xref: /freebsd/sys/kern/subr_prof.c (revision 48991a368427cadb9cdac39581d1676c29619c52)
1 /*-
2  * Copyright (c) 1982, 1986, 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  *	@(#)subr_prof.c	8.3 (Berkeley) 9/23/93
34  * $Id: subr_prof.c,v 1.10 1995/11/12 06:43:04 bde Exp $
35  */
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/sysproto.h>
40 #include <sys/kernel.h>
41 #include <sys/proc.h>
42 #include <vm/vm.h>
43 #include <sys/sysctl.h>
44 #include <sys/user.h>
45 
46 #include <machine/cpu.h>
47 
48 #ifdef GPROF
49 #include <sys/malloc.h>
50 #include <sys/gmon.h>
51 
52 static void kmstartup __P((void *));
53 SYSINIT(kmem, SI_SUB_KPROF, SI_ORDER_FIRST, kmstartup, NULL)
54 
55 struct gmonparam _gmonparam = { GMON_PROF_OFF };
56 
57 extern char btext[];
58 extern char etext[];
59 
60 static void
61 kmstartup(udata)
62 	void *udata;
63 {
64 	char *cp;
65 	struct gmonparam *p = &_gmonparam;
66 	/*
67 	 * Round lowpc and highpc to multiples of the density we're using
68 	 * so the rest of the scaling (here and in gprof) stays in ints.
69 	 */
70 	p->lowpc = ROUNDDOWN((u_long)btext, HISTFRACTION * sizeof(HISTCOUNTER));
71 	p->highpc = ROUNDUP((u_long)etext, HISTFRACTION * sizeof(HISTCOUNTER));
72 	p->textsize = p->highpc - p->lowpc;
73 	printf("Profiling kernel, textsize=%d [%x..%x]\n",
74 	       p->textsize, p->lowpc, p->highpc);
75 	p->kcountsize = p->textsize / HISTFRACTION;
76 	p->hashfraction = HASHFRACTION;
77 	p->fromssize = p->textsize / HASHFRACTION;
78 	p->tolimit = p->textsize * ARCDENSITY / 100;
79 	if (p->tolimit < MINARCS)
80 		p->tolimit = MINARCS;
81 	else if (p->tolimit > MAXARCS)
82 		p->tolimit = MAXARCS;
83 	p->tossize = p->tolimit * sizeof(struct tostruct);
84 	cp = (char *)malloc(p->kcountsize + p->fromssize + p->tossize,
85 	    M_GPROF, M_NOWAIT);
86 	if (cp == 0) {
87 		printf("No memory for profiling.\n");
88 		return;
89 	}
90 	bzero(cp, p->kcountsize + p->tossize + p->fromssize);
91 	p->tos = (struct tostruct *)cp;
92 	cp += p->tossize;
93 	p->kcount = (u_short *)cp;
94 	cp += p->kcountsize;
95 	p->froms = (u_short *)cp;
96 }
97 
98 /*
99  * Return kernel profiling information.
100  */
101 static int
102 sysctl_kern_prof SYSCTL_HANDLER_ARGS
103 {
104 	int *name = (int *) arg1;
105 	u_int namelen = arg2;
106 	struct gmonparam *gp = &_gmonparam;
107 	int error;
108 
109 	/* all sysctl names at this level are terminal */
110 	if (namelen != 1)
111 		return (ENOTDIR);		/* overloaded */
112 
113 	switch (name[0]) {
114 	case GPROF_STATE:
115 		error = sysctl_handle_int(oidp, &gp->state, 0, req);
116 		if (error)
117 			return (error);
118 		if (gp->state == GMON_PROF_OFF)
119 			stopprofclock(&proc0);
120 		else
121 			startprofclock(&proc0);
122 		return (0);
123 	case GPROF_COUNT:
124 		return (sysctl_handle_opaque(oidp,
125 			gp->kcount, gp->kcountsize, req));
126 	case GPROF_FROMS:
127 		return (sysctl_handle_opaque(oidp,
128 			gp->froms, gp->fromssize, req));
129 	case GPROF_TOS:
130 		return (sysctl_handle_opaque(oidp,
131 			gp->tos, gp->tossize, req));
132 	case GPROF_GMONPARAM:
133 		return (sysctl_handle_opaque(oidp, gp, sizeof *gp, req));
134 	default:
135 		return (EOPNOTSUPP);
136 	}
137 	/* NOTREACHED */
138 }
139 
140 SYSCTL_NODE(_kern, KERN_PROF, prof, CTLFLAG_RW, sysctl_kern_prof, "");
141 #endif /* GPROF */
142 
143 /*
144  * Profiling system call.
145  *
146  * The scale factor is a fixed point number with 16 bits of fraction, so that
147  * 1.0 is represented as 0x10000.  A scale factor of 0 turns off profiling.
148  */
149 #ifndef _SYS_SYSPROTO_H_
150 struct profil_args {
151 	caddr_t	samples;
152 	u_int	size;
153 	u_int	offset;
154 	u_int	scale;
155 };
156 #endif
157 /* ARGSUSED */
158 int
159 profil(p, uap, retval)
160 	struct proc *p;
161 	register struct profil_args *uap;
162 	int *retval;
163 {
164 	register struct uprof *upp;
165 	int s;
166 
167 	if (uap->scale > (1 << 16))
168 		return (EINVAL);
169 	if (uap->scale == 0) {
170 		stopprofclock(p);
171 		return (0);
172 	}
173 	upp = &p->p_stats->p_prof;
174 
175 	/* Block profile interrupts while changing state. */
176 	s = splstatclock();
177 	upp->pr_off = uap->offset;
178 	upp->pr_scale = uap->scale;
179 	upp->pr_base = uap->samples;
180 	upp->pr_size = uap->size;
181 	startprofclock(p);
182 	splx(s);
183 
184 	return (0);
185 }
186 
187 /*
188  * Scale is a fixed-point number with the binary point 16 bits
189  * into the value, and is <= 1.0.  pc is at most 32 bits, so the
190  * intermediate result is at most 48 bits.
191  */
192 #define	PC_TO_INDEX(pc, prof) \
193 	((int)(((u_quad_t)((pc) - (prof)->pr_off) * \
194 	    (u_quad_t)((prof)->pr_scale)) >> 16) & ~1)
195 
196 /*
197  * Collect user-level profiling statistics; called on a profiling tick,
198  * when a process is running in user-mode.  This routine may be called
199  * from an interrupt context.  We try to update the user profiling buffers
200  * cheaply with fuswintr() and suswintr().  If that fails, we revert to
201  * an AST that will vector us to trap() with a context in which copyin
202  * and copyout will work.  Trap will then call addupc_task().
203  *
204  * Note that we may (rarely) not get around to the AST soon enough, and
205  * lose profile ticks when the next tick overwrites this one, but in this
206  * case the system is overloaded and the profile is probably already
207  * inaccurate.
208  */
209 void
210 addupc_intr(p, pc, ticks)
211 	register struct proc *p;
212 	register u_long pc;
213 	u_int ticks;
214 {
215 	register struct uprof *prof;
216 	register caddr_t addr;
217 	register u_int i;
218 	register int v;
219 
220 	if (ticks == 0)
221 		return;
222 	prof = &p->p_stats->p_prof;
223 	if (pc < prof->pr_off ||
224 	    (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size)
225 		return;			/* out of range; ignore */
226 
227 	addr = prof->pr_base + i;
228 	if ((v = fuswintr(addr)) == -1 || suswintr(addr, v + ticks) == -1) {
229 		prof->pr_addr = pc;
230 		prof->pr_ticks = ticks;
231 		need_proftick(p);
232 	}
233 }
234 
235 /*
236  * Much like before, but we can afford to take faults here.  If the
237  * update fails, we simply turn off profiling.
238  */
239 void
240 addupc_task(p, pc, ticks)
241 	register struct proc *p;
242 	register u_long pc;
243 	u_int ticks;
244 {
245 	register struct uprof *prof;
246 	register caddr_t addr;
247 	register u_int i;
248 	u_short v;
249 
250 	/* Testing P_PROFIL may be unnecessary, but is certainly safe. */
251 	if ((p->p_flag & P_PROFIL) == 0 || ticks == 0)
252 		return;
253 
254 	prof = &p->p_stats->p_prof;
255 	if (pc < prof->pr_off ||
256 	    (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size)
257 		return;
258 
259 	addr = prof->pr_base + i;
260 	if (copyin(addr, (caddr_t)&v, sizeof(v)) == 0) {
261 		v += ticks;
262 		if (copyout((caddr_t)&v, addr, sizeof(v)) == 0)
263 			return;
264 	}
265 	stopprofclock(p);
266 }
267