overhead
139 - p->cputime_overhead;
140 else if (delta != 0) {
141 if (p->cputime_overhead_resid >= CALIB_SCALE) {
142 p->cputime_overhead_resid -= CALIB_SCALE;
147 if (p->mcount_overhead_resid >= CALIB_SCALE) {
148 p->mcount_overhead_resid -= CALIB_SCALE;
152 KCOUNT(p, frompci) += delta;
154 *p->mcount_count += p->mcount_overhead_sub;
156 *p->cputime_count += p->cputime_overhead;
162 * When we are called from an exception handler, frompc is faked
163 * to be for where the exception occurred. We've just solidified
164 * the count for there. Now convert frompci to the index of btrap()
165 * for trap handlers and bintr() for interrupt handlers to make
166 * exceptions appear in the call graph as calls from btrap() and
167 * bintr() instead of calls from all over.
169 if ((uintfptr_t)selfpc >= (uintfptr_t)btrap
170 && (uintfptr_t)selfpc < (uintfptr_t)eintr) {
171 if ((uintfptr_t)selfpc >= (uintfptr_t)bintr)
172 frompci = (uintfptr_t)bintr - p->lowpc;
174 frompci = (uintfptr_t)btrap - p->lowpc;
179 * check that frompc is a reasonable pc value.
180 * for example: signal catchers get called from the stack,
181 * not from text space. too bad.
183 if (frompci >= p->textsize)
186 frompcindex = &p->froms[frompci / (p->hashfraction * sizeof(*p->froms))];
187 toindex = *frompcindex;
190 * first time traversing this arc
192 toindex = ++p->tos[0].link;
193 if (toindex >= p->tolimit)
194 /* halt further profiling */
197 *frompcindex = toindex;
198 top = &p->tos[toindex];
199 top->selfpc = selfpc;
204 top = &p->tos[toindex];
205 if (top->selfpc == selfpc) {
207 * arc at front of chain; usual case.
213 * have to go looking down chain for it.
214 * top points to what we are looking at,
215 * prevtop points to previous top.
216 * we know it is not at the head of the chain.
218 for (; /* goto done */; ) {
219 if (top->link == 0) {
221 * top is end of the chain and none of the chain
222 * had top->selfpc == selfpc.
223 * so we allocate a new tostruct
224 * and link it to the head of the chain.
226 toindex = ++p->tos[0].link;
227 if (toindex >= p->tolimit)
230 top = &p->tos[toindex];
231 top->selfpc = selfpc;
233 top->link = *frompcindex;
234 *frompcindex = toindex;
238 * otherwise, check the next arc on the chain.
241 top = &p->tos[top->link];
242 if (top->selfpc == selfpc) {
245 * increment its count
246 * move it to the head of the chain.
249 toindex = prevtop->link;
250 prevtop->link = top->link;
251 top->link = *frompcindex;
252 *frompcindex = toindex;
261 atomic_store_rel_int(&p->state, GMON_PROF_ON);
265 atomic_store_rel_int(&p->state, GMON_PROF_ERROR);
273 * Actual definition of mcount function. Defined in <machine/profile.h>,
274 * which is included by <sys/gmon.h>.
284 uintfptr_t selfpcdiff;
287 selfpcdiff = selfpc - (uintfptr_t)p->lowpc;
288 if (selfpcdiff < p->textsize) {
292 * Solidify the count for the current function.
294 delta = cputime() - p->mexitcount_overhead;
295 p->cputime_overhead_resid += p->cputime_overhead_frac;
296 p->mexitcount_overhead_resid += p->mexitcount_overhead_frac;
298 *p->mexitcount_count += delta + p->mexitcount_overhead
299 - p->cputime_overhead;
300 else if (delta != 0) {
301 if (p->cputime_overhead_resid >= CALIB_SCALE) {
302 p->cputime_overhead_resid -= CALIB_SCALE;
307 if (p->mexitcount_overhead_resid
309 p->mexitcount_overhead_resid
311 ++*p->mexitcount_count;
314 KCOUNT(p, selfpcdiff) += delta;
316 *p->mexitcount_count += p->mexitcount_overhead_sub;
318 *p->cputime_count += p->cputime_overhead;