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
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD: src/sys/i386/isa/prof_machdep.c,v 1.20 2004/05/26 09:43:38 bde Exp $");
29
30 #ifdef GUPROF
31 #include "opt_i586_guprof.h"
32 #include "opt_perfmon.h"
33
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/gmon.h>
37 #include <sys/kernel.h>
38 #include <sys/sysctl.h>
39
40 #include <machine/clock.h>
41 #include <machine/perfmon.h>
42 #include <machine/profile.h>
43 #undef MCOUNT
44 #endif
45
46 #include <machine/asmacros.h>
47
48 #ifdef PC98
49 #include <pc98/pc98/pc98.h>
50 #else
51 #include <i386/isa/isa.h>
52 #endif
53 #include <i386/isa/timerreg.h>
54
55 #ifdef GUPROF
56 #define CPUTIME_CLOCK_UNINITIALIZED 0
57 #define CPUTIME_CLOCK_I8254 1
58 #define CPUTIME_CLOCK_TSC 2
59 #define CPUTIME_CLOCK_I586_PMC 3
60 #define CPUTIME_CLOCK_I8254_SHIFT 7
61
62 int cputime_bias = 1; /* initialize for locality of reference */
63
64 static int cputime_clock = CPUTIME_CLOCK_UNINITIALIZED;
65 #ifdef I586_PMC_GUPROF
66 static u_int cputime_clock_pmc_conf = I586_PMC_GUPROF;
67 static int cputime_clock_pmc_init;
68 static struct gmonparam saved_gmp;
69 #endif
70 #endif /* GUPROF */
71
72 #if defined(__GNUC__) || defined(__INTEL_COMPILER)
73 __asm(" \n\
74 GM_STATE = 0 \n\
75 GMON_PROF_OFF = 3 \n\
76 \n\
77 .text \n\
78 .p2align 4,0x90 \n\
79 .globl __mcount \n\
80 .type __mcount,@function \n\
81 __mcount: \n\
82 # \n\
83 # Check that we are profiling. Do it early for speed. \n\
84 # \n\
85 cmpl $GMON_PROF_OFF," __XSTRING(CNAME(_gmonparam)) "+GM_STATE \n\
86 je .mcount_exit \n\
87 # \n\
88 # __mcount is the same as [.]mcount except the caller \n\
89 # hasn't changed the stack except to call here, so the \n\
90 # caller's raddr is above our raddr. \n\
91 # \n\
92 movl 4(%esp),%edx \n\
93 jmp .got_frompc \n\
94 \n\
95 .p2align 4,0x90 \n\
96 .globl " __XSTRING(HIDENAME(mcount)) " \n\
97 " __XSTRING(HIDENAME(mcount)) ": \n\
98 .globl __cyg_profile_func_enter \n\
99 __cyg_profile_func_enter: \n\
100 cmpl $GMON_PROF_OFF," __XSTRING(CNAME(_gmonparam)) "+GM_STATE \n\
101 je .mcount_exit \n\
102 # \n\
103 # The caller's stack frame has already been built, so \n\
104 # %ebp is the caller's frame pointer. The caller's \n\
105 # raddr is in the caller's frame following the caller's \n\
106 # caller's frame pointer. \n\
107 # \n\
108 movl 4(%ebp),%edx \n\
109 .got_frompc: \n\
110 # \n\
111 # Our raddr is the caller's pc. \n\
112 # \n\
113 movl (%esp),%eax \n\
114 \n\
115 pushfl \n\
116 pushl %eax \n\
117 pushl %edx \n\
118 cli \n\
119 call " __XSTRING(CNAME(mcount)) " \n\
120 addl $8,%esp \n\
121 popfl \n\
122 .mcount_exit: \n\
123 ret \n\
124 ");
125 #else /* !(__GNUC__ || __INTEL_COMPILER) */
126 #error
127 #endif /* __GNUC__ || __INTEL_COMPILER */
128
129 #ifdef GUPROF
130 /*
131 * [.]mexitcount saves the return register(s), loads selfpc and calls
132 * mexitcount(selfpc) to do the work. Someday it should be in a machine
133 * dependent file together with cputime(), __mcount and [.]mcount. cputime()
134 * can't just be put in machdep.c because it has to be compiled without -pg.
135 */
136 #if defined(__GNUC__) || defined(__INTEL_COMPILER)
137 __asm(" \n\
138 .text \n\
139 # \n\
140 # Dummy label to be seen when gprof -u hides [.]mexitcount. \n\
141 # \n\
142 .p2align 4,0x90 \n\
143 .globl __mexitcount \n\
144 .type __mexitcount,@function \n\
145 __mexitcount: \n\
146 nop \n\
147 \n\
148 GMON_PROF_HIRES = 4 \n\
149 \n\
150 .p2align 4,0x90 \n\
151 .globl " __XSTRING(HIDENAME(mexitcount)) " \n\
152 " __XSTRING(HIDENAME(mexitcount)) ": \n\
153 .globl __cyg_profile_func_exit \n\
154 __cyg_profile_func_exit: \n\
155 cmpl $GMON_PROF_HIRES," __XSTRING(CNAME(_gmonparam)) "+GM_STATE \n\
156 jne .mexitcount_exit \n\
157 pushl %edx \n\
158 pushl %eax \n\
159 movl 8(%esp),%eax \n\
160 pushfl \n\
161 pushl %eax \n\
162 cli \n\
163 call " __XSTRING(CNAME(mexitcount)) " \n\
164 addl $4,%esp \n\
165 popfl \n\
166 popl %eax \n\
167 popl %edx \n\
168 .mexitcount_exit: \n\
169 ret \n\
170 ");
171 #else /* !(__GNUC__ || __INTEL_COMPILER) */
172 #error
173 #endif /* __GNUC__ || __INTEL_COMPILER */
174
175 /*
176 * Return the time elapsed since the last call. The units are machine-
177 * dependent.
178 */
179 int
180 cputime()
181 {
182 u_int count;
183 int delta;
184 #if (defined(I586_CPU) || defined(I686_CPU)) && !defined(SMP) && \
185 defined(PERFMON) && defined(I586_PMC_GUPROF)
186 u_quad_t event_count;
187 #endif
188 u_char high, low;
189 static u_int prev_count;
190
191 #if (defined(I586_CPU) || defined(I686_CPU)) && !defined(SMP)
192 if (cputime_clock == CPUTIME_CLOCK_TSC) {
193 /*
194 * Scale the TSC a little to make cputime()'s frequency
195 * fit in an int, assuming that the TSC frequency fits
196 * in a u_int. Use a fixed scale since dynamic scaling
197 * would be slower and we can't really use the low bit
198 * of precision.
199 */
200 count = (u_int)rdtsc() & ~1u;
201 delta = (int)(count - prev_count) >> 1;
202 prev_count = count;
203 return (delta);
204 }
205 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
206 if (cputime_clock == CPUTIME_CLOCK_I586_PMC) {
207 /*
208 * XXX permon_read() should be inlined so that the
209 * perfmon module doesn't need to be compiled with
210 * profiling disabled and so that it is fast.
211 */
212 perfmon_read(0, &event_count);
213
214 count = (u_int)event_count;
215 delta = (int)(count - prev_count);
216 prev_count = count;
217 return (delta);
218 }
219 #endif /* PERFMON && I586_PMC_GUPROF */
220 #endif /* (I586_CPU || I686_CPU) && !SMP */
221
222 /*
223 * Read the current value of the 8254 timer counter 0.
224 */
225 outb(TIMER_MODE, TIMER_SEL0 | TIMER_LATCH);
226 low = inb(TIMER_CNTR0);
227 high = inb(TIMER_CNTR0);
228 count = ((high << 8) | low) << CPUTIME_CLOCK_I8254_SHIFT;
229
230 /*
231 * The timer counts down from TIMER_CNTR0_MAX to 0 and then resets.
232 * While profiling is enabled, this routine is called at least twice
233 * per timer reset (for mcounting and mexitcounting hardclock()),
234 * so at most one reset has occurred since the last call, and one
235 * has occurred iff the current count is larger than the previous
236 * count. This allows counter underflow to be detected faster
237 * than in microtime().
238 */
239 delta = prev_count - count;
240 prev_count = count;
241 if ((int) delta <= 0)
242 return (delta + (timer0_max_count << CPUTIME_CLOCK_I8254_SHIFT));
243 return (delta);
244 }
245
246 static int
247 sysctl_machdep_cputime_clock(SYSCTL_HANDLER_ARGS)
248 {
249 int clock;
250 int error;
251 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
252 int event;
253 struct pmc pmc;
254 #endif
255
256 clock = cputime_clock;
257 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
258 if (clock == CPUTIME_CLOCK_I586_PMC) {
259 pmc.pmc_val = cputime_clock_pmc_conf;
260 clock += pmc.pmc_event;
261 }
262 #endif
263 error = sysctl_handle_opaque(oidp, &clock, sizeof clock, req);
264 if (error == 0 && req->newptr != NULL) {
265 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
266 if (clock >= CPUTIME_CLOCK_I586_PMC) {
267 event = clock - CPUTIME_CLOCK_I586_PMC;
268 if (event >= 256)
269 return (EINVAL);
270 pmc.pmc_num = 0;
271 pmc.pmc_event = event;
272 pmc.pmc_unit = 0;
273 pmc.pmc_flags = PMCF_E | PMCF_OS | PMCF_USR;
274 pmc.pmc_mask = 0;
275 cputime_clock_pmc_conf = pmc.pmc_val;
276 cputime_clock = CPUTIME_CLOCK_I586_PMC;
277 } else
278 #endif
279 {
280 if (clock < 0 || clock >= CPUTIME_CLOCK_I586_PMC)
281 return (EINVAL);
282 cputime_clock = clock;
283 }
284 }
285 return (error);
286 }
287
288 SYSCTL_PROC(_machdep, OID_AUTO, cputime_clock, CTLTYPE_INT | CTLFLAG_RW,
289 0, sizeof(u_int), sysctl_machdep_cputime_clock, "I", "");
290
291 /*
292 * The start and stop routines need not be here since we turn off profiling
293 * before calling them. They are here for convenience.
294 */
295
296 void
297 startguprof(gp)
298 struct gmonparam *gp;
299 {
300 if (cputime_clock == CPUTIME_CLOCK_UNINITIALIZED) {
301 cputime_clock = CPUTIME_CLOCK_I8254;
302 #if (defined(I586_CPU) || defined(I686_CPU)) && !defined(SMP)
303 if (tsc_freq != 0)
304 cputime_clock = CPUTIME_CLOCK_TSC;
305 #endif
306 }
307 gp->profrate = timer_freq << CPUTIME_CLOCK_I8254_SHIFT;
308 #if (defined(I586_CPU) || defined(I686_CPU)) && !defined(SMP)
309 if (cputime_clock == CPUTIME_CLOCK_TSC)
310 gp->profrate = tsc_freq >> 1;
311 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
312 else if (cputime_clock == CPUTIME_CLOCK_I586_PMC) {
313 if (perfmon_avail() &&
314 perfmon_setup(0, cputime_clock_pmc_conf) == 0) {
315 if (perfmon_start(0) != 0)
316 perfmon_fini(0);
317 else {
318 /* XXX 1 event == 1 us. */
319 gp->profrate = 1000000;
320
321 saved_gmp = *gp;
322
323 /* Zap overheads. They are invalid. */
324 gp->cputime_overhead = 0;
325 gp->mcount_overhead = 0;
326 gp->mcount_post_overhead = 0;
327 gp->mcount_pre_overhead = 0;
328 gp->mexitcount_overhead = 0;
329 gp->mexitcount_post_overhead = 0;
330 gp->mexitcount_pre_overhead = 0;
331
332 cputime_clock_pmc_init = TRUE;
333 }
334 }
335 }
336 #endif /* PERFMON && I586_PMC_GUPROF */
337 #endif /* (I586_CPU || I686_CPU) && !SMP */
338 cputime_bias = 0;
339 cputime();
340 }
341
342 void
343 stopguprof(gp)
344 struct gmonparam *gp;
345 {
346 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
347 if (cputime_clock_pmc_init) {
348 *gp = saved_gmp;
349 perfmon_fini(0);
350 cputime_clock_pmc_init = FALSE;
351 }
352 #endif
353 }
354
355 #else /* !GUPROF */
356 #if defined(__GNUC__) || defined(__INTEL_COMPILER)
357 __asm(" \n\
358 .text \n\
359 .p2align 4,0x90 \n\
360 .globl " __XSTRING(HIDENAME(mexitcount)) " \n\
361 " __XSTRING(HIDENAME(mexitcount)) ": \n\
362 ret \n\
363 ");
364 #else /* !(__GNUC__ || __INTEL_COMPILER) */
365 #error
366 #endif /* __GNUC__ || __INTEL_COMPILER */
367 #endif /* GUPROF */
Cache object: 2eb987085b531aaa274b005babc854ac
|