xref: /netbsd-src/sys/kern/subr_prof.c (revision d0fed6c87ddc40a8bffa6f99e7433ddfc864dd83)
1 /*	$NetBSD: subr_prof.c,v 1.15 1996/12/18 20:12:58 gwr Exp $	*/
2 
3 /*-
4  * Copyright (c) 1982, 1986, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *	This product includes software developed by the University of
18  *	California, Berkeley and its contributors.
19  * 4. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)subr_prof.c	8.3 (Berkeley) 9/23/93
36  */
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/kernel.h>
41 #include <sys/proc.h>
42 #include <sys/user.h>
43 #include <sys/mount.h>
44 #include <sys/syscallargs.h>
45 
46 #include <machine/cpu.h>
47 
48 #ifdef GPROF
49 #include <sys/malloc.h>
50 #include <sys/gmon.h>
51 
52 /*
53  * Froms is actually a bunch of unsigned shorts indexing tos
54  */
55 struct gmonparam _gmonparam = { GMON_PROF_OFF };
56 
57 /* Actual start of the kernel text segment. */
58 extern char kernel_text[];
59 
60 extern char etext[];
61 
62 
63 void
64 kmstartup()
65 {
66 	char *cp;
67 	struct gmonparam *p = &_gmonparam;
68 	/*
69 	 * Round lowpc and highpc to multiples of the density we're using
70 	 * so the rest of the scaling (here and in gprof) stays in ints.
71 	 */
72 	p->lowpc = ROUNDDOWN(((u_long)kernel_text),
73 		HISTFRACTION * sizeof(HISTCOUNTER));
74 	p->highpc = ROUNDUP((u_long)etext,
75 		HISTFRACTION * sizeof(HISTCOUNTER));
76 	p->textsize = p->highpc - p->lowpc;
77 	printf("Profiling kernel, textsize=%ld [%lx..%lx]\n",
78 	       p->textsize, p->lowpc, p->highpc);
79 	p->kcountsize = p->textsize / HISTFRACTION;
80 	p->hashfraction = HASHFRACTION;
81 	p->fromssize = p->textsize / HASHFRACTION;
82 	p->tolimit = p->textsize * ARCDENSITY / 100;
83 	if (p->tolimit < MINARCS)
84 		p->tolimit = MINARCS;
85 	else if (p->tolimit > MAXARCS)
86 		p->tolimit = MAXARCS;
87 	p->tossize = p->tolimit * sizeof(struct tostruct);
88 	cp = (char *)malloc(p->kcountsize + p->fromssize + p->tossize,
89 	    M_GPROF, M_NOWAIT);
90 	if (cp == 0) {
91 		printf("No memory for profiling.\n");
92 		return;
93 	}
94 	bzero(cp, p->kcountsize + p->tossize + p->fromssize);
95 	p->tos = (struct tostruct *)cp;
96 	cp += p->tossize;
97 	p->kcount = (u_short *)cp;
98 	cp += p->kcountsize;
99 	p->froms = (u_short *)cp;
100 }
101 
102 /*
103  * Return kernel profiling information.
104  */
105 int
106 sysctl_doprof(name, namelen, oldp, oldlenp, newp, newlen)
107 	int *name;
108 	u_int namelen;
109 	void *oldp;
110 	size_t *oldlenp;
111 	void *newp;
112 	size_t newlen;
113 {
114 	struct gmonparam *gp = &_gmonparam;
115 	int error;
116 
117 	/* all sysctl names at this level are terminal */
118 	if (namelen != 1)
119 		return (ENOTDIR);		/* overloaded */
120 
121 	switch (name[0]) {
122 	case GPROF_STATE:
123 		error = sysctl_int(oldp, oldlenp, newp, newlen, &gp->state);
124 		if (error)
125 			return (error);
126 		if (gp->state == GMON_PROF_OFF)
127 			stopprofclock(&proc0);
128 		else
129 			startprofclock(&proc0);
130 		return (0);
131 	case GPROF_COUNT:
132 		return (sysctl_struct(oldp, oldlenp, newp, newlen,
133 		    gp->kcount, gp->kcountsize));
134 	case GPROF_FROMS:
135 		return (sysctl_struct(oldp, oldlenp, newp, newlen,
136 		    gp->froms, gp->fromssize));
137 	case GPROF_TOS:
138 		return (sysctl_struct(oldp, oldlenp, newp, newlen,
139 		    gp->tos, gp->tossize));
140 	case GPROF_GMONPARAM:
141 		return (sysctl_rdstruct(oldp, oldlenp, newp, gp, sizeof *gp));
142 	default:
143 		return (EOPNOTSUPP);
144 	}
145 	/* NOTREACHED */
146 }
147 #endif /* GPROF */
148 
149 /*
150  * Profiling system call.
151  *
152  * The scale factor is a fixed point number with 16 bits of fraction, so that
153  * 1.0 is represented as 0x10000.  A scale factor of 0 turns off profiling.
154  */
155 /* ARGSUSED */
156 int
157 sys_profil(p, v, retval)
158 	struct proc *p;
159 	void *v;
160 	register_t *retval;
161 {
162 	register struct sys_profil_args /* {
163 		syscallarg(caddr_t) samples;
164 		syscallarg(u_int) size;
165 		syscallarg(u_int) offset;
166 		syscallarg(u_int) scale;
167 	} */ *uap = v;
168 	register struct uprof *upp;
169 	int s;
170 
171 	if (SCARG(uap, scale) > (1 << 16))
172 		return (EINVAL);
173 	if (SCARG(uap, scale) == 0) {
174 		stopprofclock(p);
175 		return (0);
176 	}
177 	upp = &p->p_stats->p_prof;
178 
179 	/* Block profile interrupts while changing state. */
180 	s = splstatclock();
181 	upp->pr_off = SCARG(uap, offset);
182 	upp->pr_scale = SCARG(uap, scale);
183 	upp->pr_base = SCARG(uap, samples);
184 	upp->pr_size = SCARG(uap, size);
185 	startprofclock(p);
186 	splx(s);
187 
188 	return (0);
189 }
190 
191 /*
192  * Scale is a fixed-point number with the binary point 16 bits
193  * into the value, and is <= 1.0.  pc is at most 32 bits, so the
194  * intermediate result is at most 48 bits.
195  */
196 #define	PC_TO_INDEX(pc, prof) \
197 	((int)(((u_quad_t)((pc) - (prof)->pr_off) * \
198 	    (u_quad_t)((prof)->pr_scale)) >> 16) & ~1)
199 
200 /*
201  * Collect user-level profiling statistics; called on a profiling tick,
202  * when a process is running in user-mode.  This routine may be called
203  * from an interrupt context.  We try to update the user profiling buffers
204  * cheaply with fuswintr() and suswintr().  If that fails, we revert to
205  * an AST that will vector us to trap() with a context in which copyin
206  * and copyout will work.  Trap will then call addupc_task().
207  *
208  * Note that we may (rarely) not get around to the AST soon enough, and
209  * lose profile ticks when the next tick overwrites this one, but in this
210  * case the system is overloaded and the profile is probably already
211  * inaccurate.
212  */
213 void
214 addupc_intr(p, pc, ticks)
215 	register struct proc *p;
216 	register u_long pc;
217 	u_int ticks;
218 {
219 	register struct uprof *prof;
220 	register caddr_t addr;
221 	register u_int i;
222 	register int v;
223 
224 	if (ticks == 0)
225 		return;
226 	prof = &p->p_stats->p_prof;
227 	if (pc < prof->pr_off ||
228 	    (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size)
229 		return;			/* out of range; ignore */
230 
231 	addr = prof->pr_base + i;
232 	if ((v = fuswintr(addr)) == -1 || suswintr(addr, v + ticks) == -1) {
233 		prof->pr_addr = pc;
234 		prof->pr_ticks = ticks;
235 		need_proftick(p);
236 	}
237 }
238 
239 /*
240  * Much like before, but we can afford to take faults here.  If the
241  * update fails, we simply turn off profiling.
242  */
243 void
244 addupc_task(p, pc, ticks)
245 	register struct proc *p;
246 	register u_long pc;
247 	u_int ticks;
248 {
249 	register struct uprof *prof;
250 	register caddr_t addr;
251 	register u_int i;
252 	u_short v;
253 
254 	/* Testing P_PROFIL may be unnecessary, but is certainly safe. */
255 	if ((p->p_flag & P_PROFIL) == 0 || ticks == 0)
256 		return;
257 
258 	prof = &p->p_stats->p_prof;
259 	if (pc < prof->pr_off ||
260 	    (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size)
261 		return;
262 
263 	addr = prof->pr_base + i;
264 	if (copyin(addr, (caddr_t)&v, sizeof(v)) == 0) {
265 		v += ticks;
266 		if (copyout((caddr_t)&v, addr, sizeof(v)) == 0)
267 			return;
268 	}
269 	stopprofclock(p);
270 }
271