xref: /freebsd/sys/kern/subr_prof.c (revision 8e6b01171e30297084bb0b4457c4183c2746aacc)
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 1.8 1995/08/29 03:09:05 bde Exp $
35  */
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
40 #include <sys/proc.h>
41 #include <vm/vm.h>
42 #include <sys/sysctl.h>
43 #include <sys/user.h>
44 
45 #include <machine/cpu.h>
46 
47 #ifdef GPROF
48 #include <sys/malloc.h>
49 #include <sys/gmon.h>
50 
51 static void kmstartup __P((void *));
52 SYSINIT(kmem, SI_SUB_KPROF, SI_ORDER_FIRST, kmstartup, NULL)
53 
54 struct gmonparam _gmonparam = { GMON_PROF_OFF };
55 
56 extern char btext[];
57 extern char etext[];
58 
59 static void
60 kmstartup(udata)
61 	void *udata;
62 {
63 	char *cp;
64 	struct gmonparam *p = &_gmonparam;
65 	/*
66 	 * Round lowpc and highpc to multiples of the density we're using
67 	 * so the rest of the scaling (here and in gprof) stays in ints.
68 	 */
69 	p->lowpc = ROUNDDOWN((u_long)btext, HISTFRACTION * sizeof(HISTCOUNTER));
70 	p->highpc = ROUNDUP((u_long)etext, HISTFRACTION * sizeof(HISTCOUNTER));
71 	p->textsize = p->highpc - p->lowpc;
72 	printf("Profiling kernel, textsize=%d [%x..%x]\n",
73 	       p->textsize, p->lowpc, p->highpc);
74 	p->kcountsize = p->textsize / HISTFRACTION;
75 	p->hashfraction = HASHFRACTION;
76 	p->fromssize = p->textsize / HASHFRACTION;
77 	p->tolimit = p->textsize * ARCDENSITY / 100;
78 	if (p->tolimit < MINARCS)
79 		p->tolimit = MINARCS;
80 	else if (p->tolimit > MAXARCS)
81 		p->tolimit = MAXARCS;
82 	p->tossize = p->tolimit * sizeof(struct tostruct);
83 	cp = (char *)malloc(p->kcountsize + p->fromssize + p->tossize,
84 	    M_GPROF, M_NOWAIT);
85 	if (cp == 0) {
86 		printf("No memory for profiling.\n");
87 		return;
88 	}
89 	bzero(cp, p->kcountsize + p->tossize + p->fromssize);
90 	p->tos = (struct tostruct *)cp;
91 	cp += p->tossize;
92 	p->kcount = (u_short *)cp;
93 	cp += p->kcountsize;
94 	p->froms = (u_short *)cp;
95 }
96 
97 /*
98  * Return kernel profiling information.
99  */
100 int
101 sysctl_doprof(name, namelen, oldp, oldlenp, newp, newlen)
102 	int *name;
103 	u_int namelen;
104 	void *oldp;
105 	size_t *oldlenp;
106 	void *newp;
107 	size_t newlen;
108 {
109 	struct gmonparam *gp = &_gmonparam;
110 	int error;
111 
112 	/* all sysctl names at this level are terminal */
113 	if (namelen != 1)
114 		return (ENOTDIR);		/* overloaded */
115 
116 	switch (name[0]) {
117 	case GPROF_STATE:
118 		error = sysctl_int(oldp, oldlenp, newp, newlen, &gp->state);
119 		if (error)
120 			return (error);
121 		if (gp->state == GMON_PROF_OFF)
122 			stopprofclock(&proc0);
123 		else
124 			startprofclock(&proc0);
125 		return (0);
126 	case GPROF_COUNT:
127 		return (sysctl_struct(oldp, oldlenp, newp, newlen,
128 		    gp->kcount, gp->kcountsize));
129 	case GPROF_FROMS:
130 		return (sysctl_struct(oldp, oldlenp, newp, newlen,
131 		    gp->froms, gp->fromssize));
132 	case GPROF_TOS:
133 		return (sysctl_struct(oldp, oldlenp, newp, newlen,
134 		    gp->tos, gp->tossize));
135 	case GPROF_GMONPARAM:
136 		return (sysctl_rdstruct(oldp, oldlenp, newp, gp, sizeof *gp));
137 	default:
138 		return (EOPNOTSUPP);
139 	}
140 	/* NOTREACHED */
141 }
142 #endif /* GPROF */
143 
144 /*
145  * Profiling system call.
146  *
147  * The scale factor is a fixed point number with 16 bits of fraction, so that
148  * 1.0 is represented as 0x10000.  A scale factor of 0 turns off profiling.
149  */
150 struct profil_args {
151 	caddr_t	samples;
152 	u_int	size;
153 	u_int	offset;
154 	u_int	scale;
155 };
156 /* ARGSUSED */
157 int
158 profil(p, uap, retval)
159 	struct proc *p;
160 	register struct profil_args *uap;
161 	int *retval;
162 {
163 	register struct uprof *upp;
164 	int s;
165 
166 	if (uap->scale > (1 << 16))
167 		return (EINVAL);
168 	if (uap->scale == 0) {
169 		stopprofclock(p);
170 		return (0);
171 	}
172 	upp = &p->p_stats->p_prof;
173 
174 	/* Block profile interrupts while changing state. */
175 	s = splstatclock();
176 	upp->pr_off = uap->offset;
177 	upp->pr_scale = uap->scale;
178 	upp->pr_base = uap->samples;
179 	upp->pr_size = uap->size;
180 	startprofclock(p);
181 	splx(s);
182 
183 	return (0);
184 }
185 
186 /*
187  * Scale is a fixed-point number with the binary point 16 bits
188  * into the value, and is <= 1.0.  pc is at most 32 bits, so the
189  * intermediate result is at most 48 bits.
190  */
191 #define	PC_TO_INDEX(pc, prof) \
192 	((int)(((u_quad_t)((pc) - (prof)->pr_off) * \
193 	    (u_quad_t)((prof)->pr_scale)) >> 16) & ~1)
194 
195 /*
196  * Collect user-level profiling statistics; called on a profiling tick,
197  * when a process is running in user-mode.  This routine may be called
198  * from an interrupt context.  We try to update the user profiling buffers
199  * cheaply with fuswintr() and suswintr().  If that fails, we revert to
200  * an AST that will vector us to trap() with a context in which copyin
201  * and copyout will work.  Trap will then call addupc_task().
202  *
203  * Note that we may (rarely) not get around to the AST soon enough, and
204  * lose profile ticks when the next tick overwrites this one, but in this
205  * case the system is overloaded and the profile is probably already
206  * inaccurate.
207  */
208 void
209 addupc_intr(p, pc, ticks)
210 	register struct proc *p;
211 	register u_long pc;
212 	u_int ticks;
213 {
214 	register struct uprof *prof;
215 	register caddr_t addr;
216 	register u_int i;
217 	register int v;
218 
219 	if (ticks == 0)
220 		return;
221 	prof = &p->p_stats->p_prof;
222 	if (pc < prof->pr_off ||
223 	    (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size)
224 		return;			/* out of range; ignore */
225 
226 	addr = prof->pr_base + i;
227 	if ((v = fuswintr(addr)) == -1 || suswintr(addr, v + ticks) == -1) {
228 		prof->pr_addr = pc;
229 		prof->pr_ticks = ticks;
230 		need_proftick(p);
231 	}
232 }
233 
234 /*
235  * Much like before, but we can afford to take faults here.  If the
236  * update fails, we simply turn off profiling.
237  */
238 void
239 addupc_task(p, pc, ticks)
240 	register struct proc *p;
241 	register u_long pc;
242 	u_int ticks;
243 {
244 	register struct uprof *prof;
245 	register caddr_t addr;
246 	register u_int i;
247 	u_short v;
248 
249 	/* Testing P_PROFIL may be unnecessary, but is certainly safe. */
250 	if ((p->p_flag & P_PROFIL) == 0 || ticks == 0)
251 		return;
252 
253 	prof = &p->p_stats->p_prof;
254 	if (pc < prof->pr_off ||
255 	    (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size)
256 		return;
257 
258 	addr = prof->pr_base + i;
259 	if (copyin(addr, (caddr_t)&v, sizeof(v)) == 0) {
260 		v += ticks;
261 		if (copyout((caddr_t)&v, addr, sizeof(v)) == 0)
262 			return;
263 	}
264 	stopprofclock(p);
265 }
266