subr_prof.c revision 1.33.20.2
1 /* $NetBSD: subr_prof.c,v 1.33.20.2 2006/11/17 16:34:37 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 1.33.20.2 2006/11/17 16:34:37 ad 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 <machine/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 = { 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); 93 if (cp == 0) { 94 printf("No memory for profiling.\n"); 95 return; 96 } 97 memset(cp, 0, p->kcountsize + p->tossize + p->fromssize); 98 p->tos = (struct tostruct *)cp; 99 cp += p->tossize; 100 p->kcount = (u_short *)cp; 101 cp += p->kcountsize; 102 p->froms = (u_short *)cp; 103 } 104 105 /* 106 * Return kernel profiling information. 107 */ 108 /* 109 * sysctl helper routine for kern.profiling subtree. enables/disables 110 * kernel profiling and gives out copies of the profiling data. 111 */ 112 static int 113 sysctl_kern_profiling(SYSCTLFN_ARGS) 114 { 115 struct gmonparam *gp = &_gmonparam; 116 int error; 117 struct sysctlnode node; 118 119 node = *rnode; 120 121 switch (node.sysctl_num) { 122 case GPROF_STATE: 123 node.sysctl_data = &gp->state; 124 break; 125 case GPROF_COUNT: 126 node.sysctl_data = gp->kcount; 127 node.sysctl_size = gp->kcountsize; 128 break; 129 case GPROF_FROMS: 130 node.sysctl_data = gp->froms; 131 node.sysctl_size = gp->fromssize; 132 break; 133 case GPROF_TOS: 134 node.sysctl_data = gp->tos; 135 node.sysctl_size = gp->tossize; 136 break; 137 case GPROF_GMONPARAM: 138 node.sysctl_data = gp; 139 node.sysctl_size = sizeof(*gp); 140 break; 141 default: 142 return (EOPNOTSUPP); 143 } 144 145 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 146 if (error || newp == NULL) 147 return (error); 148 149 if (node.sysctl_num == GPROF_STATE) { 150 mutex_enter(proc0.p_smutex); 151 if (gp->state == GMON_PROF_OFF) 152 stopprofclock(&proc0); 153 else 154 startprofclock(&proc0); 155 mutex_exit(proc0.p_smutex); 156 } 157 158 return (0); 159 } 160 161 SYSCTL_SETUP(sysctl_kern_gprof_setup, "sysctl kern.profiling subtree setup") 162 { 163 164 sysctl_createv(clog, 0, NULL, NULL, 165 CTLFLAG_PERMANENT, 166 CTLTYPE_NODE, "kern", NULL, 167 NULL, 0, NULL, 0, 168 CTL_KERN, CTL_EOL); 169 sysctl_createv(clog, 0, NULL, NULL, 170 CTLFLAG_PERMANENT, 171 CTLTYPE_NODE, "profiling", 172 SYSCTL_DESCR("Profiling information (available)"), 173 NULL, 0, NULL, 0, 174 CTL_KERN, KERN_PROF, CTL_EOL); 175 176 sysctl_createv(clog, 0, NULL, NULL, 177 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 178 CTLTYPE_INT, "state", 179 SYSCTL_DESCR("Profiling state"), 180 sysctl_kern_profiling, 0, NULL, 0, 181 CTL_KERN, KERN_PROF, GPROF_STATE, CTL_EOL); 182 sysctl_createv(clog, 0, NULL, NULL, 183 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 184 CTLTYPE_STRUCT, "count", 185 SYSCTL_DESCR("Array of statistical program counters"), 186 sysctl_kern_profiling, 0, NULL, 0, 187 CTL_KERN, KERN_PROF, GPROF_COUNT, CTL_EOL); 188 sysctl_createv(clog, 0, NULL, NULL, 189 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 190 CTLTYPE_STRUCT, "froms", 191 SYSCTL_DESCR("Array indexed by program counter of " 192 "call-from points"), 193 sysctl_kern_profiling, 0, NULL, 0, 194 CTL_KERN, KERN_PROF, GPROF_FROMS, CTL_EOL); 195 sysctl_createv(clog, 0, NULL, NULL, 196 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 197 CTLTYPE_STRUCT, "tos", 198 SYSCTL_DESCR("Array of structures describing " 199 "destination of calls and their counts"), 200 sysctl_kern_profiling, 0, NULL, 0, 201 CTL_KERN, KERN_PROF, GPROF_TOS, CTL_EOL); 202 sysctl_createv(clog, 0, NULL, NULL, 203 CTLFLAG_PERMANENT, 204 CTLTYPE_STRUCT, "gmonparam", 205 SYSCTL_DESCR("Structure giving the sizes of the above " 206 "arrays"), 207 sysctl_kern_profiling, 0, NULL, 0, 208 CTL_KERN, KERN_PROF, GPROF_GMONPARAM, CTL_EOL); 209 } 210 #endif /* GPROF */ 211 212 /* 213 * Profiling system call. 214 * 215 * The scale factor is a fixed point number with 16 bits of fraction, so that 216 * 1.0 is represented as 0x10000. A scale factor of 0 turns off profiling. 217 */ 218 /* ARGSUSED */ 219 int 220 sys_profil(struct lwp *l, void *v, register_t *retval) 221 { 222 struct sys_profil_args /* { 223 syscallarg(caddr_t) samples; 224 syscallarg(u_int) size; 225 syscallarg(u_int) offset; 226 syscallarg(u_int) scale; 227 } */ *uap = v; 228 struct proc *p = l->l_proc; 229 struct uprof *upp; 230 int s; 231 232 if (SCARG(uap, scale) > (1 << 16)) 233 return (EINVAL); 234 if (SCARG(uap, scale) == 0) { 235 mutex_enter(&p->p_smutex); 236 stopprofclock(p); 237 mutex_exit(&p->p_smutex); 238 return (0); 239 } 240 upp = &p->p_stats->p_prof; 241 242 /* Block profile interrupts while changing state. */ 243 mutex_enter(&p->p_smutex); 244 s = splstatclock(); /* XXX */ 245 upp->pr_off = SCARG(uap, offset); 246 upp->pr_scale = SCARG(uap, scale); 247 upp->pr_base = SCARG(uap, samples); 248 upp->pr_size = SCARG(uap, size); 249 startprofclock(p); 250 splx(s); 251 mutex_exit(&p->p_smutex); 252 253 return (0); 254 } 255 256 /* 257 * Scale is a fixed-point number with the binary point 16 bits 258 * into the value, and is <= 1.0. pc is at most 32 bits, so the 259 * intermediate result is at most 48 bits. 260 */ 261 #define PC_TO_INDEX(pc, prof) \ 262 ((int)(((u_quad_t)((pc) - (prof)->pr_off) * \ 263 (u_quad_t)((prof)->pr_scale)) >> 16) & ~1) 264 265 /* 266 * Collect user-level profiling statistics; called on a profiling tick, 267 * when a process is running in user-mode. This routine may be called 268 * from an interrupt context. We try to update the user profiling buffers 269 * cheaply with fuswintr() and suswintr(). If that fails, we revert to 270 * an AST that will vector us to trap() with a context in which copyin 271 * and copyout will work. Trap will then call addupc_task(). 272 * 273 * Note that we may (rarely) not get around to the AST soon enough, and 274 * lose profile ticks when the next tick overwrites this one, but in this 275 * case the system is overloaded and the profile is probably already 276 * inaccurate. 277 */ 278 void 279 addupc_intr(struct lwp *l, u_long pc) 280 { 281 struct uprof *prof; 282 struct proc *p; 283 caddr_t addr; 284 u_int i; 285 int v; 286 287 p = l->l_proc; 288 289 LOCK_ASSERT(mutex_owned(&p->p_smutex)); 290 291 prof = &p->p_stats->p_prof; 292 if (pc < prof->pr_off || 293 (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size) 294 return; /* out of range; ignore */ 295 296 addr = prof->pr_base + i; 297 mutex_exit(&p->p_smutex); 298 if ((v = fuswintr(addr)) == -1 || suswintr(addr, v + 1) == -1) { 299 prof->pr_addr = pc; 300 prof->pr_ticks++; 301 cpu_need_proftick(l); 302 } 303 mutex_enter(&p->p_smutex); 304 } 305 306 /* 307 * Much like before, but we can afford to take faults here. If the 308 * update fails, we simply turn off profiling. 309 * 310 * XXXSMP unlocked 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 u_int i; 319 u_short v; 320 321 p = l->l_proc; 322 323 /* Testing P_PROFIL may be unnecessary, but is certainly safe. */ 324 if ((p->p_sflag & PS_PROFIL) == 0 || ticks == 0) 325 return; 326 327 prof = &p->p_stats->p_prof; 328 if (pc < prof->pr_off || 329 (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size) 330 return; 331 332 addr = prof->pr_base + i; 333 if (copyin(addr, (caddr_t)&v, sizeof(v)) == 0) { 334 v += ticks; 335 if (copyout((caddr_t)&v, addr, sizeof(v)) == 0) 336 return; 337 } 338 339 mutex_enter(&p->p_smutex); 340 stopprofclock(p); 341 mutex_exit(&p->p_smutex); 342 } 343
Indexes created Tue Sep 30 20:09:53 GMT 2025