xref: /netbsd-src/sys/kern/subr_prof.c (revision 5aefcfdc06931dd97e76246d2fe0302f7b3fe094)
1 /*	$NetBSD: subr_prof.c,v 1.22 2000/12/10 19:29:31 mycroft 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	8.4 (Berkeley) 2/14/95
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 #include <uvm/uvm_extern.h>
46 #include <sys/sysctl.h>
47 
48 #include <machine/cpu.h>
49 
50 #ifdef GPROF
51 #include <sys/malloc.h>
52 #include <sys/gmon.h>
53 
54 /*
55  * Froms is actually a bunch of unsigned shorts indexing tos
56  */
57 struct gmonparam _gmonparam = { GMON_PROF_OFF };
58 
59 /* Actual start of the kernel text segment. */
60 extern char kernel_text[];
61 
62 extern char etext[];
63 
64 
65 void
66 kmstartup()
67 {
68 	char *cp;
69 	struct gmonparam *p = &_gmonparam;
70 	/*
71 	 * Round lowpc and highpc to multiples of the density we're using
72 	 * so the rest of the scaling (here and in gprof) stays in ints.
73 	 */
74 	p->lowpc = ROUNDDOWN(((u_long)kernel_text),
75 		HISTFRACTION * sizeof(HISTCOUNTER));
76 	p->highpc = ROUNDUP((u_long)etext,
77 		HISTFRACTION * sizeof(HISTCOUNTER));
78 	p->textsize = p->highpc - p->lowpc;
79 	printf("Profiling kernel, textsize=%ld [%lx..%lx]\n",
80 	       p->textsize, p->lowpc, p->highpc);
81 	p->kcountsize = p->textsize / HISTFRACTION;
82 	p->hashfraction = HASHFRACTION;
83 	p->fromssize = p->textsize / HASHFRACTION;
84 	p->tolimit = p->textsize * ARCDENSITY / 100;
85 	if (p->tolimit < MINARCS)
86 		p->tolimit = MINARCS;
87 	else if (p->tolimit > MAXARCS)
88 		p->tolimit = MAXARCS;
89 	p->tossize = p->tolimit * sizeof(struct tostruct);
90 	cp = (char *)malloc(p->kcountsize + p->fromssize + p->tossize,
91 	    M_GPROF, M_NOWAIT);
92 	if (cp == 0) {
93 		printf("No memory for profiling.\n");
94 		return;
95 	}
96 	memset(cp, 0, p->kcountsize + p->tossize + p->fromssize);
97 	p->tos = (struct tostruct *)cp;
98 	cp += p->tossize;
99 	p->kcount = (u_short *)cp;
100 	cp += p->kcountsize;
101 	p->froms = (u_short *)cp;
102 }
103 
104 /*
105  * Return kernel profiling information.
106  */
107 int
108 sysctl_doprof(name, namelen, oldp, oldlenp, newp, newlen)
109 	int *name;
110 	u_int namelen;
111 	void *oldp;
112 	size_t *oldlenp;
113 	void *newp;
114 	size_t newlen;
115 {
116 	struct gmonparam *gp = &_gmonparam;
117 	int error;
118 
119 	/* all sysctl names at this level are terminal */
120 	if (namelen != 1)
121 		return (ENOTDIR);		/* overloaded */
122 
123 	switch (name[0]) {
124 	case GPROF_STATE:
125 		error = sysctl_int(oldp, oldlenp, newp, newlen, &gp->state);
126 		if (error)
127 			return (error);
128 		if (gp->state == GMON_PROF_OFF)
129 			stopprofclock(&proc0);
130 		else
131 			startprofclock(&proc0);
132 		return (0);
133 	case GPROF_COUNT:
134 		return (sysctl_struct(oldp, oldlenp, newp, newlen,
135 		    gp->kcount, gp->kcountsize));
136 	case GPROF_FROMS:
137 		return (sysctl_struct(oldp, oldlenp, newp, newlen,
138 		    gp->froms, gp->fromssize));
139 	case GPROF_TOS:
140 		return (sysctl_struct(oldp, oldlenp, newp, newlen,
141 		    gp->tos, gp->tossize));
142 	case GPROF_GMONPARAM:
143 		return (sysctl_rdstruct(oldp, oldlenp, newp, gp, sizeof(*gp)));
144 	default:
145 		return (EOPNOTSUPP);
146 	}
147 	/* NOTREACHED */
148 }
149 #endif /* GPROF */
150 
151 /*
152  * Profiling system call.
153  *
154  * The scale factor is a fixed point number with 16 bits of fraction, so that
155  * 1.0 is represented as 0x10000.  A scale factor of 0 turns off profiling.
156  */
157 /* ARGSUSED */
158 int
159 sys_profil(p, v, retval)
160 	struct proc *p;
161 	void *v;
162 	register_t *retval;
163 {
164 	struct sys_profil_args /* {
165 		syscallarg(caddr_t) samples;
166 		syscallarg(u_int) size;
167 		syscallarg(u_int) offset;
168 		syscallarg(u_int) scale;
169 	} */ *uap = v;
170 	struct uprof *upp;
171 	int s;
172 
173 	if (SCARG(uap, scale) > (1 << 16))
174 		return (EINVAL);
175 	if (SCARG(uap, scale) == 0) {
176 		stopprofclock(p);
177 		return (0);
178 	}
179 	upp = &p->p_stats->p_prof;
180 
181 	/* Block profile interrupts while changing state. */
182 	s = splstatclock();
183 	upp->pr_off = SCARG(uap, offset);
184 	upp->pr_scale = SCARG(uap, scale);
185 	upp->pr_base = SCARG(uap, samples);
186 	upp->pr_size = SCARG(uap, size);
187 	startprofclock(p);
188 	splx(s);
189 
190 	return (0);
191 }
192 
193 /*
194  * Scale is a fixed-point number with the binary point 16 bits
195  * into the value, and is <= 1.0.  pc is at most 32 bits, so the
196  * intermediate result is at most 48 bits.
197  */
198 #define	PC_TO_INDEX(pc, prof) \
199 	((int)(((u_quad_t)((pc) - (prof)->pr_off) * \
200 	    (u_quad_t)((prof)->pr_scale)) >> 16) & ~1)
201 
202 /*
203  * Collect user-level profiling statistics; called on a profiling tick,
204  * when a process is running in user-mode.  This routine may be called
205  * from an interrupt context.  We try to update the user profiling buffers
206  * cheaply with fuswintr() and suswintr().  If that fails, we revert to
207  * an AST that will vector us to trap() with a context in which copyin
208  * and copyout will work.  Trap will then call addupc_task().
209  *
210  * Note that we may (rarely) not get around to the AST soon enough, and
211  * lose profile ticks when the next tick overwrites this one, but in this
212  * case the system is overloaded and the profile is probably already
213  * inaccurate.
214  */
215 void
216 addupc_intr(p, pc)
217 	struct proc *p;
218 	u_long pc;
219 {
220 	struct uprof *prof;
221 	caddr_t addr;
222 	u_int i;
223 	int v;
224 
225 	prof = &p->p_stats->p_prof;
226 	if (pc < prof->pr_off ||
227 	    (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size)
228 		return;			/* out of range; ignore */
229 
230 	addr = prof->pr_base + i;
231 	if ((v = fuswintr(addr)) == -1 || suswintr(addr, v + 1) == -1) {
232 		prof->pr_addr = pc;
233 		prof->pr_ticks++;
234 		need_proftick(p);
235 	}
236 }
237 
238 /*
239  * Much like before, but we can afford to take faults here.  If the
240  * update fails, we simply turn off profiling.
241  */
242 void
243 addupc_task(p, pc, ticks)
244 	struct proc *p;
245 	u_long pc;
246 	u_int ticks;
247 {
248 	struct uprof *prof;
249 	caddr_t addr;
250 	u_int i;
251 	u_short v;
252 
253 	/* Testing P_PROFIL may be unnecessary, but is certainly safe. */
254 	if ((p->p_flag & P_PROFIL) == 0 || ticks == 0)
255 		return;
256 
257 	prof = &p->p_stats->p_prof;
258 	if (pc < prof->pr_off ||
259 	    (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size)
260 		return;
261 
262 	addr = prof->pr_base + i;
263 	if (copyin(addr, (caddr_t)&v, sizeof(v)) == 0) {
264 		v += ticks;
265 		if (copyout((caddr_t)&v, addr, sizeof(v)) == 0)
266 			return;
267 	}
268 	stopprofclock(p);
269 }
270