1 /* $NetBSD: subr_prof.c,v 1.27 2003/02/01 06:23:44 thorpej 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. All advertising materials mentioning features or use of this software 16 * must display the following acknowledgement: 17 * This product includes software developed by the University of 18 * California, Berkeley and its contributors. 19 * 4. Neither the name of the University nor the names of its contributors 20 * may be used to endorse or promote products derived from this software 21 * without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 * 35 * @(#)subr_prof.c 8.4 (Berkeley) 2/14/95 36 */ 37 38 #include <sys/cdefs.h> 39 __KERNEL_RCSID(0, "$NetBSD: subr_prof.c,v 1.27 2003/02/01 06:23:44 thorpej Exp $"); 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/kernel.h> 44 #include <sys/proc.h> 45 #include <sys/user.h> 46 #include <sys/mount.h> 47 #include <sys/sa.h> 48 #include <sys/syscallargs.h> 49 #include <sys/sysctl.h> 50 51 #include <machine/cpu.h> 52 53 #ifdef GPROF 54 #include <sys/malloc.h> 55 #include <sys/gmon.h> 56 57 MALLOC_DEFINE(M_GPROF, "gprof", "kernel profiling buffer"); 58 59 /* 60 * Froms is actually a bunch of unsigned shorts indexing tos 61 */ 62 struct gmonparam _gmonparam = { GMON_PROF_OFF }; 63 64 /* Actual start of the kernel text segment. */ 65 extern char kernel_text[]; 66 67 extern char etext[]; 68 69 70 void 71 kmstartup() 72 { 73 char *cp; 74 struct gmonparam *p = &_gmonparam; 75 /* 76 * Round lowpc and highpc to multiples of the density we're using 77 * so the rest of the scaling (here and in gprof) stays in ints. 78 */ 79 p->lowpc = ROUNDDOWN(((u_long)kernel_text), 80 HISTFRACTION * sizeof(HISTCOUNTER)); 81 p->highpc = ROUNDUP((u_long)etext, 82 HISTFRACTION * sizeof(HISTCOUNTER)); 83 p->textsize = p->highpc - p->lowpc; 84 printf("Profiling kernel, textsize=%ld [%lx..%lx]\n", 85 p->textsize, p->lowpc, p->highpc); 86 p->kcountsize = p->textsize / HISTFRACTION; 87 p->hashfraction = HASHFRACTION; 88 p->fromssize = p->textsize / HASHFRACTION; 89 p->tolimit = p->textsize * ARCDENSITY / 100; 90 if (p->tolimit < MINARCS) 91 p->tolimit = MINARCS; 92 else if (p->tolimit > MAXARCS) 93 p->tolimit = MAXARCS; 94 p->tossize = p->tolimit * sizeof(struct tostruct); 95 cp = (char *)malloc(p->kcountsize + p->fromssize + p->tossize, 96 M_GPROF, M_NOWAIT); 97 if (cp == 0) { 98 printf("No memory for profiling.\n"); 99 return; 100 } 101 memset(cp, 0, p->kcountsize + p->tossize + p->fromssize); 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 int 113 sysctl_doprof(name, namelen, oldp, oldlenp, newp, newlen) 114 int *name; 115 u_int namelen; 116 void *oldp; 117 size_t *oldlenp; 118 void *newp; 119 size_t newlen; 120 { 121 struct gmonparam *gp = &_gmonparam; 122 int error; 123 124 /* all sysctl names at this level are terminal */ 125 if (namelen != 1) 126 return (ENOTDIR); /* overloaded */ 127 128 /* Check we got the necessary memory at startup. */ 129 if (gp->kcount == NULL) 130 return (EOPNOTSUPP); 131 132 switch (name[0]) { 133 case GPROF_STATE: 134 error = sysctl_int(oldp, oldlenp, newp, newlen, &gp->state); 135 if (error) 136 return (error); 137 if (gp->state == GMON_PROF_OFF) 138 stopprofclock(&proc0); 139 else 140 startprofclock(&proc0); 141 return (0); 142 case GPROF_COUNT: 143 return (sysctl_struct(oldp, oldlenp, newp, newlen, 144 gp->kcount, gp->kcountsize)); 145 case GPROF_FROMS: 146 return (sysctl_struct(oldp, oldlenp, newp, newlen, 147 gp->froms, gp->fromssize)); 148 case GPROF_TOS: 149 return (sysctl_struct(oldp, oldlenp, newp, newlen, 150 gp->tos, gp->tossize)); 151 case GPROF_GMONPARAM: 152 return (sysctl_rdstruct(oldp, oldlenp, newp, gp, sizeof(*gp))); 153 default: 154 return (EOPNOTSUPP); 155 } 156 /* NOTREACHED */ 157 } 158 #endif /* GPROF */ 159 160 /* 161 * Profiling system call. 162 * 163 * The scale factor is a fixed point number with 16 bits of fraction, so that 164 * 1.0 is represented as 0x10000. A scale factor of 0 turns off profiling. 165 */ 166 /* ARGSUSED */ 167 int 168 sys_profil(l, v, retval) 169 struct lwp *l; 170 void *v; 171 register_t *retval; 172 { 173 struct sys_profil_args /* { 174 syscallarg(caddr_t) samples; 175 syscallarg(u_int) size; 176 syscallarg(u_int) offset; 177 syscallarg(u_int) scale; 178 } */ *uap = v; 179 struct proc *p = l->l_proc; 180 struct uprof *upp; 181 int s; 182 183 if (SCARG(uap, scale) > (1 << 16)) 184 return (EINVAL); 185 if (SCARG(uap, scale) == 0) { 186 stopprofclock(p); 187 return (0); 188 } 189 upp = &p->p_stats->p_prof; 190 191 /* Block profile interrupts while changing state. */ 192 s = splstatclock(); 193 upp->pr_off = SCARG(uap, offset); 194 upp->pr_scale = SCARG(uap, scale); 195 upp->pr_base = SCARG(uap, samples); 196 upp->pr_size = SCARG(uap, size); 197 startprofclock(p); 198 splx(s); 199 200 return (0); 201 } 202 203 /* 204 * Scale is a fixed-point number with the binary point 16 bits 205 * into the value, and is <= 1.0. pc is at most 32 bits, so the 206 * intermediate result is at most 48 bits. 207 */ 208 #define PC_TO_INDEX(pc, prof) \ 209 ((int)(((u_quad_t)((pc) - (prof)->pr_off) * \ 210 (u_quad_t)((prof)->pr_scale)) >> 16) & ~1) 211 212 /* 213 * Collect user-level profiling statistics; called on a profiling tick, 214 * when a process is running in user-mode. This routine may be called 215 * from an interrupt context. We try to update the user profiling buffers 216 * cheaply with fuswintr() and suswintr(). If that fails, we revert to 217 * an AST that will vector us to trap() with a context in which copyin 218 * and copyout will work. Trap will then call addupc_task(). 219 * 220 * Note that we may (rarely) not get around to the AST soon enough, and 221 * lose profile ticks when the next tick overwrites this one, but in this 222 * case the system is overloaded and the profile is probably already 223 * inaccurate. 224 */ 225 void 226 addupc_intr(p, pc) 227 struct proc *p; 228 u_long pc; 229 { 230 struct uprof *prof; 231 caddr_t addr; 232 u_int i; 233 int v; 234 235 prof = &p->p_stats->p_prof; 236 if (pc < prof->pr_off || 237 (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size) 238 return; /* out of range; ignore */ 239 240 addr = prof->pr_base + i; 241 if ((v = fuswintr(addr)) == -1 || suswintr(addr, v + 1) == -1) { 242 prof->pr_addr = pc; 243 prof->pr_ticks++; 244 need_proftick(p); 245 } 246 } 247 248 /* 249 * Much like before, but we can afford to take faults here. If the 250 * update fails, we simply turn off profiling. 251 */ 252 void 253 addupc_task(p, pc, ticks) 254 struct proc *p; 255 u_long pc; 256 u_int ticks; 257 { 258 struct uprof *prof; 259 caddr_t addr; 260 u_int i; 261 u_short v; 262 263 /* Testing P_PROFIL may be unnecessary, but is certainly safe. */ 264 if ((p->p_flag & P_PROFIL) == 0 || ticks == 0) 265 return; 266 267 prof = &p->p_stats->p_prof; 268 if (pc < prof->pr_off || 269 (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size) 270 return; 271 272 addr = prof->pr_base + i; 273 if (copyin(addr, (caddr_t)&v, sizeof(v)) == 0) { 274 v += ticks; 275 if (copyout((caddr_t)&v, addr, sizeof(v)) == 0) 276 return; 277 } 278 stopprofclock(p); 279 } 280