subr_prof.c revision 1.47 
1/*	$NetBSD: subr_prof.c,v 1.47 2014/07/10 21:13:52 christos 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. Neither the name of the University nor the names of its contributors
16 *    may be used to endorse or promote products derived from this software
17 *    without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 *	@(#)subr_prof.c	8.4 (Berkeley) 2/14/95
32 */
33
34#include <sys/cdefs.h>
35__KERNEL_RCSID(0, "$NetBSD: subr_prof.c,v 1.47 2014/07/10 21:13:52 christos Exp $");
36
37#include <sys/param.h>
38#include <sys/systm.h>
39#include <sys/kernel.h>
40#include <sys/proc.h>
41#include <sys/mount.h>
42#include <sys/syscallargs.h>
43#include <sys/sysctl.h>
44
45#include <sys/cpu.h>
46
47#ifdef GPROF
48#include <sys/malloc.h>
49#include <sys/gmon.h>
50
51MALLOC_DEFINE(M_GPROF, "gprof", "kernel profiling buffer");
52
53/*
54 * Froms is actually a bunch of unsigned shorts indexing tos
55 */
56struct gmonparam _gmonparam = { .state = GMON_PROF_OFF };
57
58/* Actual start of the kernel text segment. */
59extern char kernel_text[];
60
61extern char etext[];
62
63
64void
65kmstartup(void)
66{
67	char *cp;
68	struct gmonparam *p = &_gmonparam;
69	/*
70	 * Round lowpc and highpc to multiples of the density we're using
71	 * so the rest of the scaling (here and in gprof) stays in ints.
72	 */
73	p->lowpc = rounddown(((u_long)kernel_text),
74		HISTFRACTION * sizeof(HISTCOUNTER));
75	p->highpc = roundup((u_long)etext,
76		HISTFRACTION * sizeof(HISTCOUNTER));
77	p->textsize = p->highpc - p->lowpc;
78	printf("Profiling kernel, textsize=%ld [%lx..%lx]\n",
79	       p->textsize, p->lowpc, p->highpc);
80	p->kcountsize = p->textsize / HISTFRACTION;
81	p->hashfraction = HASHFRACTION;
82	p->fromssize = p->textsize / HASHFRACTION;
83	p->tolimit = p->textsize * ARCDENSITY / 100;
84	if (p->tolimit < MINARCS)
85		p->tolimit = MINARCS;
86	else if (p->tolimit > MAXARCS)
87		p->tolimit = MAXARCS;
88	p->tossize = p->tolimit * sizeof(struct tostruct);
89	cp = malloc(p->kcountsize + p->fromssize + p->tossize,
90	    M_GPROF, M_NOWAIT | M_ZERO);
91	if (cp == 0) {
92		printf("No memory for profiling.\n");
93		return;
94	}
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/*
106 * sysctl helper routine for kern.profiling subtree.  enables/disables
107 * kernel profiling and gives out copies of the profiling data.
108 */
109static int
110sysctl_kern_profiling(SYSCTLFN_ARGS)
111{
112	struct gmonparam *gp = &_gmonparam;
113	int error;
114	struct sysctlnode node;
115
116	node = *rnode;
117
118	switch (node.sysctl_num) {
119	case GPROF_STATE:
120		node.sysctl_data = &gp->state;
121		break;
122	case GPROF_COUNT:
123		node.sysctl_data = gp->kcount;
124		node.sysctl_size = gp->kcountsize;
125		break;
126	case GPROF_FROMS:
127		node.sysctl_data = gp->froms;
128		node.sysctl_size = gp->fromssize;
129		break;
130	case GPROF_TOS:
131		node.sysctl_data = gp->tos;
132		node.sysctl_size = gp->tossize;
133		break;
134	case GPROF_GMONPARAM:
135		node.sysctl_data = gp;
136		node.sysctl_size = sizeof(*gp);
137		break;
138	default:
139		return (EOPNOTSUPP);
140	}
141
142	error = sysctl_lookup(SYSCTLFN_CALL(&node));
143	if (error || newp == NULL)
144		return (error);
145
146	if (node.sysctl_num == GPROF_STATE) {
147		mutex_spin_enter(&proc0.p_stmutex);
148		if (gp->state == GMON_PROF_OFF)
149			stopprofclock(&proc0);
150		else
151			startprofclock(&proc0);
152		mutex_spin_exit(&proc0.p_stmutex);
153	}
154
155	return (0);
156}
157
158SYSCTL_SETUP(sysctl_kern_gprof_setup, "sysctl kern.profiling subtree setup")
159{
160
161	sysctl_createv(clog, 0, NULL, NULL,
162		       CTLFLAG_PERMANENT,
163		       CTLTYPE_NODE, "profiling",
164		       SYSCTL_DESCR("Profiling information (available)"),
165		       NULL, 0, NULL, 0,
166		       CTL_KERN, KERN_PROF, CTL_EOL);
167
168	sysctl_createv(clog, 0, NULL, NULL,
169		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
170		       CTLTYPE_INT, "state",
171		       SYSCTL_DESCR("Profiling state"),
172		       sysctl_kern_profiling, 0, NULL, 0,
173		       CTL_KERN, KERN_PROF, GPROF_STATE, CTL_EOL);
174	sysctl_createv(clog, 0, NULL, NULL,
175		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
176		       CTLTYPE_STRUCT, "count",
177		       SYSCTL_DESCR("Array of statistical program counters"),
178		       sysctl_kern_profiling, 0, NULL, 0,
179		       CTL_KERN, KERN_PROF, GPROF_COUNT, CTL_EOL);
180	sysctl_createv(clog, 0, NULL, NULL,
181		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
182		       CTLTYPE_STRUCT, "froms",
183		       SYSCTL_DESCR("Array indexed by program counter of "
184				    "call-from points"),
185		       sysctl_kern_profiling, 0, NULL, 0,
186		       CTL_KERN, KERN_PROF, GPROF_FROMS, CTL_EOL);
187	sysctl_createv(clog, 0, NULL, NULL,
188		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
189		       CTLTYPE_STRUCT, "tos",
190		       SYSCTL_DESCR("Array of structures describing "
191				    "destination of calls and their counts"),
192		       sysctl_kern_profiling, 0, NULL, 0,
193		       CTL_KERN, KERN_PROF, GPROF_TOS, CTL_EOL);
194	sysctl_createv(clog, 0, NULL, NULL,
195		       CTLFLAG_PERMANENT,
196		       CTLTYPE_STRUCT, "gmonparam",
197		       SYSCTL_DESCR("Structure giving the sizes of the above "
198				    "arrays"),
199		       sysctl_kern_profiling, 0, NULL, 0,
200		       CTL_KERN, KERN_PROF, GPROF_GMONPARAM, CTL_EOL);
201}
202#endif /* GPROF */
203
204/*
205 * Profiling system call.
206 *
207 * The scale factor is a fixed point number with 16 bits of fraction, so that
208 * 1.0 is represented as 0x10000.  A scale factor of 0 turns off profiling.
209 */
210/* ARGSUSED */
211int
212sys_profil(struct lwp *l, const struct sys_profil_args *uap, register_t *retval)
213{
214	/* {
215		syscallarg(char *) samples;
216		syscallarg(size_t) size;
217		syscallarg(u_long) offset;
218		syscallarg(u_int) scale;
219	} */
220	struct proc *p = l->l_proc;
221	struct uprof *upp;
222
223	if (SCARG(uap, scale) > (1 << 16))
224		return (EINVAL);
225	if (SCARG(uap, scale) == 0) {
226		mutex_spin_enter(&p->p_stmutex);
227		stopprofclock(p);
228		mutex_spin_exit(&p->p_stmutex);
229		return (0);
230	}
231	upp = &p->p_stats->p_prof;
232
233	/* Block profile interrupts while changing state. */
234	mutex_spin_enter(&p->p_stmutex);
235	upp->pr_off = SCARG(uap, offset);
236	upp->pr_scale = SCARG(uap, scale);
237	upp->pr_base = SCARG(uap, samples);
238	upp->pr_size = SCARG(uap, size);
239	startprofclock(p);
240	mutex_spin_exit(&p->p_stmutex);
241
242	return (0);
243}
244
245/*
246 * Scale is a fixed-point number with the binary point 16 bits
247 * into the value, and is <= 1.0.  pc is at most 32 bits, so the
248 * intermediate result is at most 48 bits.
249 */
250#define	PC_TO_INDEX(pc, prof) \
251	((int)(((u_quad_t)((pc) - (prof)->pr_off) * \
252	    (u_quad_t)((prof)->pr_scale)) >> 16) & ~1)
253
254/*
255 * Collect user-level profiling statistics; called on a profiling tick,
256 * when a process is running in user-mode.  This routine may be called
257 * from an interrupt context.  We try to update the user profiling buffers
258 * cheaply with fuswintr() and suswintr().  If that fails, we revert to
259 * an AST that will vector us to trap() with a context in which copyin
260 * and copyout will work.  Trap will then call addupc_task().
261 *
262 * Note that we may (rarely) not get around to the AST soon enough, and
263 * lose profile ticks when the next tick overwrites this one, but in this
264 * case the system is overloaded and the profile is probably already
265 * inaccurate.
266 */
267void
268addupc_intr(struct lwp *l, u_long pc)
269{
270	struct uprof *prof;
271	struct proc *p;
272	void *addr;
273	u_int i;
274	int v;
275
276	p = l->l_proc;
277
278	KASSERT(mutex_owned(&p->p_stmutex));
279
280	prof = &p->p_stats->p_prof;
281	if (pc < prof->pr_off ||
282	    (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size)
283		return;			/* out of range; ignore */
284
285	addr = prof->pr_base + i;
286	mutex_spin_exit(&p->p_stmutex);
287	if ((v = fuswintr(addr)) == -1 || suswintr(addr, v + 1) == -1) {
288		/* XXXSMP */
289		prof->pr_addr = pc;
290		prof->pr_ticks++;
291		cpu_need_proftick(l);
292	}
293	mutex_spin_enter(&p->p_stmutex);
294}
295
296/*
297 * Much like before, but we can afford to take faults here.  If the
298 * update fails, we simply turn off profiling.
299 */
300void
301addupc_task(struct lwp *l, u_long pc, u_int ticks)
302{
303	struct uprof *prof;
304	struct proc *p;
305	void *addr;
306	int error;
307	u_int i;
308	u_short v;
309
310	p = l->l_proc;
311
312	if (ticks == 0)
313		return;
314
315	mutex_spin_enter(&p->p_stmutex);
316	prof = &p->p_stats->p_prof;
317
318	/* Testing P_PROFIL may be unnecessary, but is certainly safe. */
319	if ((p->p_stflag & PST_PROFIL) == 0 || pc < prof->pr_off ||
320	    (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size) {
321		mutex_spin_exit(&p->p_stmutex);
322		return;
323	}
324
325	addr = prof->pr_base + i;
326	mutex_spin_exit(&p->p_stmutex);
327	if ((error = copyin(addr, (void *)&v, sizeof(v))) == 0) {
328		v += ticks;
329		error = copyout((void *)&v, addr, sizeof(v));
330	}
331	if (error != 0) {
332		mutex_spin_enter(&p->p_stmutex);
333		stopprofclock(p);
334		mutex_spin_exit(&p->p_stmutex);
335	}
336}
337