subr_prof.c revision 1.33.20.8
1 /* $NetBSD: subr_prof.c,v 1.33.20.8 2007/01/31 23:38:38 ad 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 1.33.20.8 2007/01/31 23:38:38 ad Exp $"); 36 37 #include "opt_lockdebug.h" 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/kernel.h> 42 #include <sys/proc.h> 43 #include <sys/user.h> 44 #include <sys/mount.h> 45 #include <sys/syscallargs.h> 46 #include <sys/sysctl.h> 47 48 #include <machine/cpu.h> 49 50 #ifdef GPROF 51 #include <sys/malloc.h> 52 #include <sys/gmon.h> 53 54 MALLOC_DEFINE(M_GPROF, "gprof", "kernel profiling buffer"); 55 56 #ifdef LOCKDEBUG 57 #error The LOCKDEBUG option can not be used with a profiling kernel 58 #endif 59 60 /* 61 * Froms is actually a bunch of unsigned shorts indexing tos 62 */ 63 struct gmonparam _gmonparam = { .state = GMON_PROF_OFF }; 64 65 /* Actual start of the kernel text segment. */ 66 extern char kernel_text[]; 67 68 extern char etext[]; 69 70 71 void 72 kmstartup(void) 73 { 74 char *cp; 75 struct gmonparam *p = &_gmonparam; 76 /* 77 * Round lowpc and highpc to multiples of the density we're using 78 * so the rest of the scaling (here and in gprof) stays in ints. 79 */ 80 p->lowpc = rounddown(((u_long)kernel_text), 81 HISTFRACTION * sizeof(HISTCOUNTER)); 82 p->highpc = roundup((u_long)etext, 83 HISTFRACTION * sizeof(HISTCOUNTER)); 84 p->textsize = p->highpc - p->lowpc; 85 printf("Profiling kernel, textsize=%ld [%lx..%lx]\n", 86 p->textsize, p->lowpc, p->highpc); 87 p->kcountsize = p->textsize / HISTFRACTION; 88 p->hashfraction = HASHFRACTION; 89 p->fromssize = p->textsize / HASHFRACTION; 90 p->tolimit = p->textsize * ARCDENSITY / 100; 91 if (p->tolimit < MINARCS) 92 p->tolimit = MINARCS; 93 else if (p->tolimit > MAXARCS) 94 p->tolimit = MAXARCS; 95 p->tossize = p->tolimit * sizeof(struct tostruct); 96 cp = (char *)malloc(p->kcountsize + p->fromssize + p->tossize, 97 M_GPROF, M_NOWAIT | M_ZERO); 98 if (cp == 0) { 99 printf("No memory for profiling.\n"); 100 return; 101 } 102 p->tos = (struct tostruct *)cp; 103 cp += p->tossize; 104 p->kcount = (u_short *)cp; 105 cp += p->kcountsize; 106 p->froms = (u_short *)cp; 107 } 108 109 /* 110 * Return kernel profiling information. 111 */ 112 /* 113 * sysctl helper routine for kern.profiling subtree. enables/disables 114 * kernel profiling and gives out copies of the profiling data. 115 */ 116 static int 117 sysctl_kern_profiling(SYSCTLFN_ARGS) 118 { 119 struct gmonparam *gp = &_gmonparam; 120 int error; 121 struct sysctlnode node; 122 123 node = *rnode; 124 125 switch (node.sysctl_num) { 126 case GPROF_STATE: 127 node.sysctl_data = &gp->state; 128 break; 129 case GPROF_COUNT: 130 node.sysctl_data = gp->kcount; 131 node.sysctl_size = gp->kcountsize; 132 break; 133 case GPROF_FROMS: 134 node.sysctl_data = gp->froms; 135 node.sysctl_size = gp->fromssize; 136 break; 137 case GPROF_TOS: 138 node.sysctl_data = gp->tos; 139 node.sysctl_size = gp->tossize; 140 break; 141 case GPROF_GMONPARAM: 142 node.sysctl_data = gp; 143 node.sysctl_size = sizeof(*gp); 144 break; 145 default: 146 return (EOPNOTSUPP); 147 } 148 149 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 150 if (error || newp == NULL) 151 return (error); 152 153 if (node.sysctl_num == GPROF_STATE) { 154 mutex_enter(&proc0.p_stmutex); 155 if (gp->state == GMON_PROF_OFF) 156 stopprofclock(&proc0); 157 else 158 startprofclock(&proc0); 159 mutex_exit(&proc0.p_stmutex); 160 } 161 162 return (0); 163 } 164 165 SYSCTL_SETUP(sysctl_kern_gprof_setup, "sysctl kern.profiling subtree setup") 166 { 167 168 sysctl_createv(clog, 0, NULL, NULL, 169 CTLFLAG_PERMANENT, 170 CTLTYPE_NODE, "kern", NULL, 171 NULL, 0, NULL, 0, 172 CTL_KERN, CTL_EOL); 173 sysctl_createv(clog, 0, NULL, NULL, 174 CTLFLAG_PERMANENT, 175 CTLTYPE_NODE, "profiling", 176 SYSCTL_DESCR("Profiling information (available)"), 177 NULL, 0, NULL, 0, 178 CTL_KERN, KERN_PROF, CTL_EOL); 179 180 sysctl_createv(clog, 0, NULL, NULL, 181 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 182 CTLTYPE_INT, "state", 183 SYSCTL_DESCR("Profiling state"), 184 sysctl_kern_profiling, 0, NULL, 0, 185 CTL_KERN, KERN_PROF, GPROF_STATE, CTL_EOL); 186 sysctl_createv(clog, 0, NULL, NULL, 187 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 188 CTLTYPE_STRUCT, "count", 189 SYSCTL_DESCR("Array of statistical program counters"), 190 sysctl_kern_profiling, 0, NULL, 0, 191 CTL_KERN, KERN_PROF, GPROF_COUNT, CTL_EOL); 192 sysctl_createv(clog, 0, NULL, NULL, 193 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 194 CTLTYPE_STRUCT, "froms", 195 SYSCTL_DESCR("Array indexed by program counter of " 196 "call-from points"), 197 sysctl_kern_profiling, 0, NULL, 0, 198 CTL_KERN, KERN_PROF, GPROF_FROMS, CTL_EOL); 199 sysctl_createv(clog, 0, NULL, NULL, 200 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 201 CTLTYPE_STRUCT, "tos", 202 SYSCTL_DESCR("Array of structures describing " 203 "destination of calls and their counts"), 204 sysctl_kern_profiling, 0, NULL, 0, 205 CTL_KERN, KERN_PROF, GPROF_TOS, CTL_EOL); 206 sysctl_createv(clog, 0, NULL, NULL, 207 CTLFLAG_PERMANENT, 208 CTLTYPE_STRUCT, "gmonparam", 209 SYSCTL_DESCR("Structure giving the sizes of the above " 210 "arrays"), 211 sysctl_kern_profiling, 0, NULL, 0, 212 CTL_KERN, KERN_PROF, GPROF_GMONPARAM, CTL_EOL); 213 } 214 #endif /* GPROF */ 215 216 /* 217 * Profiling system call. 218 * 219 * The scale factor is a fixed point number with 16 bits of fraction, so that 220 * 1.0 is represented as 0x10000. A scale factor of 0 turns off profiling. 221 */ 222 /* ARGSUSED */ 223 int 224 sys_profil(struct lwp *l, void *v, register_t *retval) 225 { 226 struct sys_profil_args /* { 227 syscallarg(caddr_t) samples; 228 syscallarg(u_int) size; 229 syscallarg(u_int) offset; 230 syscallarg(u_int) scale; 231 } */ *uap = v; 232 struct proc *p = l->l_proc; 233 struct uprof *upp; 234 235 if (SCARG(uap, scale) > (1 << 16)) 236 return (EINVAL); 237 if (SCARG(uap, scale) == 0) { 238 mutex_enter(&p->p_stmutex); 239 stopprofclock(p); 240 mutex_exit(&p->p_stmutex); 241 return (0); 242 } 243 upp = &p->p_stats->p_prof; 244 245 /* Block profile interrupts while changing state. */ 246 mutex_enter(&p->p_stmutex); 247 upp->pr_off = SCARG(uap, offset); 248 upp->pr_scale = SCARG(uap, scale); 249 upp->pr_base = SCARG(uap, samples); 250 upp->pr_size = SCARG(uap, size); 251 startprofclock(p); 252 mutex_exit(&p->p_stmutex); 253 254 return (0); 255 } 256 257 /* 258 * Scale is a fixed-point number with the binary point 16 bits 259 * into the value, and is <= 1.0. pc is at most 32 bits, so the 260 * intermediate result is at most 48 bits. 261 */ 262 #define PC_TO_INDEX(pc, prof) \ 263 ((int)(((u_quad_t)((pc) - (prof)->pr_off) * \ 264 (u_quad_t)((prof)->pr_scale)) >> 16) & ~1) 265 266 /* 267 * Collect user-level profiling statistics; called on a profiling tick, 268 * when a process is running in user-mode. This routine may be called 269 * from an interrupt context. We try to update the user profiling buffers 270 * cheaply with fuswintr() and suswintr(). If that fails, we revert to 271 * an AST that will vector us to trap() with a context in which copyin 272 * and copyout will work. Trap will then call addupc_task(). 273 * 274 * Note that we may (rarely) not get around to the AST soon enough, and 275 * lose profile ticks when the next tick overwrites this one, but in this 276 * case the system is overloaded and the profile is probably already 277 * inaccurate. 278 */ 279 void 280 addupc_intr(struct lwp *l, u_long pc) 281 { 282 struct uprof *prof; 283 struct proc *p; 284 caddr_t addr; 285 u_int i; 286 int v; 287 288 p = l->l_proc; 289 290 LOCK_ASSERT(mutex_owned(&p->p_stmutex)); 291 292 prof = &p->p_stats->p_prof; 293 if (pc < prof->pr_off || 294 (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size) 295 return; /* out of range; ignore */ 296 297 addr = prof->pr_base + i; 298 mutex_exit(&p->p_stmutex); 299 if ((v = fuswintr(addr)) == -1 || suswintr(addr, v + 1) == -1) { 300 /* XXXSMP */ 301 prof->pr_addr = pc; 302 prof->pr_ticks++; 303 cpu_need_proftick(l); 304 } 305 mutex_enter(&p->p_stmutex); 306 } 307 308 /* 309 * Much like before, but we can afford to take faults here. If the 310 * update fails, we simply turn off profiling. 311 */ 312 void 313 addupc_task(struct lwp *l, u_long pc, u_int ticks) 314 { 315 struct uprof *prof; 316 struct proc *p; 317 caddr_t addr; 318 int error; 319 u_int i; 320 u_short v; 321 322 p = l->l_proc; 323 324 if (ticks == 0) 325 return; 326 327 mutex_enter(&p->p_stmutex); 328 prof = &p->p_stats->p_prof; 329 330 /* Testing P_PROFIL may be unnecessary, but is certainly safe. */ 331 if ((p->p_stflag & PST_PROFIL) == 0 || pc < prof->pr_off || 332 (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size) { 333 mutex_exit(&p->p_stmutex); 334 return; 335 } 336 337 addr = prof->pr_base + i; 338 mutex_exit(&p->p_stmutex); 339 if ((error = copyin(addr, (caddr_t)&v, sizeof(v))) == 0) { 340 v += ticks; 341 error = copyout((caddr_t)&v, addr, sizeof(v)); 342 } 343 if (error != 0) { 344 mutex_enter(&p->p_stmutex); 345 stopprofclock(p); 346 mutex_exit(&p->p_stmutex); 347 } 348 } 349
Indexes created Tue Sep 30 20:09:53 GMT 2025