subr_prof.c revision 1.43.12.1
1 /* $NetBSD: subr_prof.c,v 1.43.12.1 2008/05/10 23:49:05 wrstuden 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 1.43.12.1 2008/05/10 23:49:05 wrstuden Exp $"); 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/kernel.h> 40 #include <sys/proc.h> 41 #include <sys/user.h> 42 #include <sys/mount.h> 43 #include <sys/sa.h> 44 #include <sys/syscallargs.h> 45 #include <sys/sysctl.h> 46 47 #include <sys/cpu.h> 48 49 #ifdef GPROF 50 #include <sys/malloc.h> 51 #include <sys/gmon.h> 52 53 MALLOC_DEFINE(M_GPROF, "gprof", "kernel profiling buffer"); 54 55 /* 56 * Froms is actually a bunch of unsigned shorts indexing tos 57 */ 58 struct gmonparam _gmonparam = { .state = GMON_PROF_OFF }; 59 60 /* Actual start of the kernel text segment. */ 61 extern char kernel_text[]; 62 63 extern char etext[]; 64 65 66 void 67 kmstartup(void) 68 { 69 char *cp; 70 struct gmonparam *p = &_gmonparam; 71 /* 72 * Round lowpc and highpc to multiples of the density we're using 73 * so the rest of the scaling (here and in gprof) stays in ints. 74 */ 75 p->lowpc = rounddown(((u_long)kernel_text), 76 HISTFRACTION * sizeof(HISTCOUNTER)); 77 p->highpc = roundup((u_long)etext, 78 HISTFRACTION * sizeof(HISTCOUNTER)); 79 p->textsize = p->highpc - p->lowpc; 80 printf("Profiling kernel, textsize=%ld [%lx..%lx]\n", 81 p->textsize, p->lowpc, p->highpc); 82 p->kcountsize = p->textsize / HISTFRACTION; 83 p->hashfraction = HASHFRACTION; 84 p->fromssize = p->textsize / HASHFRACTION; 85 p->tolimit = p->textsize * ARCDENSITY / 100; 86 if (p->tolimit < MINARCS) 87 p->tolimit = MINARCS; 88 else if (p->tolimit > MAXARCS) 89 p->tolimit = MAXARCS; 90 p->tossize = p->tolimit * sizeof(struct tostruct); 91 cp = (char *)malloc(p->kcountsize + p->fromssize + p->tossize, 92 M_GPROF, M_NOWAIT | M_ZERO); 93 if (cp == 0) { 94 printf("No memory for profiling.\n"); 95 return; 96 } 97 p->tos = (struct tostruct *)cp; 98 cp += p->tossize; 99 p->kcount = (u_short *)cp; 100 cp += p->kcountsize; 101 p->froms = (u_short *)cp; 102 } 103 104 /* 105 * Return kernel profiling information. 106 */ 107 /* 108 * sysctl helper routine for kern.profiling subtree. enables/disables 109 * kernel profiling and gives out copies of the profiling data. 110 */ 111 static int 112 sysctl_kern_profiling(SYSCTLFN_ARGS) 113 { 114 struct gmonparam *gp = &_gmonparam; 115 int error; 116 struct sysctlnode node; 117 118 node = *rnode; 119 120 switch (node.sysctl_num) { 121 case GPROF_STATE: 122 node.sysctl_data = &gp->state; 123 break; 124 case GPROF_COUNT: 125 node.sysctl_data = gp->kcount; 126 node.sysctl_size = gp->kcountsize; 127 break; 128 case GPROF_FROMS: 129 node.sysctl_data = gp->froms; 130 node.sysctl_size = gp->fromssize; 131 break; 132 case GPROF_TOS: 133 node.sysctl_data = gp->tos; 134 node.sysctl_size = gp->tossize; 135 break; 136 case GPROF_GMONPARAM: 137 node.sysctl_data = gp; 138 node.sysctl_size = sizeof(*gp); 139 break; 140 default: 141 return (EOPNOTSUPP); 142 } 143 144 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 145 if (error || newp == NULL) 146 return (error); 147 148 if (node.sysctl_num == GPROF_STATE) { 149 mutex_spin_enter(&proc0.p_stmutex); 150 if (gp->state == GMON_PROF_OFF) 151 stopprofclock(&proc0); 152 else 153 startprofclock(&proc0); 154 mutex_spin_exit(&proc0.p_stmutex); 155 } 156 157 return (0); 158 } 159 160 SYSCTL_SETUP(sysctl_kern_gprof_setup, "sysctl kern.profiling subtree setup") 161 { 162 163 sysctl_createv(clog, 0, NULL, NULL, 164 CTLFLAG_PERMANENT, 165 CTLTYPE_NODE, "kern", NULL, 166 NULL, 0, NULL, 0, 167 CTL_KERN, CTL_EOL); 168 sysctl_createv(clog, 0, NULL, NULL, 169 CTLFLAG_PERMANENT, 170 CTLTYPE_NODE, "profiling", 171 SYSCTL_DESCR("Profiling information (available)"), 172 NULL, 0, NULL, 0, 173 CTL_KERN, KERN_PROF, CTL_EOL); 174 175 sysctl_createv(clog, 0, NULL, NULL, 176 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 177 CTLTYPE_INT, "state", 178 SYSCTL_DESCR("Profiling state"), 179 sysctl_kern_profiling, 0, NULL, 0, 180 CTL_KERN, KERN_PROF, GPROF_STATE, CTL_EOL); 181 sysctl_createv(clog, 0, NULL, NULL, 182 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 183 CTLTYPE_STRUCT, "count", 184 SYSCTL_DESCR("Array of statistical program counters"), 185 sysctl_kern_profiling, 0, NULL, 0, 186 CTL_KERN, KERN_PROF, GPROF_COUNT, CTL_EOL); 187 sysctl_createv(clog, 0, NULL, NULL, 188 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 189 CTLTYPE_STRUCT, "froms", 190 SYSCTL_DESCR("Array indexed by program counter of " 191 "call-from points"), 192 sysctl_kern_profiling, 0, NULL, 0, 193 CTL_KERN, KERN_PROF, GPROF_FROMS, CTL_EOL); 194 sysctl_createv(clog, 0, NULL, NULL, 195 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 196 CTLTYPE_STRUCT, "tos", 197 SYSCTL_DESCR("Array of structures describing " 198 "destination of calls and their counts"), 199 sysctl_kern_profiling, 0, NULL, 0, 200 CTL_KERN, KERN_PROF, GPROF_TOS, CTL_EOL); 201 sysctl_createv(clog, 0, NULL, NULL, 202 CTLFLAG_PERMANENT, 203 CTLTYPE_STRUCT, "gmonparam", 204 SYSCTL_DESCR("Structure giving the sizes of the above " 205 "arrays"), 206 sysctl_kern_profiling, 0, NULL, 0, 207 CTL_KERN, KERN_PROF, GPROF_GMONPARAM, CTL_EOL); 208 } 209 #endif /* GPROF */ 210 211 /* 212 * Profiling system call. 213 * 214 * The scale factor is a fixed point number with 16 bits of fraction, so that 215 * 1.0 is represented as 0x10000. A scale factor of 0 turns off profiling. 216 */ 217 /* ARGSUSED */ 218 int 219 sys_profil(struct lwp *l, const struct sys_profil_args *uap, register_t *retval) 220 { 221 /* { 222 syscallarg(char *) samples; 223 syscallarg(u_int) size; 224 syscallarg(u_int) offset; 225 syscallarg(u_int) scale; 226 } */ 227 struct proc *p = l->l_proc; 228 struct uprof *upp; 229 230 if (SCARG(uap, scale) > (1 << 16)) 231 return (EINVAL); 232 if (SCARG(uap, scale) == 0) { 233 mutex_spin_enter(&p->p_stmutex); 234 stopprofclock(p); 235 mutex_spin_exit(&p->p_stmutex); 236 return (0); 237 } 238 upp = &p->p_stats->p_prof; 239 240 /* Block profile interrupts while changing state. */ 241 mutex_spin_enter(&p->p_stmutex); 242 upp->pr_off = SCARG(uap, offset); 243 upp->pr_scale = SCARG(uap, scale); 244 upp->pr_base = SCARG(uap, samples); 245 upp->pr_size = SCARG(uap, size); 246 startprofclock(p); 247 mutex_spin_exit(&p->p_stmutex); 248 249 return (0); 250 } 251 252 /* 253 * Scale is a fixed-point number with the binary point 16 bits 254 * into the value, and is <= 1.0. pc is at most 32 bits, so the 255 * intermediate result is at most 48 bits. 256 */ 257 #define PC_TO_INDEX(pc, prof) \ 258 ((int)(((u_quad_t)((pc) - (prof)->pr_off) * \ 259 (u_quad_t)((prof)->pr_scale)) >> 16) & ~1) 260 261 /* 262 * Collect user-level profiling statistics; called on a profiling tick, 263 * when a process is running in user-mode. This routine may be called 264 * from an interrupt context. We try to update the user profiling buffers 265 * cheaply with fuswintr() and suswintr(). If that fails, we revert to 266 * an AST that will vector us to trap() with a context in which copyin 267 * and copyout will work. Trap will then call addupc_task(). 268 * 269 * Note that we may (rarely) not get around to the AST soon enough, and 270 * lose profile ticks when the next tick overwrites this one, but in this 271 * case the system is overloaded and the profile is probably already 272 * inaccurate. 273 */ 274 void 275 addupc_intr(struct lwp *l, u_long pc) 276 { 277 struct uprof *prof; 278 struct proc *p; 279 void *addr; 280 u_int i; 281 int v; 282 283 p = l->l_proc; 284 285 KASSERT(mutex_owned(&p->p_stmutex)); 286 287 prof = &p->p_stats->p_prof; 288 if (pc < prof->pr_off || 289 (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size) 290 return; /* out of range; ignore */ 291 292 addr = prof->pr_base + i; 293 mutex_spin_exit(&p->p_stmutex); 294 if ((v = fuswintr(addr)) == -1 || suswintr(addr, v + 1) == -1) { 295 /* XXXSMP */ 296 prof->pr_addr = pc; 297 prof->pr_ticks++; 298 cpu_need_proftick(l); 299 } 300 mutex_spin_enter(&p->p_stmutex); 301 } 302 303 /* 304 * Much like before, but we can afford to take faults here. If the 305 * update fails, we simply turn off profiling. 306 */ 307 void 308 addupc_task(struct lwp *l, u_long pc, u_int ticks) 309 { 310 struct uprof *prof; 311 struct proc *p; 312 void *addr; 313 int error; 314 u_int i; 315 u_short v; 316 317 p = l->l_proc; 318 319 if (ticks == 0) 320 return; 321 322 mutex_spin_enter(&p->p_stmutex); 323 prof = &p->p_stats->p_prof; 324 325 /* Testing P_PROFIL may be unnecessary, but is certainly safe. */ 326 if ((p->p_stflag & PST_PROFIL) == 0 || pc < prof->pr_off || 327 (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size) { 328 mutex_spin_exit(&p->p_stmutex); 329 return; 330 } 331 332 addr = prof->pr_base + i; 333 mutex_spin_exit(&p->p_stmutex); 334 if ((error = copyin(addr, (void *)&v, sizeof(v))) == 0) { 335 v += ticks; 336 error = copyout((void *)&v, addr, sizeof(v)); 337 } 338 if (error != 0) { 339 mutex_spin_enter(&p->p_stmutex); 340 stopprofclock(p); 341 mutex_spin_exit(&p->p_stmutex); 342 } 343 } 344
Indexes created Tue Sep 30 20:09:53 GMT 2025