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 1.9 2007/01/12 22:12:50 dillon 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 * XXX doesn't work properly with vkernel
106 if (frompci
>= p
->textsize
) {
107 if (frompci
+ p
->lowpc
108 >= (uintfptr_t
)(VM_MAX_USER_ADDRESS
+ UPAGES
* PAGE_SIZE
))
110 frompci
= (uintfptr_t
)user
- p
->lowpc
;
111 if (frompci
>= p
->textsize
)
117 if (p
->state
== GMON_PROF_HIRES
) {
119 * Count the time since cputime() was previously called
120 * against `frompc'. Compensate for overheads.
122 * cputime() sets its prev_count variable to the count when
123 * it is called. This in effect starts a counter for
124 * the next period of execution (normally from now until
125 * the next call to mcount() or mexitcount()). We set
126 * cputime_bias to compensate for our own overhead.
128 * We use the usual sampling counters since they can be
129 * located efficiently. 4-byte counters are usually
130 * necessary. gprof will add up the scattered counts
131 * just like it does for statistical profiling. All
132 * counts are signed so that underflow in the subtractions
133 * doesn't matter much (negative counts are normally
134 * compensated for by larger counts elsewhere). Underflow
135 * shouldn't occur, but may be caused by slightly wrong
136 * calibrations or from not clearing cputime_bias.
138 delta
= cputime() - cputime_bias
- p
->mcount_pre_overhead
;
139 cputime_bias
= p
->mcount_post_overhead
;
140 KCOUNT(p
, frompci
) += delta
;
141 *p
->cputime_count
+= p
->cputime_overhead
;
142 *p
->mcount_count
+= p
->mcount_overhead
;
148 * When we are called from an exception handler, frompc is faked
149 * to be for where the exception occurred. We've just solidified
150 * the count for there. Now convert frompci to the index of btrap()
151 * for trap handlers and bintr() for interrupt handlers to make
152 * exceptions appear in the call graph as calls from btrap() and
153 * bintr() instead of calls from all over.
155 if ((uintfptr_t
)selfpc
>= (uintfptr_t
)btrap
156 && (uintfptr_t
)selfpc
< (uintfptr_t
)eintr
) {
157 if ((uintfptr_t
)selfpc
>= (uintfptr_t
)bintr
)
158 frompci
= (uintfptr_t
)bintr
- p
->lowpc
;
160 frompci
= (uintfptr_t
)btrap
- p
->lowpc
;
165 * check that frompc is a reasonable pc value.
166 * for example: signal catchers get called from the stack,
167 * not from text space. too bad.
169 if (frompci
>= p
->textsize
)
172 frompcindex
= &p
->froms
[frompci
/ (p
->hashfraction
* sizeof(*p
->froms
))];
173 toindex
= *frompcindex
;
176 * first time traversing this arc
178 toindex
= ++p
->tos
[0].link
;
179 if (toindex
>= p
->tolimit
)
180 /* halt further profiling */
183 *frompcindex
= toindex
;
184 top
= &p
->tos
[toindex
];
185 top
->selfpc
= selfpc
;
190 top
= &p
->tos
[toindex
];
191 if (top
->selfpc
== selfpc
) {
193 * arc at front of chain; usual case.
199 * have to go looking down chain for it.
200 * top points to what we are looking at,
201 * prevtop points to previous top.
202 * we know it is not at the head of the chain.
204 for (; /* goto done */; ) {
205 if (top
->link
== 0) {
207 * top is end of the chain and none of the chain
208 * had top->selfpc == selfpc.
209 * so we allocate a new tostruct
210 * and link it to the head of the chain.
212 toindex
= ++p
->tos
[0].link
;
213 if (toindex
>= p
->tolimit
)
216 top
= &p
->tos
[toindex
];
217 top
->selfpc
= selfpc
;
219 top
->link
= *frompcindex
;
220 *frompcindex
= toindex
;
224 * otherwise, check the next arc on the chain.
227 top
= &p
->tos
[top
->link
];
228 if (top
->selfpc
== selfpc
) {
231 * increment its count
232 * move it to the head of the chain.
235 toindex
= prevtop
->link
;
236 prevtop
->link
= top
->link
;
237 top
->link
= *frompcindex
;
238 *frompcindex
= toindex
;
247 p
->state
= GMON_PROF_ON
;
251 p
->state
= GMON_PROF_ERROR
;
259 * Actual definition of mcount function. Defined in <machine/profile.h>,
260 * which is included by <sys/gmon.h>.
266 mexitcount(uintfptr_t selfpc
)
269 uintfptr_t selfpcdiff
;
272 selfpcdiff
= selfpc
- (uintfptr_t
)p
->lowpc
;
273 if (selfpcdiff
< p
->textsize
) {
277 * Count the time since cputime() was previously called
278 * against `selfpc'. Compensate for overheads.
280 delta
= cputime() - cputime_bias
- p
->mexitcount_pre_overhead
;
281 cputime_bias
= p
->mexitcount_post_overhead
;
282 KCOUNT(p
, selfpcdiff
) += delta
;
283 *p
->cputime_count
+= p
->cputime_overhead
;
284 *p
->mexitcount_count
+= p
->mexitcount_overhead
;
293 for (i
= 0; i
< CALIB_SCALE
; i
++)
307 for (i
= 0; i
< CALIB_SCALE
; i
++)