sys/platform/pc32/isa/prof_machdep.c

   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  * $FreeBSD: src/sys/i386/isa/prof_machdep.c,v 1.14.2.1 2000/08/03 00:09:30 ps Exp $
  27  * $DragonFly: src/sys/platform/pc32/isa/prof_machdep.c,v 1.10 2008/08/02 01:14:43 dillon Exp $
  28  */
  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 #include <bus/isa/isa.h>
  49 #include <machine_base/isa/timerreg.h>
  50
  51 #ifdef GUPROF
  52 #define CPUTIME_CLOCK_UNINITIALIZED     0
  53 #define CPUTIME_CLOCK_I8254             1
  54 #define CPUTIME_CLOCK_TSC               2
  55 #define CPUTIME_CLOCK_I586_PMC          3
  56 #define CPUTIME_CLOCK_I8254_SHIFT       7
  57
  58 int     cputime_bias = 1;       /* initialize for locality of reference */
  59
  60 static int      cputime_clock = CPUTIME_CLOCK_UNINITIALIZED;
  61 #ifdef I586_PMC_GUPROF
  62 static u_int    cputime_clock_pmc_conf = I586_PMC_GUPROF;
  63 static int      cputime_clock_pmc_init;
  64 static struct gmonparam saved_gmp;
  65 #endif
  66 #endif /* GUPROF */
  67
  68 #ifdef __GNUC__
  69 __asm("                                                         \n\
  70 GM_STATE        =       0                                       \n\
  71 GMON_PROF_OFF   =       3                                       \n\
  72                                                                 \n\
  73         .text                                                   \n\
  74         .p2align 4,0x90                                         \n\
  75         .globl  __mcount                                        \n\
  76         .type   __mcount,@function                              \n\
  77 __mcount:                                                       \n\
  78         #                                                       \n\
  79         # Check that we are profiling.  Do it early for speed.  \n\
  80         #                                                       \n\
  81         cmpl    $GMON_PROF_OFF," __XSTRING(CNAME(_gmonparam)) "+GM_STATE \n\
  82         je      .mcount_exit                                    \n\
  83         #                                                       \n\
  84         # __mcount is the same as [.]mcount except the caller   \n\
  85         # hasn't changed the stack except to call here, so the  \n\
  86         # caller's raddr is above our raddr.                    \n\
  87         #                                                       \n\
  88         pushl   %eax                                            \n\
  89         pushl   %ecx                                            \n\
  90         pushl   %edx                                            \n\
  91         movl    12+4(%esp),%edx                                 \n\
  92         jmp     .got_frompc                                     \n\
  93                                                                 \n\
  94         .p2align 4,0x90                                         \n\
  95         .globl  " __XSTRING(HIDENAME(mcount)) "                 \n\
  96 " __XSTRING(HIDENAME(mcount)) ":                                \n\
  97         cmpl    $GMON_PROF_OFF," __XSTRING(CNAME(_gmonparam)) "+GM_STATE \n\
  98         je      .mcount_exit                                    \n\
  99         #                                                       \n\
 100         # The caller's stack frame has already been built, so   \n\
 101         # %ebp is the caller's frame pointer.  The caller's     \n\
 102         # raddr is in the caller's frame following the caller's \n\
 103         # caller's frame pointer.                               \n\
 104         #                                                       \n\
 105         pushl   %eax                                            \n\
 106         pushl   %ecx                                            \n\
 107         pushl   %edx                                            \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         popl    %edx                                            \n\
 123         popl    %ecx                                            \n\
 124         popl    %eax                                            \n\
 125 .mcount_exit:                                                   \n\
 126         ret                                                     \n\
 127 ");
 128 #else /* !__GNUC__ */
 129 #error
 130 #endif /* __GNUC__ */
 131
 132 #ifdef GUPROF
 133 /*
 134  * [.]mexitcount saves the return register(s), loads selfpc and calls
 135  * mexitcount(selfpc) to do the work.  Someday it should be in a machine
 136  * dependent file together with cputime(), __mcount and [.]mcount.  cputime()
 137  * can't just be put in machdep.c because it has to be compiled without -pg.
 138  */
 139 #ifdef __GNUC__
 140 __asm("                                                         \n\
 141         .text                                                   \n\
 142 #                                                               \n\
 143 # Dummy label to be seen when gprof -u hides [.]mexitcount.     \n\
 144 #                                                               \n\
 145         .p2align 4,0x90                                         \n\
 146         .globl  __mexitcount                                    \n\
 147         .type   __mexitcount,@function                          \n\
 148 __mexitcount:                                                   \n\
 149         nop                                                     \n\
 150                                                                 \n\
 151 GMON_PROF_HIRES =       4                                       \n\
 152                                                                 \n\
 153         .p2align 4,0x90                                         \n\
 154         .globl  " __XSTRING(HIDENAME(mexitcount)) "             \n\
 155 " __XSTRING(HIDENAME(mexitcount)) ":                            \n\
 156         cmpl    $GMON_PROF_HIRES," __XSTRING(CNAME(_gmonparam)) "+GM_STATE \n\
 157         jne     .mexitcount_exit                                \n\
 158         pushl   %edx                                            \n\
 159         pushl   %ecx                                            \n\
 160         pushl   %eax                                            \n\
 161         movl    12(%esp),%eax                                   \n\
 162         pushfl                                                  \n\
 163         pushl   %eax                                            \n\
 164         cli                                                     \n\
 165         call    " __XSTRING(CNAME(mexitcount)) "                \n\
 166         addl    $4,%esp                                         \n\
 167         popfl                                                   \n\
 168         popl    %eax                                            \n\
 169         popl    %ecx                                            \n\
 170         popl    %edx                                            \n\
 171 .mexitcount_exit:                                               \n\
 172         ret                                                     \n\
 173 ");
 174 #else /* !__GNUC__ */
 175 #error
 176 #endif /* __GNUC__ */
 177
 178 /*
 179  * Return the time elapsed since the last call.  The units are machine-
 180  * dependent.
 181  */
 182 int
 183 cputime(void)
 184 {
 185         u_int count;
 186         int delta;
 187 #if (defined(I586_CPU) || defined(I686_CPU)) && !defined(SMP) && \
 188     defined(PERFMON) && defined(I586_PMC_GUPROF)
 189         u_quad_t event_count;
 190 #endif
 191         u_char high, low;
 192         static u_int prev_count;
 193
 194 #if (defined(I586_CPU) || defined(I686_CPU)) && !defined(SMP)
 195         if (cputime_clock == CPUTIME_CLOCK_TSC) {
 196                 count = (u_int)rdtsc();
 197                 delta = (int)(count - prev_count);
 198                 prev_count = count;
 199                 return (delta);
 200         }
 201 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
 202         if (cputime_clock == CPUTIME_CLOCK_I586_PMC) {
 203                 /*
 204                  * XXX permon_read() should be inlined so that the
 205                  * perfmon module doesn't need to be compiled with
 206                  * profiling disabled and so that it is fast.
 207                  */
 208                 perfmon_read(0, &event_count);
 209
 210                 count = (u_int)event_count;
 211                 delta = (int)(count - prev_count);
 212                 prev_count = count;
 213                 return (delta);
 214         }
 215 #endif /* PERFMON && I586_PMC_GUPROF */
 216 #endif /* (I586_CPU || I686_CPU) && !SMP */
 217
 218         /*
 219          * Read the current value of the 8254 timer counter 0.
 220          */
 221         outb(TIMER_MODE, TIMER_SEL0 | TIMER_LATCH);
 222         low = inb(TIMER_CNTR0);
 223         high = inb(TIMER_CNTR0);
 224         count = ((high << 8) | low) << CPUTIME_CLOCK_I8254_SHIFT;
 225
 226         /*
 227          * The timer counts down from TIMER_CNTR0_MAX to 0 and then resets.
 228          * While profiling is enabled, this routine is called at least twice
 229          * per timer reset (for mcounting and mexitcounting hardclock()),
 230          * so at most one reset has occurred since the last call, and one
 231          * has occurred iff the current count is larger than the previous
 232          * count.  This allows counter underflow to be detected faster
 233          * than in microtime().
 234          */
 235         delta = prev_count - count;
 236         prev_count = count;
 237         if ((int) delta <= 0)
 238                 return (delta + (timer0_max_count << CPUTIME_CLOCK_I8254_SHIFT));
 239         return (delta);
 240 }
 241
 242 static int
 243 sysctl_machdep_cputime_clock(SYSCTL_HANDLER_ARGS)
 244 {
 245         int clock;
 246         int error;
 247 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
 248         int event;
 249         struct pmc pmc;
 250 #endif
 251
 252         clock = cputime_clock;
 253 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
 254         if (clock == CPUTIME_CLOCK_I586_PMC) {
 255                 pmc.pmc_val = cputime_clock_pmc_conf;
 256                 clock += pmc.pmc_event;
 257         }
 258 #endif
 259         error = sysctl_handle_opaque(oidp, &clock, sizeof clock, req);
 260         if (error == 0 && req->newptr != NULL) {
 261 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
 262                 if (clock >= CPUTIME_CLOCK_I586_PMC) {
 263                         event = clock - CPUTIME_CLOCK_I586_PMC;
 264                         if (event >= 256)
 265                                 return (EINVAL);
 266                         pmc.pmc_num = 0;
 267                         pmc.pmc_event = event;
 268                         pmc.pmc_unit = 0;
 269                         pmc.pmc_flags = PMCF_E | PMCF_OS | PMCF_USR;
 270                         pmc.pmc_mask = 0;
 271                         cputime_clock_pmc_conf = pmc.pmc_val;
 272                         cputime_clock = CPUTIME_CLOCK_I586_PMC;
 273                 } else
 274 #endif
 275                 {
 276                         if (clock < 0 || clock >= CPUTIME_CLOCK_I586_PMC)
 277                                 return (EINVAL);
 278                         cputime_clock = clock;
 279                 }
 280         }
 281         return (error);
 282 }
 283
 284 SYSCTL_PROC(_machdep, OID_AUTO, cputime_clock, CTLTYPE_INT | CTLFLAG_RW,
 285             0, sizeof(u_int), sysctl_machdep_cputime_clock, "I", "");
 286
 287 /*
 288  * The start and stop routines need not be here since we turn off profiling
 289  * before calling them.  They are here for convenience.
 290  */
 291
 292 void
 293 startguprof(struct gmonparam *gp)
 294 {
 295         if (cputime_clock == CPUTIME_CLOCK_UNINITIALIZED) {
 296                 cputime_clock = CPUTIME_CLOCK_I8254;
 297 #if (defined(I586_CPU) || defined(I686_CPU)) && !defined(SMP)
 298                 if (tsc_frequency != 0)
 299                         cputime_clock = CPUTIME_CLOCK_TSC;
 300 #endif
 301         }
 302         gp->profrate = timer_freq << CPUTIME_CLOCK_I8254_SHIFT;
 303 #if (defined(I586_CPU) || defined(I686_CPU)) && !defined(SMP)
 304         if (cputime_clock == CPUTIME_CLOCK_TSC)
 305                 gp->profrate = (u_int)tsc_frequency;    /* XXX */
 306 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
 307         else if (cputime_clock == CPUTIME_CLOCK_I586_PMC) {
 308                 if (perfmon_avail() &&
 309                     perfmon_setup(0, cputime_clock_pmc_conf) == 0) {
 310                         if (perfmon_start(0) != 0)
 311                                 perfmon_fini(0);
 312                         else {
 313                                 /* XXX 1 event == 1 us. */
 314                                 gp->profrate = 1000000;
 315
 316                                 saved_gmp = *gp;
 317
 318                                 /* Zap overheads.  They are invalid. */
 319                                 gp->cputime_overhead = 0;
 320                                 gp->mcount_overhead = 0;
 321                                 gp->mcount_post_overhead = 0;
 322                                 gp->mcount_pre_overhead = 0;
 323                                 gp->mexitcount_overhead = 0;
 324                                 gp->mexitcount_post_overhead = 0;
 325                                 gp->mexitcount_pre_overhead = 0;
 326
 327                                 cputime_clock_pmc_init = TRUE;
 328                         }
 329                 }
 330         }
 331 #endif /* PERFMON && I586_PMC_GUPROF */
 332 #endif /* (I586_CPU || I686_CPU) && !SMP */
 333         cputime_bias = 0;
 334         cputime();
 335 }
 336
 337 void
 338 stopguprof(struct gmonparam *gp)
 339 {
 340 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
 341         if (cputime_clock_pmc_init) {
 342                 *gp = saved_gmp;
 343                 perfmon_fini(0);
 344                 cputime_clock_pmc_init = FALSE;
 345         }
 346 #endif
 347 }
 348
 349 #else /* !GUPROF */
 350 #ifdef __GNUC__
 351 __asm("                                                         \n\
 352         .text                                                   \n\
 353         .p2align 4,0x90                                         \n\
 354         .globl  " __XSTRING(HIDENAME(mexitcount)) "             \n\
 355 " __XSTRING(HIDENAME(mexitcount)) ":                            \n\
 356         ret                                                     \n\
 357 ");
 358 #else /* !__GNUC__ */
 359 #error
 360 #endif /* __GNUC__ */
 361 #endif /* GUPROF */