sys/platform/pc64/isa/prof_machdep.c

   1 /*-
   2  * Copyright (c) 1996 Bruce D. Evans.
   3  * Copyright (c) 2008 The DragonFly Project.
   4  * All rights reserved.
   5  *
   6  * Redistribution and use in source and binary forms, with or without
   7  * modification, are permitted provided that the following conditions
   8  * are met:
   9  * 1. Redistributions of source code must retain the above copyright
  10  *    notice, this list of conditions and the following disclaimer.
  11  * 2. Redistributions in binary form must reproduce the above copyright
  12  *    notice, this list of conditions and the following disclaimer in the
  13  *    documentation and/or other materials provided with the distribution.
  14  *
  15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  25  * SUCH DAMAGE.
  26  *
  27  * $FreeBSD: src/sys/i386/isa/prof_machdep.c,v 1.14.2.1 2000/08/03 00:09:30 ps Exp $
  28  * $DragonFly: src/sys/platform/pc64/isa/prof_machdep.c,v 1.1 2008/08/29 17:07:20 dillon Exp $
  29  */
  30
  31 #ifdef GUPROF
  32 #include "opt_i586_guprof.h"
  33 #include "opt_perfmon.h"
  34
  35 #include <sys/param.h>
  36 #include <sys/systm.h>
  37 #include <sys/gmon.h>
  38 #include <sys/kernel.h>
  39 #include <sys/sysctl.h>
  40
  41 #include <machine/clock.h>
  42 #include <machine/perfmon.h>
  43 #include <machine/profile.h>
  44 #undef MCOUNT
  45 #endif
  46
  47 #include <machine/asmacros.h>
  48
  49 #include <bus/isa/isa.h>
  50 #include <machine_base/isa/timerreg.h>
  51
  52 #ifdef GUPROF
  53 #define CPUTIME_CLOCK_UNINITIALIZED     0
  54 #define CPUTIME_CLOCK_I8254             1
  55 #define CPUTIME_CLOCK_TSC               2
  56 #define CPUTIME_CLOCK_I586_PMC          3
  57 #define CPUTIME_CLOCK_I8254_SHIFT       7
  58
  59 int     cputime_bias = 1;       /* initialize for locality of reference */
  60
  61 static int      cputime_clock = CPUTIME_CLOCK_UNINITIALIZED;
  62 #ifdef I586_PMC_GUPROF
  63 static u_int    cputime_clock_pmc_conf = I586_PMC_GUPROF;
  64 static int      cputime_clock_pmc_init;
  65 static struct gmonparam saved_gmp;
  66 #endif
  67 #endif /* GUPROF */
  68
  69 #ifdef __GNUC__
  70 __asm("                                                         \n\
  71 GM_STATE        =       0                                       \n\
  72 GMON_PROF_OFF   =       3                                       \n\
  73                                                                 \n\
  74         .text                                                   \n\
  75         .p2align 4,0x90                                         \n\
  76         .globl  __mcount                                        \n\
  77         .type   __mcount,@function                              \n\
  78 __mcount:                                                       \n\
  79         #                                                       \n\
  80         # Check that we are profiling.  Do it early for speed.  \n\
  81         #                                                       \n\
  82         cmpl    $GMON_PROF_OFF," __XSTRING(CNAME(_gmonparam)) "+GM_STATE \n\
  83         je      .mcount_exit                                    \n\
  84         #                                                       \n\
  85         # __mcount is the same as [.]mcount except the caller   \n\
  86         # hasn't changed the stack except to call here, so the  \n\
  87         # caller's raddr is above our raddr.                    \n\
  88         #                                                       \n\
  89         pushl   %eax                                            \n\
  90         pushl   %ecx                                            \n\
  91         pushl   %edx                                            \n\
  92         movl    12+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         cmpl    $GMON_PROF_OFF," __XSTRING(CNAME(_gmonparam)) "+GM_STATE \n\
  99         je      .mcount_exit                                    \n\
 100         #                                                       \n\
 101         # The caller's stack frame has already been built, so   \n\
 102         # %ebp is the caller's frame pointer.  The caller's     \n\
 103         # raddr is in the caller's frame following the caller's \n\
 104         # caller's frame pointer.                               \n\
 105         #                                                       \n\
 106         pushl   %eax                                            \n\
 107         pushl   %ecx                                            \n\
 108         pushl   %edx                                            \n\
 109         movl    4(%ebp),%edx                                    \n\
 110 .got_frompc:                                                    \n\
 111         #                                                       \n\
 112         # Our raddr is the caller's pc.                         \n\
 113         #                                                       \n\
 114         movl    (%esp),%eax                                     \n\
 115                                                                 \n\
 116         pushfl                                                  \n\
 117         pushl   %eax                                            \n\
 118         pushl   %edx                                            \n\
 119         cli                                                     \n\
 120         call    " __XSTRING(CNAME(mcount)) "                    \n\
 121         addl    $8,%esp                                         \n\
 122         popfl                                                   \n\
 123         popl    %edx                                            \n\
 124         popl    %ecx                                            \n\
 125         popl    %eax                                            \n\
 126 .mcount_exit:                                                   \n\
 127         ret                                                     \n\
 128 ");
 129 #else /* !__GNUC__ */
 130 #error
 131 #endif /* __GNUC__ */
 132
 133 #ifdef GUPROF
 134 /*
 135  * [.]mexitcount saves the return register(s), loads selfpc and calls
 136  * mexitcount(selfpc) to do the work.  Someday it should be in a machine
 137  * dependent file together with cputime(), __mcount and [.]mcount.  cputime()
 138  * can't just be put in machdep.c because it has to be compiled without -pg.
 139  */
 140 #ifdef __GNUC__
 141 __asm("                                                         \n\
 142         .text                                                   \n\
 143 #                                                               \n\
 144 # Dummy label to be seen when gprof -u hides [.]mexitcount.     \n\
 145 #                                                               \n\
 146         .p2align 4,0x90                                         \n\
 147         .globl  __mexitcount                                    \n\
 148         .type   __mexitcount,@function                          \n\
 149 __mexitcount:                                                   \n\
 150         nop                                                     \n\
 151                                                                 \n\
 152 GMON_PROF_HIRES =       4                                       \n\
 153                                                                 \n\
 154         .p2align 4,0x90                                         \n\
 155         .globl  " __XSTRING(HIDENAME(mexitcount)) "             \n\
 156 " __XSTRING(HIDENAME(mexitcount)) ":                            \n\
 157         cmpl    $GMON_PROF_HIRES," __XSTRING(CNAME(_gmonparam)) "+GM_STATE \n\
 158         jne     .mexitcount_exit                                \n\
 159         pushl   %edx                                            \n\
 160         pushl   %ecx                                            \n\
 161         pushl   %eax                                            \n\
 162         movl    12(%esp),%eax                                   \n\
 163         pushfl                                                  \n\
 164         pushl   %eax                                            \n\
 165         cli                                                     \n\
 166         call    " __XSTRING(CNAME(mexitcount)) "                \n\
 167         addl    $4,%esp                                         \n\
 168         popfl                                                   \n\
 169         popl    %eax                                            \n\
 170         popl    %ecx                                            \n\
 171         popl    %edx                                            \n\
 172 .mexitcount_exit:                                               \n\
 173         ret                                                     \n\
 174 ");
 175 #else /* !__GNUC__ */
 176 #error
 177 #endif /* __GNUC__ */
 178
 179 /*
 180  * Return the time elapsed since the last call.  The units are machine-
 181  * dependent.
 182  */
 183 int
 184 cputime(void)
 185 {
 186         u_int count;
 187         int delta;
 188 #if (defined(I586_CPU) || defined(I686_CPU)) && !defined(SMP) && \
 189     defined(PERFMON) && defined(I586_PMC_GUPROF)
 190         u_quad_t event_count;
 191 #endif
 192         u_char high, low;
 193         static u_int prev_count;
 194
 195 #if (defined(I586_CPU) || defined(I686_CPU)) && !defined(SMP)
 196         if (cputime_clock == CPUTIME_CLOCK_TSC) {
 197                 count = (u_int)rdtsc();
 198                 delta = (int)(count - prev_count);
 199                 prev_count = count;
 200                 return (delta);
 201         }
 202 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
 203         if (cputime_clock == CPUTIME_CLOCK_I586_PMC) {
 204                 /*
 205                  * XXX permon_read() should be inlined so that the
 206                  * perfmon module doesn't need to be compiled with
 207                  * profiling disabled and so that it is fast.
 208                  */
 209                 perfmon_read(0, &event_count);
 210
 211                 count = (u_int)event_count;
 212                 delta = (int)(count - prev_count);
 213                 prev_count = count;
 214                 return (delta);
 215         }
 216 #endif /* PERFMON && I586_PMC_GUPROF */
 217 #endif /* (I586_CPU || I686_CPU) && !SMP */
 218
 219         /*
 220          * Read the current value of the 8254 timer counter 0.
 221          */
 222         outb(TIMER_MODE, TIMER_SEL0 | TIMER_LATCH);
 223         low = inb(TIMER_CNTR0);
 224         high = inb(TIMER_CNTR0);
 225         count = ((high << 8) | low) << CPUTIME_CLOCK_I8254_SHIFT;
 226
 227         /*
 228          * The timer counts down from TIMER_CNTR0_MAX to 0 and then resets.
 229          * While profiling is enabled, this routine is called at least twice
 230          * per timer reset (for mcounting and mexitcounting hardclock()),
 231          * so at most one reset has occurred since the last call, and one
 232          * has occurred iff the current count is larger than the previous
 233          * count.  This allows counter underflow to be detected faster
 234          * than in microtime().
 235          */
 236         delta = prev_count - count;
 237         prev_count = count;
 238         if ((int) delta <= 0)
 239                 return (delta + (timer0_max_count << CPUTIME_CLOCK_I8254_SHIFT));
 240         return (delta);
 241 }
 242
 243 static int
 244 sysctl_machdep_cputime_clock(SYSCTL_HANDLER_ARGS)
 245 {
 246         int clock;
 247         int error;
 248 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
 249         int event;
 250         struct pmc pmc;
 251 #endif
 252
 253         clock = cputime_clock;
 254 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
 255         if (clock == CPUTIME_CLOCK_I586_PMC) {
 256                 pmc.pmc_val = cputime_clock_pmc_conf;
 257                 clock += pmc.pmc_event;
 258         }
 259 #endif
 260         error = sysctl_handle_opaque(oidp, &clock, sizeof clock, req);
 261         if (error == 0 && req->newptr != NULL) {
 262 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
 263                 if (clock >= CPUTIME_CLOCK_I586_PMC) {
 264                         event = clock - CPUTIME_CLOCK_I586_PMC;
 265                         if (event >= 256)
 266                                 return (EINVAL);
 267                         pmc.pmc_num = 0;
 268                         pmc.pmc_event = event;
 269                         pmc.pmc_unit = 0;
 270                         pmc.pmc_flags = PMCF_E | PMCF_OS | PMCF_USR;
 271                         pmc.pmc_mask = 0;
 272                         cputime_clock_pmc_conf = pmc.pmc_val;
 273                         cputime_clock = CPUTIME_CLOCK_I586_PMC;
 274                 } else
 275 #endif
 276                 {
 277                         if (clock < 0 || clock >= CPUTIME_CLOCK_I586_PMC)
 278                                 return (EINVAL);
 279                         cputime_clock = clock;
 280                 }
 281         }
 282         return (error);
 283 }
 284
 285 SYSCTL_PROC(_machdep, OID_AUTO, cputime_clock, CTLTYPE_INT | CTLFLAG_RW,
 286             0, sizeof(u_int), sysctl_machdep_cputime_clock, "I", "");
 287
 288 /*
 289  * The start and stop routines need not be here since we turn off profiling
 290  * before calling them.  They are here for convenience.
 291  */
 292
 293 void
 294 startguprof(struct gmonparam *gp)
 295 {
 296         if (cputime_clock == CPUTIME_CLOCK_UNINITIALIZED) {
 297                 cputime_clock = CPUTIME_CLOCK_I8254;
 298 #if (defined(I586_CPU) || defined(I686_CPU)) && !defined(SMP)
 299                 if (tsc_frequency != 0)
 300                         cputime_clock = CPUTIME_CLOCK_TSC;
 301 #endif
 302         }
 303         gp->profrate = timer_freq << CPUTIME_CLOCK_I8254_SHIFT;
 304 #if (defined(I586_CPU) || defined(I686_CPU)) && !defined(SMP)
 305         if (cputime_clock == CPUTIME_CLOCK_TSC)
 306                 gp->profrate = (u_int)tsc_frequency;    /* XXX */
 307 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
 308         else if (cputime_clock == CPUTIME_CLOCK_I586_PMC) {
 309                 if (perfmon_avail() &&
 310                     perfmon_setup(0, cputime_clock_pmc_conf) == 0) {
 311                         if (perfmon_start(0) != 0)
 312                                 perfmon_fini(0);
 313                         else {
 314                                 /* XXX 1 event == 1 us. */
 315                                 gp->profrate = 1000000;
 316
 317                                 saved_gmp = *gp;
 318
 319                                 /* Zap overheads.  They are invalid. */
 320                                 gp->cputime_overhead = 0;
 321                                 gp->mcount_overhead = 0;
 322                                 gp->mcount_post_overhead = 0;
 323                                 gp->mcount_pre_overhead = 0;
 324                                 gp->mexitcount_overhead = 0;
 325                                 gp->mexitcount_post_overhead = 0;
 326                                 gp->mexitcount_pre_overhead = 0;
 327
 328                                 cputime_clock_pmc_init = TRUE;
 329                         }
 330                 }
 331         }
 332 #endif /* PERFMON && I586_PMC_GUPROF */
 333 #endif /* (I586_CPU || I686_CPU) && !SMP */
 334         cputime_bias = 0;
 335         cputime();
 336 }
 337
 338 void
 339 stopguprof(struct gmonparam *gp)
 340 {
 341 #if defined(PERFMON) && defined(I586_PMC_GUPROF)
 342         if (cputime_clock_pmc_init) {
 343                 *gp = saved_gmp;
 344                 perfmon_fini(0);
 345                 cputime_clock_pmc_init = FALSE;
 346         }
 347 #endif
 348 }
 349
 350 #else /* !GUPROF */
 351 #ifdef __GNUC__
 352 __asm("                                                         \n\
 353         .text                                                   \n\
 354         .p2align 4,0x90                                         \n\
 355         .globl  " __XSTRING(HIDENAME(mexitcount)) "             \n\
 356 " __XSTRING(HIDENAME(mexitcount)) ":                            \n\
 357         ret                                                     \n\
 358 ");
 359 #else /* !__GNUC__ */
 360 #error
 361 #endif /* __GNUC__ */
 362 #endif /* GUPROF */