prof_machdep.c revision 31395 
1/*-
2 * Copyright (c) 1996 Bruce D. Evans.
3 * 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 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 *
26 *	$Id: prof_machdep.c,v 1.6 1997/02/22 09:36:59 peter Exp $
27 */
28
29#ifdef GUPROF
30#include "opt_i586_guprof.h"
31#include "opt_perfmon.h"
32
33#include <sys/param.h>
34#include <sys/systm.h>
35#include <sys/gmon.h>
36#include <sys/kernel.h>
37#include <sys/sysctl.h>
38
39#include <machine/clock.h>
40#include <machine/perfmon.h>
41#include <machine/profile.h>
42#endif
43
44#ifdef PC98
45#include <pc98/pc98/pc98.h>
46#else
47#include <i386/isa/isa.h>
48#endif
49#include <i386/isa/timerreg.h>
50
51#ifdef GUPROF
52#define	CPUTIME_CLOCK_UNINITIALIZED	0
53#define	CPUTIME_CLOCK_I8254		1
54#define	CPUTIME_CLOCK_I586_CTR		2
55#define	CPUTIME_CLOCK_I586_PMC		3
56#define	CPUTIME_CLOCK_I8254_SHIFT	7
57
58int	cputime_bias = 1;	/* initialize for locality of reference */
59
60static int	cputime_clock = CPUTIME_CLOCK_UNINITIALIZED;
61#ifdef I586_PMC_GUPROF
62static u_int	cputime_clock_pmc_conf = I586_PMC_GUPROF;
63static int	cputime_clock_pmc_init;
64static struct gmonparam saved_gmp;
65#endif
66#endif /* GUPROF */
67
68#ifdef __GNUC__
69asm("
70GM_STATE	=	0
71GMON_PROF_OFF	=	3
72
73	.text
74	.align	4,0x90
75	.globl	__mcount
76__mcount:
77	#
78	# Check that we are profiling.  Do it early for speed.
79	#
80	cmpl	$GMON_PROF_OFF,__gmonparam+GM_STATE
81 	je	Lmcount_exit
82 	#
83 	# __mcount is the same as mcount except the caller hasn't changed
84 	# the stack except to call here, so the caller's raddr is above
85 	# our raddr.
86 	#
87 	movl	4(%esp),%edx
88 	jmp	Lgot_frompc
89
90 	.align	4,0x90
91 	.globl	mcount
92mcount:
93	cmpl	$GMON_PROF_OFF,__gmonparam+GM_STATE
94	je	Lmcount_exit
95	#
96	# The caller's stack frame has already been built, so %ebp is
97	# the caller's frame pointer.  The caller's raddr is in the
98	# caller's frame following the caller's caller's frame pointer.
99	#
100	movl	4(%ebp),%edx
101Lgot_frompc:
102	#
103	# Our raddr is the caller's pc.
104	#
105	movl	(%esp),%eax
106
107	pushfl
108	pushl	%eax
109	pushl	%edx
110	cli
111	call	_mcount
112	addl	$8,%esp
113	popfl
114Lmcount_exit:
115	ret
116");
117#else /* !__GNUC__ */
118#error
119#endif /* __GNUC__ */
120
121#ifdef GUPROF
122/*
123 * mexitcount saves the return register(s), loads selfpc and calls
124 * mexitcount(selfpc) to do the work.  Someday it should be in a machine
125 * dependent file together with cputime(), __mcount and mcount.  cputime()
126 * can't just be put in machdep.c because it has to be compiled without -pg.
127 */
128#ifdef __GNUC__
129asm("
130	.text
131#
132# Dummy label to be seen when gprof -u hides mexitcount.
133#
134	.align	4,0x90
135	.globl	__mexitcount
136__mexitcount:
137	nop
138
139GMON_PROF_HIRES	=	4
140
141	.align	4,0x90
142	.globl	mexitcount
143mexitcount:
144	cmpl	$GMON_PROF_HIRES,__gmonparam+GM_STATE
145	jne	Lmexitcount_exit
146	pushl	%edx
147	pushl	%eax
148	movl	8(%esp),%eax
149	pushfl
150	pushl	%eax
151	cli
152	call	_mexitcount
153	addl	$4,%esp
154	popfl
155	popl	%eax
156	popl	%edx
157Lmexitcount_exit:
158	ret
159");
160#else /* !__GNUC__ */
161#error
162#endif /* __GNUC__ */
163
164/*
165 * Return the time elapsed since the last call.  The units are machine-
166 * dependent.
167 */
168int
169cputime()
170{
171	u_int count;
172	int delta;
173#ifdef I586_PMC_GUPROF
174	u_quad_t event_count;
175#endif
176	u_char high, low;
177	static u_int prev_count;
178
179#if (defined(I586_CPU) || defined(I686_CPU)) && !defined(SMP)
180	if (cputime_clock == CPUTIME_CLOCK_I586_CTR) {
181		count = (u_int)rdtsc();
182		delta = (int)(count - prev_count);
183		prev_count = count;
184		return (delta);
185	}
186#if defined(PERFMON) && defined(I586_PMC_GUPROF)
187	if (cputime_clock == CPUTIME_CLOCK_I586_PMC) {
188		/*
189		 * XXX permon_read() should be inlined so that the
190		 * perfmon module doesn't need to be compiled with
191		 * profiling disabled and so that it is fast.
192		 */
193		perfmon_read(0, &event_count);
194
195		count = (u_int)event_count;
196		delta = (int)(count - prev_count);
197		prev_count = count;
198		return (delta);
199	}
200#endif /* PERFMON && I586_PMC_GUPROF */
201#endif /* (I586_CPU || I686_CPU) && !SMP */
202
203	/*
204	 * Read the current value of the 8254 timer counter 0.
205	 */
206	outb(TIMER_MODE, TIMER_SEL0 | TIMER_LATCH);
207	low = inb(TIMER_CNTR0);
208	high = inb(TIMER_CNTR0);
209	count = ((high << 8) | low) << CPUTIME_CLOCK_I8254_SHIFT;
210
211	/*
212	 * The timer counts down from TIMER_CNTR0_MAX to 0 and then resets.
213	 * While profiling is enabled, this routine is called at least twice
214	 * per timer reset (for mcounting and mexitcounting hardclock()),
215	 * so at most one reset has occurred since the last call, and one
216	 * has occurred iff the current count is larger than the previous
217	 * count.  This allows counter underflow to be detected faster
218	 * than in microtime().
219	 */
220	delta = prev_count - count;
221	prev_count = count;
222	if ((int) delta <= 0)
223		return (delta + (timer0_max_count << CPUTIME_CLOCK_I8254_SHIFT));
224	return (delta);
225}
226
227static int
228sysctl_machdep_cputime_clock SYSCTL_HANDLER_ARGS
229{
230	int clock;
231	int event;
232	int error;
233	struct pmc pmc;
234
235	clock = cputime_clock;
236#if defined(PERFMON) && defined(I586_PMC_GUPROF)
237	if (clock == CPUTIME_CLOCK_I586_PMC) {
238		pmc.pmc_val = cputime_clock_pmc_conf;
239		clock += pmc.pmc_event;
240	}
241#endif
242	error = sysctl_handle_opaque(oidp, &clock, sizeof clock, req);
243	if (error == 0 && req->newptr != NULL) {
244#if defined(PERFMON) && defined(I586_PMC_GUPROF)
245		if (clock >= CPUTIME_CLOCK_I586_PMC) {
246			event = clock - CPUTIME_CLOCK_I586_PMC;
247			if (event >= 256)
248				return (EINVAL);
249			pmc.pmc_num = 0;
250			pmc.pmc_event = event;
251			pmc.pmc_unit = 0;
252			pmc.pmc_flags = PMCF_E | PMCF_OS | PMCF_USR;
253			pmc.pmc_mask = 0;
254			cputime_clock_pmc_conf = pmc.pmc_val;
255			cputime_clock = CPUTIME_CLOCK_I586_PMC;
256		} else
257#endif
258		{
259			if (clock < 0 || clock >= CPUTIME_CLOCK_I586_PMC)
260				return (EINVAL);
261			cputime_clock = clock;
262		}
263	}
264	return (error);
265}
266
267SYSCTL_PROC(_machdep, OID_AUTO, cputime_clock, CTLTYPE_INT | CTLFLAG_RW,
268	    0, sizeof(u_int), sysctl_machdep_cputime_clock, "I", "");
269
270/*
271 * The start and stop routines need not be here since we turn off profiling
272 * before calling them.  They are here for convenience.
273 */
274
275void
276startguprof(gp)
277	struct gmonparam *gp;
278{
279	if (cputime_clock == CPUTIME_CLOCK_UNINITIALIZED) {
280		cputime_clock = CPUTIME_CLOCK_I8254;
281#if (defined(I586_CPU) || defined(I686_CPU)) && !defined(SMP)
282		if (i586_ctr_freq != 0)
283			cputime_clock = CPUTIME_CLOCK_I586_CTR;
284#endif
285	}
286	gp->profrate = timer_freq << CPUTIME_CLOCK_I8254_SHIFT;
287#if (defined(I586_CPU) || defined(I686_CPU)) && !defined(SMP)
288	if (cputime_clock == CPUTIME_CLOCK_I586_CTR)
289		gp->profrate = i586_ctr_freq;
290#if defined(PERFMON) && defined(I586_PMC_GUPROF)
291	else if (cputime_clock == CPUTIME_CLOCK_I586_PMC) {
292		if (perfmon_avail() &&
293		    perfmon_setup(0, cputime_clock_pmc_conf) == 0) {
294			if (perfmon_start(0) != 0)
295				perfmon_fini(0);
296			else {
297				/* XXX 1 event == 1 us. */
298				gp->profrate = 1000000;
299
300				saved_gmp = *gp;
301
302				/* Zap overheads.  They are invalid. */
303				gp->cputime_overhead = 0;
304				gp->mcount_overhead = 0;
305				gp->mcount_post_overhead = 0;
306				gp->mcount_pre_overhead = 0;
307				gp->mexitcount_overhead = 0;
308				gp->mexitcount_post_overhead = 0;
309				gp->mexitcount_pre_overhead = 0;
310
311				cputime_clock_pmc_init = TRUE;
312			}
313		}
314	}
315#endif /* PERFMON && I586_PMC_GUPROF */
316#endif /* (I586_CPU || I686_CPU) && !SMP */
317	cputime_bias = 0;
318	cputime();
319}
320
321void
322stopguprof(gp)
323	struct gmonparam *gp;
324{
325#if defined(PERFMON) && defined(I586_PMC_GUPROF)
326	if (cputime_clock_pmc_init) {
327		*gp = saved_gmp;
328		perfmon_fini(0);
329		cputime_clock_pmc_init = FALSE;
330	}
331#endif
332}
333
334#else /* !GUPROF */
335#ifdef __GNUC__
336asm("
337	.text
338	.align	4,0x90
339	.globl	mexitcount
340mexitcount:
341	ret
342");
343#else /* !__GNUC__ */
344#error
345#endif /* __GNUC__ */
346#endif /* GUPROF */
347