1 /*- 2 * Copyright (c) 1990 The Regents of the University of California. 3 * All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * the University of Utah, and William Jolitz. 7 * 8 * %sccs.include.redist.c% 9 * 10 * @(#)trap.c 7.6 (Berkeley) 11/20/91 11 */ 12 13 /* 14 * 386 Trap and System call handleing 15 */ 16 17 #include "machine/cpu.h" 18 #include "machine/psl.h" 19 #include "machine/reg.h" 20 21 #include "param.h" 22 #include "systm.h" 23 #include "proc.h" 24 #include "user.h" 25 #include "acct.h" 26 #include "kernel.h" 27 #ifdef KTRACE 28 #include "ktrace.h" 29 #endif 30 31 #include "vm/vm_param.h" 32 #include "vm/pmap.h" 33 #include "vm/vm_map.h" 34 35 #include "machine/trap.h" 36 #include "machine/dbg.h" 37 38 39 struct sysent sysent[]; 40 int nsysent; 41 unsigned rcr2(); 42 extern short cpl; 43 44 45 /* 46 * trap(frame): 47 * Exception, fault, and trap interface to BSD kernel. This 48 * common code is called from assembly language IDT gate entry 49 * routines that prepare a suitable stack frame, and restore this 50 * frame after the exception has been processed. Note that the 51 * effect is as if the arguments were passed call by reference. 52 */ 53 54 /*ARGSUSED*/ 55 trap(frame) 56 struct trapframe frame; 57 { 58 register int i; 59 register struct proc *p = curproc; 60 struct timeval syst; 61 int ucode, type, code, eva; 62 extern int cold; 63 64 if(cold) goto we_re_toast; 65 frame.tf_eflags &= ~PSL_NT; /* clear nested trap XXX */ 66 type = frame.tf_trapno; 67 68 if (curpcb->pcb_onfault && frame.tf_trapno != 0xc) { 69 frame.tf_eip = (int)curpcb->pcb_onfault; 70 return; 71 } 72 73 syst = p->p_stime; 74 if (ISPL(frame.tf_cs) == SEL_UPL) { 75 type |= T_USER; 76 p->p_regs = (int *)&frame; 77 curpcb->pcb_flags |= FM_TRAP; /* used by sendsig */ 78 } 79 80 ucode=0; 81 eva = rcr2(); 82 code = frame.tf_err; 83 switch (type) { 84 85 default: 86 we_re_toast: 87 #ifdef KDB 88 if (kdb_trap(&psl)) 89 return; 90 #endif 91 92 printf("trap type %d code = %x eip = %x cs = %x eflags = %x ", 93 frame.tf_trapno, frame.tf_err, frame.tf_eip, 94 frame.tf_cs, frame.tf_eflags); 95 printf("cr2 %x cpl %x\n", eva, cpl); 96 type &= ~T_USER; 97 panic("trap"); 98 /*NOTREACHED*/ 99 100 case T_SEGNPFLT|T_USER: 101 case T_PROTFLT|T_USER: /* protection fault */ 102 copyfault: 103 ucode = code + BUS_SEGM_FAULT ; 104 i = SIGBUS; 105 break; 106 107 case T_PRIVINFLT|T_USER: /* privileged instruction fault */ 108 case T_RESADFLT|T_USER: /* reserved addressing fault */ 109 case T_RESOPFLT|T_USER: /* reserved operand fault */ 110 case T_FPOPFLT|T_USER: /* coprocessor operand fault */ 111 ucode = type &~ T_USER; 112 i = SIGILL; 113 break; 114 115 case T_ASTFLT|T_USER: /* Allow process switch */ 116 case T_ASTFLT: 117 astoff(); 118 if ((p->p_flag & SOWEUPC) && p->p_stats->p_prof.pr_scale) { 119 addupc(frame.tf_eip, &p->p_stats->p_prof, 1); 120 p->p_flag &= ~SOWEUPC; 121 } 122 goto out; 123 124 case T_DNA|T_USER: 125 #ifdef NPX 126 /* if a transparent fault (due to context switch "late") */ 127 if (npxdna()) return; 128 #endif 129 ucode = FPE_FPU_NP_TRAP; 130 i = SIGFPE; 131 break; 132 133 case T_BOUND|T_USER: 134 ucode = FPE_SUBRNG_TRAP; 135 i = SIGFPE; 136 break; 137 138 case T_OFLOW|T_USER: 139 ucode = FPE_INTOVF_TRAP; 140 i = SIGFPE; 141 break; 142 143 case T_DIVIDE|T_USER: 144 ucode = FPE_INTDIV_TRAP; 145 i = SIGFPE; 146 break; 147 148 case T_ARITHTRAP|T_USER: 149 ucode = code; 150 i = SIGFPE; 151 break; 152 153 case T_PAGEFLT: /* allow page faults in kernel mode */ 154 if (code & PGEX_P) goto we_re_toast; 155 156 /* fall into */ 157 case T_PAGEFLT|T_USER: /* page fault */ 158 { 159 register vm_offset_t va; 160 register struct vmspace *vm = p->p_vmspace; 161 register vm_map_t map; 162 int rv; 163 vm_prot_t ftype; 164 extern vm_map_t kernel_map; 165 unsigned nss,v; 166 167 va = trunc_page((vm_offset_t)eva); 168 /* 169 * It is only a kernel address space fault iff: 170 * 1. (type & T_USER) == 0 and 171 * 2. pcb_onfault not set or 172 * 3. pcb_onfault set but supervisor space fault 173 * The last can occur during an exec() copyin where the 174 * argument space is lazy-allocated. 175 */ 176 if (type == T_PAGEFLT && va >= 0xfe000000) 177 map = kernel_map; 178 else 179 map = &vm->vm_map; 180 if (code & PGEX_W) 181 ftype = VM_PROT_READ | VM_PROT_WRITE; 182 else 183 ftype = VM_PROT_READ; 184 185 #ifdef DEBUG 186 if (map == kernel_map && va == 0) { 187 printf("trap: bad kernel access at %x\n", va); 188 goto we_re_toast; 189 } 190 #endif 191 /* 192 * XXX: rude hack to make stack limits "work" 193 */ 194 nss = 0; 195 if ((caddr_t)va >= vm->vm_maxsaddr && map != kernel_map) { 196 nss = clrnd(btoc(USRSTACK-(unsigned)va)); 197 if (nss > btoc(p->p_rlimit[RLIMIT_STACK].rlim_cur)) { 198 rv = KERN_FAILURE; 199 goto nogo; 200 } 201 } 202 203 /* check if page table is mapped, if not, fault it first */ 204 #define pde_v(v) (PTD[((v)>>PD_SHIFT)&1023].pd_v) 205 if (!pde_v(va)) { 206 v = trunc_page(vtopte(va)); 207 rv = vm_fault(map, v, ftype, FALSE); 208 if (rv != KERN_SUCCESS) goto nogo; 209 /* check if page table fault, increment wiring */ 210 vm_map_pageable(map, v, round_page(v+1), FALSE); 211 } else v=0; 212 rv = vm_fault(map, va, ftype, FALSE); 213 if (rv == KERN_SUCCESS) { 214 /* 215 * XXX: continuation of rude stack hack 216 */ 217 if (nss > vm->vm_ssize) 218 vm->vm_ssize = nss; 219 va = trunc_page(vtopte(va)); 220 /* for page table, increment wiring 221 as long as not a page table fault as well */ 222 if (!v && type != T_PAGEFLT) 223 vm_map_pageable(map, va, round_page(va+1), FALSE); 224 if (type == T_PAGEFLT) 225 return; 226 goto out; 227 } 228 nogo: 229 if (type == T_PAGEFLT) { 230 if (curpcb->pcb_onfault) 231 goto copyfault; 232 printf("vm_fault(%x, %x, %x, 0) -> %x\n", 233 map, va, ftype, rv); 234 printf(" type %x, code %x\n", 235 type, code); 236 goto we_re_toast; 237 } 238 i = (rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV; 239 break; 240 } 241 242 case T_TRCTRAP: /* trace trap -- someone single stepping lcall's */ 243 frame.tf_eflags &= ~PSL_T; 244 245 /* Q: how do we turn it on again? */ 246 return; 247 248 case T_BPTFLT|T_USER: /* bpt instruction fault */ 249 case T_TRCTRAP|T_USER: /* trace trap */ 250 frame.tf_eflags &= ~PSL_T; 251 i = SIGTRAP; 252 break; 253 254 #include "isa.h" 255 #if NISA > 0 256 case T_NMI: 257 case T_NMI|T_USER: 258 /* machine/parity/power fail/"kitchen sink" faults */ 259 if(isa_nmi(code) == 0) return; 260 else goto we_re_toast; 261 #endif 262 } 263 264 trapsignal(p, i, ucode); 265 if ((type & T_USER) == 0) 266 return; 267 out: 268 while (i = CURSIG(p)) 269 psig(i); 270 p->p_pri = p->p_usrpri; 271 if (want_resched) { 272 int pl; 273 /* 274 * Since we are curproc, clock will normally just change 275 * our priority without moving us from one queue to another 276 * (since the running process is not on a queue.) 277 * If that happened after we setrq ourselves but before we 278 * swtch()'ed, we might not be on the queue indicated by 279 * our priority. 280 */ 281 pl = splclock(); 282 setrq(p); 283 p->p_stats->p_ru.ru_nivcsw++; 284 swtch(); 285 splx(pl); 286 while (i = CURSIG(p)) 287 psig(i); 288 } 289 if (p->p_stats->p_prof.pr_scale) { 290 int ticks; 291 struct timeval *tv = &p->p_stime; 292 293 ticks = ((tv->tv_sec - syst.tv_sec) * 1000 + 294 (tv->tv_usec - syst.tv_usec) / 1000) / (tick / 1000); 295 if (ticks) { 296 #ifdef PROFTIMER 297 extern int profscale; 298 addupc(frame.tf_eip, &p->p_stats->p_prof, 299 ticks * profscale); 300 #else 301 addupc(frame.tf_eip, &p->p_stats->p_prof, ticks); 302 #endif 303 } 304 } 305 curpri = p->p_pri; 306 curpcb->pcb_flags &= ~FM_TRAP; /* used by sendsig */ 307 } 308 309 /* 310 * syscall(frame): 311 * System call request from POSIX system call gate interface to kernel. 312 * Like trap(), argument is call by reference. 313 */ 314 /*ARGSUSED*/ 315 syscall(frame) 316 volatile struct syscframe frame; 317 { 318 register int *locr0 = ((int *)&frame); 319 register caddr_t params; 320 register int i; 321 register struct sysent *callp; 322 register struct proc *p = curproc; 323 struct timeval syst; 324 int error, opc; 325 int args[8], rval[2]; 326 int code; 327 328 #ifdef lint 329 r0 = 0; r0 = r0; r1 = 0; r1 = r1; 330 #endif 331 syst = p->p_stime; 332 if (ISPL(frame.sf_cs) != SEL_UPL) 333 panic("syscall"); 334 335 code = frame.sf_eax; 336 p->p_regs = (int *)&frame; 337 curpcb->pcb_flags &= ~FM_TRAP; /* used by sendsig */ 338 params = (caddr_t)frame.sf_esp + sizeof (int) ; 339 340 /* 341 * Reconstruct pc, assuming lcall $X,y is 7 bytes, as it is always. 342 */ 343 opc = frame.sf_eip - 7; 344 callp = (code >= nsysent) ? &sysent[63] : &sysent[code]; 345 if (callp == sysent) { 346 i = fuword(params); 347 params += sizeof (int); 348 callp = (code >= nsysent) ? &sysent[63] : &sysent[code]; 349 } 350 351 if ((i = callp->sy_narg * sizeof (int)) && 352 (error = copyin(params, (caddr_t)args, (u_int)i))) { 353 frame.sf_eax = error; 354 frame.sf_eflags |= PSL_C; /* carry bit */ 355 #ifdef KTRACE 356 if (KTRPOINT(p, KTR_SYSCALL)) 357 ktrsyscall(p->p_tracep, code, callp->sy_narg, &args); 358 #endif 359 goto done; 360 } 361 #ifdef KTRACE 362 if (KTRPOINT(p, KTR_SYSCALL)) 363 ktrsyscall(p->p_tracep, code, callp->sy_narg, &args); 364 #endif 365 rval[0] = 0; 366 rval[1] = frame.sf_edx; 367 error = (*callp->sy_call)(p, args, rval); 368 if (error == ERESTART) 369 frame.sf_eip = opc; 370 else if (error != EJUSTRETURN) { 371 if (error) { 372 frame.sf_eax = error; 373 frame.sf_eflags |= PSL_C; /* carry bit */ 374 } else { 375 frame.sf_eax = rval[0]; 376 frame.sf_edx = rval[1]; 377 frame.sf_eflags &= ~PSL_C; /* carry bit */ 378 } 379 } 380 /* else if (error == EJUSTRETURN) */ 381 /* nothing to do */ 382 done: 383 /* 384 * Reinitialize proc pointer `p' as it may be different 385 * if this is a child returning from fork syscall. 386 */ 387 p = curproc; 388 while (i = CURSIG(p)) 389 psig(i); 390 p->p_pri = p->p_usrpri; 391 if (want_resched) { 392 int pl; 393 /* 394 * Since we are curproc, clock will normally just change 395 * our priority without moving us from one queue to another 396 * (since the running process is not on a queue.) 397 * If that happened after we setrq ourselves but before we 398 * swtch()'ed, we might not be on the queue indicated by 399 * our priority. 400 */ 401 pl = splclock(); 402 setrq(p); 403 p->p_stats->p_ru.ru_nivcsw++; 404 swtch(); 405 splx(pl); 406 while (i = CURSIG(p)) 407 psig(i); 408 } 409 if (p->p_stats->p_prof.pr_scale) { 410 int ticks; 411 struct timeval *tv = &p->p_stime; 412 413 ticks = ((tv->tv_sec - syst.tv_sec) * 1000 + 414 (tv->tv_usec - syst.tv_usec) / 1000) / (tick / 1000); 415 if (ticks) { 416 #ifdef PROFTIMER 417 extern int profscale; 418 addupc(frame.sf_eip, &p->p_stats->p_prof, 419 ticks * profscale); 420 #else 421 addupc(frame.sf_eip, &p->p_stats->p_prof, ticks); 422 #endif 423 } 424 } 425 curpri = p->p_pri; 426 #ifdef KTRACE 427 if (KTRPOINT(p, KTR_SYSRET)) 428 ktrsysret(p->p_tracep, code, error, rval[0]); 429 #endif 430 } 431