mcount.c revision 38846
1/*-
2 * Copyright (c) 1983, 1992, 1993
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
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.
20 *
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
31 * SUCH DAMAGE.
32 */
33
34#if !defined(lint) && !defined(KERNEL) && defined(LIBC_SCCS)
35#if 0
36static char sccsid[] = "@(#)mcount.c	8.1 (Berkeley) 6/4/93";
37#endif
38static const char rcsid[] =
39	"$Id: mcount.c,v 1.13 1998/07/14 05:09:41 bde Exp $";
40#endif
41
42#ifndef __alpha__
43
44#include <sys/param.h>
45#include <sys/gmon.h>
46#ifdef KERNEL
47#include <sys/systm.h>
48#include <vm/vm.h>
49#include <vm/vm_param.h>
50#include <vm/pmap.h>
51void	bintr __P((void));
52void	btrap __P((void));
53void	eintr __P((void));
54void	user __P((void));
55#endif
56
57/*
58 * mcount is called on entry to each function compiled with the profiling
59 * switch set.  _mcount(), which is declared in a machine-dependent way
60 * with _MCOUNT_DECL, does the actual work and is either inlined into a
61 * C routine or called by an assembly stub.  In any case, this magic is
62 * taken care of by the MCOUNT definition in <machine/profile.h>.
63 *
64 * _mcount updates data structures that represent traversals of the
65 * program's call graph edges.  frompc and selfpc are the return
66 * address and function address that represents the given call graph edge.
67 *
68 * Note: the original BSD code used the same variable (frompcindex) for
69 * both frompcindex and frompc.  Any reasonable, modern compiler will
70 * perform this optimization.
71 */
72_MCOUNT_DECL(frompc, selfpc)	/* _mcount; may be static, inline, etc */
73	register uintfptr_t frompc, selfpc;
74{
75#ifdef GUPROF
76	u_int delta;
77#endif
78	register fptrdiff_t frompci;
79	register u_short *frompcindex;
80	register struct tostruct *top, *prevtop;
81	register struct gmonparam *p;
82	register long toindex;
83#ifdef KERNEL
84	MCOUNT_DECL(s)
85#endif
86
87	p = &_gmonparam;
88#ifndef GUPROF			/* XXX */
89	/*
90	 * check that we are profiling
91	 * and that we aren't recursively invoked.
92	 */
93	if (p->state != GMON_PROF_ON)
94		return;
95#endif
96#ifdef KERNEL
97	MCOUNT_ENTER(s);
98#else
99	p->state = GMON_PROF_BUSY;
100#endif
101	frompci = frompc - p->lowpc;
102
103#ifdef KERNEL
104	/*
105	 * When we are called from an exception handler, frompci may be
106	 * for a user address.  Convert such frompci's to the index of
107	 * user() to merge all user counts.
108	 */
109	if (frompci >= p->textsize) {
110		if (frompci + p->lowpc
111		    >= (uintfptr_t)(VM_MAXUSER_ADDRESS + UPAGES * PAGE_SIZE))
112			goto done;
113		frompci = (uintfptr_t)user - p->lowpc;
114		if (frompci >= p->textsize)
115		    goto done;
116	}
117#endif /* KERNEL */
118
119#ifdef GUPROF
120	if (p->state != GMON_PROF_HIRES)
121		goto skip_guprof_stuff;
122	/*
123	 * Look at the clock and add the count of clock cycles since the
124	 * clock was last looked at to a counter for frompc.  This
125	 * solidifies the count for the function containing frompc and
126	 * effectively starts another clock for the current function.
127	 * The count for the new clock will be solidified when another
128	 * function call is made or the function returns.
129	 *
130	 * We use the usual sampling counters since they can be located
131	 * efficiently.  4-byte counters are usually necessary.
132	 *
133	 * There are many complications for subtracting the profiling
134	 * overheads from the counts for normal functions and adding
135	 * them to the counts for mcount(), mexitcount() and cputime().
136	 * We attempt to handle fractional cycles, but the overheads
137	 * are usually underestimated because they are calibrated for
138	 * a simpler than usual setup.
139	 */
140	delta = cputime() - p->mcount_overhead;
141	p->cputime_overhead_resid += p->cputime_overhead_frac;
142	p->mcount_overhead_resid += p->mcount_overhead_frac;
143	if ((int)delta < 0)
144		*p->mcount_count += delta + p->mcount_overhead
145				    - p->cputime_overhead;
146	else if (delta != 0) {
147		if (p->cputime_overhead_resid >= CALIB_SCALE) {
148			p->cputime_overhead_resid -= CALIB_SCALE;
149			++*p->cputime_count;
150			--delta;
151		}
152		if (delta != 0) {
153			if (p->mcount_overhead_resid >= CALIB_SCALE) {
154				p->mcount_overhead_resid -= CALIB_SCALE;
155				++*p->mcount_count;
156				--delta;
157			}
158			KCOUNT(p, frompci) += delta;
159		}
160		*p->mcount_count += p->mcount_overhead_sub;
161	}
162	*p->cputime_count += p->cputime_overhead;
163skip_guprof_stuff:
164#endif /* GUPROF */
165
166#ifdef KERNEL
167	/*
168	 * When we are called from an exception handler, frompc is faked
169	 * to be for where the exception occurred.  We've just solidified
170	 * the count for there.  Now convert frompci to the index of btrap()
171	 * for trap handlers and bintr() for interrupt handlers to make
172	 * exceptions appear in the call graph as calls from btrap() and
173	 * bintr() instead of calls from all over.
174	 */
175	if ((uintfptr_t)selfpc >= (uintfptr_t)btrap
176	    && (uintfptr_t)selfpc < (uintfptr_t)eintr) {
177		if ((uintfptr_t)selfpc >= (uintfptr_t)bintr)
178			frompci = (uintfptr_t)bintr - p->lowpc;
179		else
180			frompci = (uintfptr_t)btrap - p->lowpc;
181	}
182#endif /* KERNEL */
183
184	/*
185	 * check that frompc is a reasonable pc value.
186	 * for example:	signal catchers get called from the stack,
187	 *		not from text space.  too bad.
188	 */
189	if (frompci >= p->textsize)
190		goto done;
191
192	frompcindex = &p->froms[frompci / (p->hashfraction * sizeof(*p->froms))];
193	toindex = *frompcindex;
194	if (toindex == 0) {
195		/*
196		 *	first time traversing this arc
197		 */
198		toindex = ++p->tos[0].link;
199		if (toindex >= p->tolimit)
200			/* halt further profiling */
201			goto overflow;
202
203		*frompcindex = toindex;
204		top = &p->tos[toindex];
205		top->selfpc = selfpc;
206		top->count = 1;
207		top->link = 0;
208		goto done;
209	}
210	top = &p->tos[toindex];
211	if (top->selfpc == selfpc) {
212		/*
213		 * arc at front of chain; usual case.
214		 */
215		top->count++;
216		goto done;
217	}
218	/*
219	 * have to go looking down chain for it.
220	 * top points to what we are looking at,
221	 * prevtop points to previous top.
222	 * we know it is not at the head of the chain.
223	 */
224	for (; /* goto done */; ) {
225		if (top->link == 0) {
226			/*
227			 * top is end of the chain and none of the chain
228			 * had top->selfpc == selfpc.
229			 * so we allocate a new tostruct
230			 * and link it to the head of the chain.
231			 */
232			toindex = ++p->tos[0].link;
233			if (toindex >= p->tolimit)
234				goto overflow;
235
236			top = &p->tos[toindex];
237			top->selfpc = selfpc;
238			top->count = 1;
239			top->link = *frompcindex;
240			*frompcindex = toindex;
241			goto done;
242		}
243		/*
244		 * otherwise, check the next arc on the chain.
245		 */
246		prevtop = top;
247		top = &p->tos[top->link];
248		if (top->selfpc == selfpc) {
249			/*
250			 * there it is.
251			 * increment its count
252			 * move it to the head of the chain.
253			 */
254			top->count++;
255			toindex = prevtop->link;
256			prevtop->link = top->link;
257			top->link = *frompcindex;
258			*frompcindex = toindex;
259			goto done;
260		}
261
262	}
263done:
264#ifdef KERNEL
265	MCOUNT_EXIT(s);
266#else
267	p->state = GMON_PROF_ON;
268#endif
269	return;
270overflow:
271	p->state = GMON_PROF_ERROR;
272#ifdef KERNEL
273	MCOUNT_EXIT(s);
274#endif
275	return;
276}
277
278/*
279 * Actual definition of mcount function.  Defined in <machine/profile.h>,
280 * which is included by <sys/gmon.h>.
281 */
282MCOUNT
283
284#ifdef GUPROF
285void
286mexitcount(selfpc)
287	uintfptr_t selfpc;
288{
289	struct gmonparam *p;
290	uintfptr_t selfpcdiff;
291
292	p = &_gmonparam;
293	selfpcdiff = selfpc - (uintfptr_t)p->lowpc;
294	if (selfpcdiff < p->textsize) {
295		u_int delta;
296
297		/*
298		 * Solidify the count for the current function.
299		 */
300		delta = cputime() - p->mexitcount_overhead;
301		p->cputime_overhead_resid += p->cputime_overhead_frac;
302		p->mexitcount_overhead_resid += p->mexitcount_overhead_frac;
303		if ((int)delta < 0)
304			*p->mexitcount_count += delta + p->mexitcount_overhead
305						- p->cputime_overhead;
306		else if (delta != 0) {
307			if (p->cputime_overhead_resid >= CALIB_SCALE) {
308				p->cputime_overhead_resid -= CALIB_SCALE;
309				++*p->cputime_count;
310				--delta;
311			}
312			if (delta != 0) {
313				if (p->mexitcount_overhead_resid
314				    >= CALIB_SCALE) {
315					p->mexitcount_overhead_resid
316					    -= CALIB_SCALE;
317					++*p->mexitcount_count;
318					--delta;
319				}
320				KCOUNT(p, selfpcdiff) += delta;
321			}
322			*p->mexitcount_count += p->mexitcount_overhead_sub;
323		}
324		*p->cputime_count += p->cputime_overhead;
325	}
326}
327#endif /* GUPROF */
328
329#endif /* !__NETBSD_SYSCALLS */
330