| 1 | /*- |
| 2 | * Copyright (C) 1994, David Greenman |
| 3 | * Copyright (c) 1990, 1993 |
| 4 | * The Regents of the University of California. All rights reserved. |
| 5 | * |
| 6 | * This code is derived from software contributed to Berkeley by |
| 7 | * the University of Utah, and William Jolitz. |
| 8 | * |
| 9 | * Redistribution and use in source and binary forms, with or without |
| 10 | * modification, are permitted provided that the following conditions |
| 11 | * are met: |
| 12 | * 1. Redistributions of source code must retain the above copyright |
| 13 | * notice, this list of conditions and the following disclaimer. |
| 14 | * 2. Redistributions in binary form must reproduce the above copyright |
| 15 | * notice, this list of conditions and the following disclaimer in the |
| 16 | * documentation and/or other materials provided with the distribution. |
| 17 | * 3. All advertising materials mentioning features or use of this software |
| 18 | * must display the following acknowledgement: |
| 19 | * This product includes software developed by the University of |
| 20 | * California, Berkeley and its contributors. |
| 21 | * 4. Neither the name of the University nor the names of its contributors |
| 22 | * may be used to endorse or promote products derived from this software |
| 23 | * without specific prior written permission. |
| 24 | * |
| 25 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND |
| 26 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 27 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 28 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE |
| 29 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 30 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
| 31 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
| 32 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
| 33 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
| 34 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 35 | * SUCH DAMAGE. |
| 36 | * |
| 37 | * from: @(#)trap.c 7.4 (Berkeley) 5/13/91 |
| 38 | * $FreeBSD: src/sys/i386/i386/trap.c,v 1.147.2.11 2003/02/27 19:09:59 luoqi Exp $ |
| 39 | * $DragonFly: src/sys/i386/i386/Attic/trap.c,v 1.18 2003/07/06 21:23:48 dillon Exp $ |
| 40 | */ |
| 41 | |
| 42 | /* |
| 43 | * 386 Trap and System call handling |
| 44 | */ |
| 45 | |
| 46 | #include "opt_cpu.h" |
| 47 | #include "opt_ddb.h" |
| 48 | #include "opt_ktrace.h" |
| 49 | #include "opt_clock.h" |
| 50 | #include "opt_trap.h" |
| 51 | |
| 52 | #include <sys/param.h> |
| 53 | #include <sys/systm.h> |
| 54 | #include <sys/proc.h> |
| 55 | #include <sys/pioctl.h> |
| 56 | #include <sys/kernel.h> |
| 57 | #include <sys/resourcevar.h> |
| 58 | #include <sys/signalvar.h> |
| 59 | #include <sys/syscall.h> |
| 60 | #include <sys/sysctl.h> |
| 61 | #include <sys/sysent.h> |
| 62 | #include <sys/uio.h> |
| 63 | #include <sys/vmmeter.h> |
| 64 | #ifdef KTRACE |
| 65 | #include <sys/ktrace.h> |
| 66 | #endif |
| 67 | |
| 68 | #include <vm/vm.h> |
| 69 | #include <vm/vm_param.h> |
| 70 | #include <sys/lock.h> |
| 71 | #include <vm/pmap.h> |
| 72 | #include <vm/vm_kern.h> |
| 73 | #include <vm/vm_map.h> |
| 74 | #include <vm/vm_page.h> |
| 75 | #include <vm/vm_extern.h> |
| 76 | |
| 77 | #include <machine/cpu.h> |
| 78 | #include <machine/ipl.h> |
| 79 | #include <machine/md_var.h> |
| 80 | #include <machine/pcb.h> |
| 81 | #ifdef SMP |
| 82 | #include <machine/smp.h> |
| 83 | #endif |
| 84 | #include <machine/tss.h> |
| 85 | #include <machine/globaldata.h> |
| 86 | |
| 87 | #include <i386/isa/intr_machdep.h> |
| 88 | |
| 89 | #ifdef POWERFAIL_NMI |
| 90 | #include <sys/syslog.h> |
| 91 | #include <machine/clock.h> |
| 92 | #endif |
| 93 | |
| 94 | #include <machine/vm86.h> |
| 95 | |
| 96 | #include <ddb/ddb.h> |
| 97 | #include <sys/thread2.h> |
| 98 | |
| 99 | #include "isa.h" |
| 100 | #include "npx.h" |
| 101 | |
| 102 | int (*pmath_emulate) __P((struct trapframe *)); |
| 103 | |
| 104 | extern void trap __P((struct trapframe frame)); |
| 105 | extern int trapwrite __P((unsigned addr)); |
| 106 | extern void syscall2 __P((struct trapframe frame)); |
| 107 | |
| 108 | static int trap_pfault __P((struct trapframe *, int, vm_offset_t)); |
| 109 | static void trap_fatal __P((struct trapframe *, vm_offset_t)); |
| 110 | void dblfault_handler __P((void)); |
| 111 | |
| 112 | extern inthand_t IDTVEC(syscall); |
| 113 | |
| 114 | #define MAX_TRAP_MSG 28 |
| 115 | static char *trap_msg[] = { |
| 116 | "", /* 0 unused */ |
| 117 | "privileged instruction fault", /* 1 T_PRIVINFLT */ |
| 118 | "", /* 2 unused */ |
| 119 | "breakpoint instruction fault", /* 3 T_BPTFLT */ |
| 120 | "", /* 4 unused */ |
| 121 | "", /* 5 unused */ |
| 122 | "arithmetic trap", /* 6 T_ARITHTRAP */ |
| 123 | "system forced exception", /* 7 T_ASTFLT */ |
| 124 | "", /* 8 unused */ |
| 125 | "general protection fault", /* 9 T_PROTFLT */ |
| 126 | "trace trap", /* 10 T_TRCTRAP */ |
| 127 | "", /* 11 unused */ |
| 128 | "page fault", /* 12 T_PAGEFLT */ |
| 129 | "", /* 13 unused */ |
| 130 | "alignment fault", /* 14 T_ALIGNFLT */ |
| 131 | "", /* 15 unused */ |
| 132 | "", /* 16 unused */ |
| 133 | "", /* 17 unused */ |
| 134 | "integer divide fault", /* 18 T_DIVIDE */ |
| 135 | "non-maskable interrupt trap", /* 19 T_NMI */ |
| 136 | "overflow trap", /* 20 T_OFLOW */ |
| 137 | "FPU bounds check fault", /* 21 T_BOUND */ |
| 138 | "FPU device not available", /* 22 T_DNA */ |
| 139 | "double fault", /* 23 T_DOUBLEFLT */ |
| 140 | "FPU operand fetch fault", /* 24 T_FPOPFLT */ |
| 141 | "invalid TSS fault", /* 25 T_TSSFLT */ |
| 142 | "segment not present fault", /* 26 T_SEGNPFLT */ |
| 143 | "stack fault", /* 27 T_STKFLT */ |
| 144 | "machine check trap", /* 28 T_MCHK */ |
| 145 | }; |
| 146 | |
| 147 | #if defined(I586_CPU) && !defined(NO_F00F_HACK) |
| 148 | extern int has_f00f_bug; |
| 149 | #endif |
| 150 | |
| 151 | #ifdef DDB |
| 152 | static int ddb_on_nmi = 1; |
| 153 | SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW, |
| 154 | &ddb_on_nmi, 0, "Go to DDB on NMI"); |
| 155 | #endif |
| 156 | static int panic_on_nmi = 1; |
| 157 | SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW, |
| 158 | &panic_on_nmi, 0, "Panic on NMI"); |
| 159 | |
| 160 | /* |
| 161 | * USER->KERNEL transition. Do not transition us out of userland from the |
| 162 | * point of view of the userland scheduler unless we actually have to |
| 163 | * switch. |
| 164 | * |
| 165 | * usertdsw is called from within a critical section, but the BGL will |
| 166 | * have already been released by lwkt_switch() so only call MP safe functions |
| 167 | * that don't block! |
| 168 | */ |
| 169 | static void |
| 170 | usertdsw(struct thread *ntd) |
| 171 | { |
| 172 | struct thread *td = curthread; |
| 173 | |
| 174 | td->td_switch = cpu_heavy_switch; |
| 175 | lwkt_setpri_self(TDPRI_KERN_USER); |
| 176 | #if 0 |
| 177 | /* |
| 178 | * This is where we might want to catch the P_CURPROC designation |
| 179 | * and fix it for *any* switchout rather then just an mi_switch() |
| 180 | * switchout (move from mi_switch()?) YYY |
| 181 | */ |
| 182 | if (p->p_flag & P_CURPROC) { |
| 183 | ... |
| 184 | } |
| 185 | #endif |
| 186 | td->td_switch(ntd); |
| 187 | } |
| 188 | |
| 189 | /* |
| 190 | * userenter() passively intercepts the thread switch function to increase |
| 191 | * the thread priority from a user priority to a kernel priority, reducing |
| 192 | * syscall and trap overhead for the case where no switch occurs. |
| 193 | */ |
| 194 | |
| 195 | static __inline void |
| 196 | userenter(void) |
| 197 | { |
| 198 | struct thread *td; |
| 199 | |
| 200 | td = curthread; |
| 201 | KASSERT(td->td_switch == cpu_heavy_switch, |
| 202 | ("userenter: bad td_switch = %p", td->td_switch)); |
| 203 | #if 0 |
| 204 | KASSERT(td->td_switch == cpu_heavy_switch || td->td_switch == usertdsw, |
| 205 | ("userenter: bad td_switch = %p", td->td_switch)); |
| 206 | #endif |
| 207 | td->td_switch = usertdsw; |
| 208 | } |
| 209 | |
| 210 | static void |
| 211 | userret(struct proc *p, struct trapframe *frame, u_quad_t oticks) |
| 212 | { |
| 213 | int sig, s; |
| 214 | struct thread *td = curthread; |
| 215 | |
| 216 | /* |
| 217 | * Post any pending signals |
| 218 | */ |
| 219 | crit_enter(); |
| 220 | while ((sig = CURSIG(p)) != 0) { |
| 221 | crit_exit(); |
| 222 | postsig(sig); |
| 223 | crit_enter(); |
| 224 | } |
| 225 | |
| 226 | /* |
| 227 | * Set our priority properly and restore our switch function. If |
| 228 | * we did not hit our lazy switch function in the first place we |
| 229 | * do not need to restore anything. |
| 230 | */ |
| 231 | if (td->td_switch == cpu_heavy_switch) { |
| 232 | switch(p->p_rtprio.type) { |
| 233 | case RTP_PRIO_IDLE: |
| 234 | lwkt_setpri_self(TDPRI_USER_IDLE); |
| 235 | break; |
| 236 | case RTP_PRIO_REALTIME: |
| 237 | case RTP_PRIO_FIFO: |
| 238 | lwkt_setpri_self(TDPRI_USER_REAL); |
| 239 | break; |
| 240 | default: |
| 241 | lwkt_setpri_self(TDPRI_USER_NORM); |
| 242 | break; |
| 243 | } |
| 244 | } else { |
| 245 | KKASSERT(td->td_switch == usertdsw); |
| 246 | td->td_switch = cpu_heavy_switch; |
| 247 | } |
| 248 | crit_exit(); |
| 249 | |
| 250 | /* |
| 251 | * If a reschedule has been requested we call chooseproc() to locate |
| 252 | * the next runnable process. When we wakeup from that we check |
| 253 | * for pending signals again. |
| 254 | */ |
| 255 | if (resched_wanted()) { |
| 256 | uio_yield(); |
| 257 | while ((sig = CURSIG(p)) != 0) |
| 258 | postsig(sig); |
| 259 | } |
| 260 | |
| 261 | /* |
| 262 | * Charge system time if profiling. |
| 263 | */ |
| 264 | if (p->p_flag & P_PROFIL) { |
| 265 | addupc_task(p, frame->tf_eip, |
| 266 | (u_int)(curthread->td_sticks - oticks) * psratio); |
| 267 | } |
| 268 | |
| 269 | /* |
| 270 | * In order to return to userland we need to be the designated |
| 271 | * current (user) process on this cpu. We have to wait for |
| 272 | * the userland scheduler to schedule as P_CURPROC. |
| 273 | */ |
| 274 | s = splhigh(); |
| 275 | while ((p->p_flag & P_CURPROC) == 0) { |
| 276 | p->p_stats->p_ru.ru_nivcsw++; |
| 277 | lwkt_deschedule_self(); |
| 278 | mi_switch(); |
| 279 | } |
| 280 | splx(s); |
| 281 | KKASSERT(mycpu->gd_uprocscheduled == 1); |
| 282 | } |
| 283 | |
| 284 | #ifdef DEVICE_POLLING |
| 285 | extern u_int32_t poll_in_trap; |
| 286 | extern int ether_poll __P((int count)); |
| 287 | #endif /* DEVICE_POLLING */ |
| 288 | |
| 289 | /* |
| 290 | * Exception, fault, and trap interface to the FreeBSD kernel. |
| 291 | * This common code is called from assembly language IDT gate entry |
| 292 | * routines that prepare a suitable stack frame, and restore this |
| 293 | * frame after the exception has been processed. |
| 294 | */ |
| 295 | |
| 296 | void |
| 297 | trap(frame) |
| 298 | struct trapframe frame; |
| 299 | { |
| 300 | struct proc *p = curproc; |
| 301 | u_quad_t sticks = 0; |
| 302 | int i = 0, ucode = 0, type, code; |
| 303 | vm_offset_t eva; |
| 304 | |
| 305 | get_mplock(); |
| 306 | |
| 307 | #ifdef DDB |
| 308 | if (db_active) { |
| 309 | eva = (frame.tf_trapno == T_PAGEFLT ? rcr2() : 0); |
| 310 | trap_fatal(&frame, eva); |
| 311 | goto out2; |
| 312 | } |
| 313 | #endif |
| 314 | |
| 315 | if (!(frame.tf_eflags & PSL_I)) { |
| 316 | /* |
| 317 | * Buggy application or kernel code has disabled interrupts |
| 318 | * and then trapped. Enabling interrupts now is wrong, but |
| 319 | * it is better than running with interrupts disabled until |
| 320 | * they are accidentally enabled later. |
| 321 | */ |
| 322 | type = frame.tf_trapno; |
| 323 | if (ISPL(frame.tf_cs) == SEL_UPL || (frame.tf_eflags & PSL_VM)) |
| 324 | printf( |
| 325 | "pid %ld (%s): trap %d with interrupts disabled\n", |
| 326 | (long)curproc->p_pid, curproc->p_comm, type); |
| 327 | else if (type != T_BPTFLT && type != T_TRCTRAP) |
| 328 | /* |
| 329 | * XXX not quite right, since this may be for a |
| 330 | * multiple fault in user mode. |
| 331 | */ |
| 332 | printf("kernel trap %d with interrupts disabled\n", |
| 333 | type); |
| 334 | cpu_enable_intr(); |
| 335 | } |
| 336 | |
| 337 | eva = 0; |
| 338 | if (frame.tf_trapno == T_PAGEFLT) { |
| 339 | /* |
| 340 | * For some Cyrix CPUs, %cr2 is clobbered by interrupts. |
| 341 | * This problem is worked around by using an interrupt |
| 342 | * gate for the pagefault handler. We are finally ready |
| 343 | * to read %cr2 and then must reenable interrupts. |
| 344 | * |
| 345 | * XXX this should be in the switch statement, but the |
| 346 | * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the |
| 347 | * flow of control too much for this to be obviously |
| 348 | * correct. |
| 349 | */ |
| 350 | eva = rcr2(); |
| 351 | cpu_enable_intr(); |
| 352 | } |
| 353 | |
| 354 | #ifdef DEVICE_POLLING |
| 355 | if (poll_in_trap) |
| 356 | ether_poll(poll_in_trap); |
| 357 | #endif /* DEVICE_POLLING */ |
| 358 | |
| 359 | #if defined(I586_CPU) && !defined(NO_F00F_HACK) |
| 360 | restart: |
| 361 | #endif |
| 362 | type = frame.tf_trapno; |
| 363 | code = frame.tf_err; |
| 364 | |
| 365 | if (in_vm86call) { |
| 366 | if (frame.tf_eflags & PSL_VM && |
| 367 | (type == T_PROTFLT || type == T_STKFLT)) { |
| 368 | i = vm86_emulate((struct vm86frame *)&frame); |
| 369 | if (i != 0) { |
| 370 | /* |
| 371 | * returns to original process |
| 372 | */ |
| 373 | vm86_trap((struct vm86frame *)&frame); |
| 374 | } |
| 375 | goto out2; |
| 376 | } |
| 377 | switch (type) { |
| 378 | /* |
| 379 | * these traps want either a process context, or |
| 380 | * assume a normal userspace trap. |
| 381 | */ |
| 382 | case T_PROTFLT: |
| 383 | case T_SEGNPFLT: |
| 384 | trap_fatal(&frame, eva); |
| 385 | goto out2; |
| 386 | case T_TRCTRAP: |
| 387 | type = T_BPTFLT; /* kernel breakpoint */ |
| 388 | /* FALL THROUGH */ |
| 389 | } |
| 390 | goto kernel_trap; /* normal kernel trap handling */ |
| 391 | } |
| 392 | |
| 393 | if ((ISPL(frame.tf_cs) == SEL_UPL) || (frame.tf_eflags & PSL_VM)) { |
| 394 | /* user trap */ |
| 395 | |
| 396 | userenter(); |
| 397 | |
| 398 | sticks = curthread->td_sticks; |
| 399 | p->p_md.md_regs = &frame; |
| 400 | |
| 401 | switch (type) { |
| 402 | case T_PRIVINFLT: /* privileged instruction fault */ |
| 403 | ucode = type; |
| 404 | i = SIGILL; |
| 405 | break; |
| 406 | |
| 407 | case T_BPTFLT: /* bpt instruction fault */ |
| 408 | case T_TRCTRAP: /* trace trap */ |
| 409 | frame.tf_eflags &= ~PSL_T; |
| 410 | i = SIGTRAP; |
| 411 | break; |
| 412 | |
| 413 | case T_ARITHTRAP: /* arithmetic trap */ |
| 414 | ucode = code; |
| 415 | i = SIGFPE; |
| 416 | break; |
| 417 | |
| 418 | case T_ASTFLT: /* Allow process switch */ |
| 419 | astoff(); |
| 420 | mycpu->gd_cnt.v_soft++; |
| 421 | if (p->p_flag & P_OWEUPC) { |
| 422 | p->p_flag &= ~P_OWEUPC; |
| 423 | addupc_task(p, p->p_stats->p_prof.pr_addr, |
| 424 | p->p_stats->p_prof.pr_ticks); |
| 425 | } |
| 426 | goto out; |
| 427 | |
| 428 | /* |
| 429 | * The following two traps can happen in |
| 430 | * vm86 mode, and, if so, we want to handle |
| 431 | * them specially. |
| 432 | */ |
| 433 | case T_PROTFLT: /* general protection fault */ |
| 434 | case T_STKFLT: /* stack fault */ |
| 435 | if (frame.tf_eflags & PSL_VM) { |
| 436 | i = vm86_emulate((struct vm86frame *)&frame); |
| 437 | if (i == 0) |
| 438 | goto out; |
| 439 | break; |
| 440 | } |
| 441 | /* FALL THROUGH */ |
| 442 | |
| 443 | case T_SEGNPFLT: /* segment not present fault */ |
| 444 | case T_TSSFLT: /* invalid TSS fault */ |
| 445 | case T_DOUBLEFLT: /* double fault */ |
| 446 | default: |
| 447 | ucode = code + BUS_SEGM_FAULT ; |
| 448 | i = SIGBUS; |
| 449 | break; |
| 450 | |
| 451 | case T_PAGEFLT: /* page fault */ |
| 452 | i = trap_pfault(&frame, TRUE, eva); |
| 453 | if (i == -1) |
| 454 | goto out; |
| 455 | #if defined(I586_CPU) && !defined(NO_F00F_HACK) |
| 456 | if (i == -2) |
| 457 | goto restart; |
| 458 | #endif |
| 459 | if (i == 0) |
| 460 | goto out; |
| 461 | |
| 462 | ucode = T_PAGEFLT; |
| 463 | break; |
| 464 | |
| 465 | case T_DIVIDE: /* integer divide fault */ |
| 466 | ucode = FPE_INTDIV; |
| 467 | i = SIGFPE; |
| 468 | break; |
| 469 | |
| 470 | #if NISA > 0 |
| 471 | case T_NMI: |
| 472 | #ifdef POWERFAIL_NMI |
| 473 | goto handle_powerfail; |
| 474 | #else /* !POWERFAIL_NMI */ |
| 475 | /* machine/parity/power fail/"kitchen sink" faults */ |
| 476 | if (isa_nmi(code) == 0) { |
| 477 | #ifdef DDB |
| 478 | /* |
| 479 | * NMI can be hooked up to a pushbutton |
| 480 | * for debugging. |
| 481 | */ |
| 482 | if (ddb_on_nmi) { |
| 483 | printf ("NMI ... going to debugger\n"); |
| 484 | kdb_trap (type, 0, &frame); |
| 485 | } |
| 486 | #endif /* DDB */ |
| 487 | goto out2; |
| 488 | } else if (panic_on_nmi) |
| 489 | panic("NMI indicates hardware failure"); |
| 490 | break; |
| 491 | #endif /* POWERFAIL_NMI */ |
| 492 | #endif /* NISA > 0 */ |
| 493 | |
| 494 | case T_OFLOW: /* integer overflow fault */ |
| 495 | ucode = FPE_INTOVF; |
| 496 | i = SIGFPE; |
| 497 | break; |
| 498 | |
| 499 | case T_BOUND: /* bounds check fault */ |
| 500 | ucode = FPE_FLTSUB; |
| 501 | i = SIGFPE; |
| 502 | break; |
| 503 | |
| 504 | case T_DNA: |
| 505 | #if NNPX > 0 |
| 506 | /* if a transparent fault (due to context switch "late") */ |
| 507 | if (npxdna()) |
| 508 | goto out; |
| 509 | #endif |
| 510 | if (!pmath_emulate) { |
| 511 | i = SIGFPE; |
| 512 | ucode = FPE_FPU_NP_TRAP; |
| 513 | break; |
| 514 | } |
| 515 | i = (*pmath_emulate)(&frame); |
| 516 | if (i == 0) { |
| 517 | if (!(frame.tf_eflags & PSL_T)) |
| 518 | goto out2; |
| 519 | frame.tf_eflags &= ~PSL_T; |
| 520 | i = SIGTRAP; |
| 521 | } |
| 522 | /* else ucode = emulator_only_knows() XXX */ |
| 523 | break; |
| 524 | |
| 525 | case T_FPOPFLT: /* FPU operand fetch fault */ |
| 526 | ucode = T_FPOPFLT; |
| 527 | i = SIGILL; |
| 528 | break; |
| 529 | |
| 530 | case T_XMMFLT: /* SIMD floating-point exception */ |
| 531 | ucode = 0; /* XXX */ |
| 532 | i = SIGFPE; |
| 533 | break; |
| 534 | } |
| 535 | } else { |
| 536 | kernel_trap: |
| 537 | /* kernel trap */ |
| 538 | |
| 539 | switch (type) { |
| 540 | case T_PAGEFLT: /* page fault */ |
| 541 | (void) trap_pfault(&frame, FALSE, eva); |
| 542 | goto out2; |
| 543 | |
| 544 | case T_DNA: |
| 545 | #if NNPX > 0 |
| 546 | /* |
| 547 | * The kernel is apparently using npx for copying. |
| 548 | * XXX this should be fatal unless the kernel has |
| 549 | * registered such use. |
| 550 | */ |
| 551 | if (npxdna()) |
| 552 | goto out2; |
| 553 | #endif |
| 554 | break; |
| 555 | |
| 556 | case T_PROTFLT: /* general protection fault */ |
| 557 | case T_SEGNPFLT: /* segment not present fault */ |
| 558 | /* |
| 559 | * Invalid segment selectors and out of bounds |
| 560 | * %eip's and %esp's can be set up in user mode. |
| 561 | * This causes a fault in kernel mode when the |
| 562 | * kernel tries to return to user mode. We want |
| 563 | * to get this fault so that we can fix the |
| 564 | * problem here and not have to check all the |
| 565 | * selectors and pointers when the user changes |
| 566 | * them. |
| 567 | */ |
| 568 | #define MAYBE_DORETI_FAULT(where, whereto) \ |
| 569 | do { \ |
| 570 | if (frame.tf_eip == (int)where) { \ |
| 571 | frame.tf_eip = (int)whereto; \ |
| 572 | goto out2; \ |
| 573 | } \ |
| 574 | } while (0) |
| 575 | |
| 576 | if (mycpu->gd_intr_nesting_level == 0) { |
| 577 | /* |
| 578 | * Invalid %fs's and %gs's can be created using |
| 579 | * procfs or PT_SETREGS or by invalidating the |
| 580 | * underlying LDT entry. This causes a fault |
| 581 | * in kernel mode when the kernel attempts to |
| 582 | * switch contexts. Lose the bad context |
| 583 | * (XXX) so that we can continue, and generate |
| 584 | * a signal. |
| 585 | */ |
| 586 | if (frame.tf_eip == (int)cpu_switch_load_gs) { |
| 587 | curthread->td_pcb->pcb_gs = 0; |
| 588 | psignal(p, SIGBUS); |
| 589 | goto out2; |
| 590 | } |
| 591 | MAYBE_DORETI_FAULT(doreti_iret, |
| 592 | doreti_iret_fault); |
| 593 | MAYBE_DORETI_FAULT(doreti_popl_ds, |
| 594 | doreti_popl_ds_fault); |
| 595 | MAYBE_DORETI_FAULT(doreti_popl_es, |
| 596 | doreti_popl_es_fault); |
| 597 | MAYBE_DORETI_FAULT(doreti_popl_fs, |
| 598 | doreti_popl_fs_fault); |
| 599 | if (curthread->td_pcb->pcb_onfault) { |
| 600 | frame.tf_eip = (int)curthread->td_pcb->pcb_onfault; |
| 601 | goto out2; |
| 602 | } |
| 603 | } |
| 604 | break; |
| 605 | |
| 606 | case T_TSSFLT: |
| 607 | /* |
| 608 | * PSL_NT can be set in user mode and isn't cleared |
| 609 | * automatically when the kernel is entered. This |
| 610 | * causes a TSS fault when the kernel attempts to |
| 611 | * `iret' because the TSS link is uninitialized. We |
| 612 | * want to get this fault so that we can fix the |
| 613 | * problem here and not every time the kernel is |
| 614 | * entered. |
| 615 | */ |
| 616 | if (frame.tf_eflags & PSL_NT) { |
| 617 | frame.tf_eflags &= ~PSL_NT; |
| 618 | goto out2; |
| 619 | } |
| 620 | break; |
| 621 | |
| 622 | case T_TRCTRAP: /* trace trap */ |
| 623 | if (frame.tf_eip == (int)IDTVEC(syscall)) { |
| 624 | /* |
| 625 | * We've just entered system mode via the |
| 626 | * syscall lcall. Continue single stepping |
| 627 | * silently until the syscall handler has |
| 628 | * saved the flags. |
| 629 | */ |
| 630 | goto out2; |
| 631 | } |
| 632 | if (frame.tf_eip == (int)IDTVEC(syscall) + 1) { |
| 633 | /* |
| 634 | * The syscall handler has now saved the |
| 635 | * flags. Stop single stepping it. |
| 636 | */ |
| 637 | frame.tf_eflags &= ~PSL_T; |
| 638 | goto out2; |
| 639 | } |
| 640 | /* |
| 641 | * Ignore debug register trace traps due to |
| 642 | * accesses in the user's address space, which |
| 643 | * can happen under several conditions such as |
| 644 | * if a user sets a watchpoint on a buffer and |
| 645 | * then passes that buffer to a system call. |
| 646 | * We still want to get TRCTRAPS for addresses |
| 647 | * in kernel space because that is useful when |
| 648 | * debugging the kernel. |
| 649 | */ |
| 650 | if (user_dbreg_trap()) { |
| 651 | /* |
| 652 | * Reset breakpoint bits because the |
| 653 | * processor doesn't |
| 654 | */ |
| 655 | load_dr6(rdr6() & 0xfffffff0); |
| 656 | goto out2; |
| 657 | } |
| 658 | /* |
| 659 | * Fall through (TRCTRAP kernel mode, kernel address) |
| 660 | */ |
| 661 | case T_BPTFLT: |
| 662 | /* |
| 663 | * If DDB is enabled, let it handle the debugger trap. |
| 664 | * Otherwise, debugger traps "can't happen". |
| 665 | */ |
| 666 | #ifdef DDB |
| 667 | if (kdb_trap (type, 0, &frame)) |
| 668 | goto out2; |
| 669 | #endif |
| 670 | break; |
| 671 | |
| 672 | #if NISA > 0 |
| 673 | case T_NMI: |
| 674 | #ifdef POWERFAIL_NMI |
| 675 | #ifndef TIMER_FREQ |
| 676 | # define TIMER_FREQ 1193182 |
| 677 | #endif |
| 678 | handle_powerfail: |
| 679 | { |
| 680 | static unsigned lastalert = 0; |
| 681 | |
| 682 | if(time_second - lastalert > 10) |
| 683 | { |
| 684 | log(LOG_WARNING, "NMI: power fail\n"); |
| 685 | sysbeep(TIMER_FREQ/880, hz); |
| 686 | lastalert = time_second; |
| 687 | } |
| 688 | /* YYY mp count */ |
| 689 | goto out2; |
| 690 | } |
| 691 | #else /* !POWERFAIL_NMI */ |
| 692 | /* machine/parity/power fail/"kitchen sink" faults */ |
| 693 | if (isa_nmi(code) == 0) { |
| 694 | #ifdef DDB |
| 695 | /* |
| 696 | * NMI can be hooked up to a pushbutton |
| 697 | * for debugging. |
| 698 | */ |
| 699 | if (ddb_on_nmi) { |
| 700 | printf ("NMI ... going to debugger\n"); |
| 701 | kdb_trap (type, 0, &frame); |
| 702 | } |
| 703 | #endif /* DDB */ |
| 704 | goto out2; |
| 705 | } else if (panic_on_nmi == 0) |
| 706 | goto out2; |
| 707 | /* FALL THROUGH */ |
| 708 | #endif /* POWERFAIL_NMI */ |
| 709 | #endif /* NISA > 0 */ |
| 710 | } |
| 711 | |
| 712 | trap_fatal(&frame, eva); |
| 713 | goto out2; |
| 714 | } |
| 715 | |
| 716 | /* Translate fault for emulators (e.g. Linux) */ |
| 717 | if (*p->p_sysent->sv_transtrap) |
| 718 | i = (*p->p_sysent->sv_transtrap)(i, type); |
| 719 | |
| 720 | trapsignal(p, i, ucode); |
| 721 | |
| 722 | #ifdef DEBUG |
| 723 | if (type <= MAX_TRAP_MSG) { |
| 724 | uprintf("fatal process exception: %s", |
| 725 | trap_msg[type]); |
| 726 | if ((type == T_PAGEFLT) || (type == T_PROTFLT)) |
| 727 | uprintf(", fault VA = 0x%lx", (u_long)eva); |
| 728 | uprintf("\n"); |
| 729 | } |
| 730 | #endif |
| 731 | |
| 732 | out: |
| 733 | #ifdef SMP |
| 734 | if (ISPL(frame.tf_cs) == SEL_UPL) |
| 735 | KASSERT(curthread->td_mpcount == 1, ("badmpcount trap from %p", (void *)frame.tf_eip)); |
| 736 | #endif |
| 737 | userret(p, &frame, sticks); |
| 738 | out2: |
| 739 | rel_mplock(); |
| 740 | } |
| 741 | |
| 742 | #ifdef notyet |
| 743 | /* |
| 744 | * This version doesn't allow a page fault to user space while |
| 745 | * in the kernel. The rest of the kernel needs to be made "safe" |
| 746 | * before this can be used. I think the only things remaining |
| 747 | * to be made safe are the iBCS2 code and the process tracing/ |
| 748 | * debugging code. |
| 749 | */ |
| 750 | static int |
| 751 | trap_pfault(frame, usermode, eva) |
| 752 | struct trapframe *frame; |
| 753 | int usermode; |
| 754 | vm_offset_t eva; |
| 755 | { |
| 756 | vm_offset_t va; |
| 757 | struct vmspace *vm = NULL; |
| 758 | vm_map_t map = 0; |
| 759 | int rv = 0; |
| 760 | vm_prot_t ftype; |
| 761 | struct proc *p = curproc; |
| 762 | |
| 763 | if (frame->tf_err & PGEX_W) |
| 764 | ftype = VM_PROT_WRITE; |
| 765 | else |
| 766 | ftype = VM_PROT_READ; |
| 767 | |
| 768 | va = trunc_page(eva); |
| 769 | if (va < VM_MIN_KERNEL_ADDRESS) { |
| 770 | vm_offset_t v; |
| 771 | vm_page_t mpte; |
| 772 | |
| 773 | if (p == NULL || |
| 774 | (!usermode && va < VM_MAXUSER_ADDRESS && |
| 775 | (mycpu->gd_intr_nesting_level != 0 || |
| 776 | curthread->td_pcb->pcb_onfault == NULL))) { |
| 777 | trap_fatal(frame, eva); |
| 778 | return (-1); |
| 779 | } |
| 780 | |
| 781 | /* |
| 782 | * This is a fault on non-kernel virtual memory. |
| 783 | * vm is initialized above to NULL. If curproc is NULL |
| 784 | * or curproc->p_vmspace is NULL the fault is fatal. |
| 785 | */ |
| 786 | vm = p->p_vmspace; |
| 787 | if (vm == NULL) |
| 788 | goto nogo; |
| 789 | |
| 790 | map = &vm->vm_map; |
| 791 | |
| 792 | /* |
| 793 | * Keep swapout from messing with us during this |
| 794 | * critical time. |
| 795 | */ |
| 796 | ++p->p_lock; |
| 797 | |
| 798 | /* |
| 799 | * Grow the stack if necessary |
| 800 | */ |
| 801 | /* grow_stack returns false only if va falls into |
| 802 | * a growable stack region and the stack growth |
| 803 | * fails. It returns true if va was not within |
| 804 | * a growable stack region, or if the stack |
| 805 | * growth succeeded. |
| 806 | */ |
| 807 | if (!grow_stack (p, va)) { |
| 808 | rv = KERN_FAILURE; |
| 809 | --p->p_lock; |
| 810 | goto nogo; |
| 811 | } |
| 812 | |
| 813 | /* Fault in the user page: */ |
| 814 | rv = vm_fault(map, va, ftype, |
| 815 | (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY |
| 816 | : VM_FAULT_NORMAL); |
| 817 | |
| 818 | --p->p_lock; |
| 819 | } else { |
| 820 | /* |
| 821 | * Don't allow user-mode faults in kernel address space. |
| 822 | */ |
| 823 | if (usermode) |
| 824 | goto nogo; |
| 825 | |
| 826 | /* |
| 827 | * Since we know that kernel virtual address addresses |
| 828 | * always have pte pages mapped, we just have to fault |
| 829 | * the page. |
| 830 | */ |
| 831 | rv = vm_fault(kernel_map, va, ftype, VM_FAULT_NORMAL); |
| 832 | } |
| 833 | |
| 834 | if (rv == KERN_SUCCESS) |
| 835 | return (0); |
| 836 | nogo: |
| 837 | if (!usermode) { |
| 838 | if (mycpu->gd_intr_nesting_level == 0 && curthread->td_pcb->pcb_onfault) { |
| 839 | frame->tf_eip = (int)curthread->td_pcb->pcb_onfault; |
| 840 | return (0); |
| 841 | } |
| 842 | trap_fatal(frame, eva); |
| 843 | return (-1); |
| 844 | } |
| 845 | |
| 846 | /* kludge to pass faulting virtual address to sendsig */ |
| 847 | frame->tf_err = eva; |
| 848 | |
| 849 | return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); |
| 850 | } |
| 851 | #endif |
| 852 | |
| 853 | int |
| 854 | trap_pfault(frame, usermode, eva) |
| 855 | struct trapframe *frame; |
| 856 | int usermode; |
| 857 | vm_offset_t eva; |
| 858 | { |
| 859 | vm_offset_t va; |
| 860 | struct vmspace *vm = NULL; |
| 861 | vm_map_t map = 0; |
| 862 | int rv = 0; |
| 863 | vm_prot_t ftype; |
| 864 | struct proc *p = curproc; |
| 865 | |
| 866 | va = trunc_page(eva); |
| 867 | if (va >= KERNBASE) { |
| 868 | /* |
| 869 | * Don't allow user-mode faults in kernel address space. |
| 870 | * An exception: if the faulting address is the invalid |
| 871 | * instruction entry in the IDT, then the Intel Pentium |
| 872 | * F00F bug workaround was triggered, and we need to |
| 873 | * treat it is as an illegal instruction, and not a page |
| 874 | * fault. |
| 875 | */ |
| 876 | #if defined(I586_CPU) && !defined(NO_F00F_HACK) |
| 877 | if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) { |
| 878 | frame->tf_trapno = T_PRIVINFLT; |
| 879 | return -2; |
| 880 | } |
| 881 | #endif |
| 882 | if (usermode) |
| 883 | goto nogo; |
| 884 | |
| 885 | map = kernel_map; |
| 886 | } else { |
| 887 | /* |
| 888 | * This is a fault on non-kernel virtual memory. |
| 889 | * vm is initialized above to NULL. If curproc is NULL |
| 890 | * or curproc->p_vmspace is NULL the fault is fatal. |
| 891 | */ |
| 892 | if (p != NULL) |
| 893 | vm = p->p_vmspace; |
| 894 | |
| 895 | if (vm == NULL) |
| 896 | goto nogo; |
| 897 | |
| 898 | map = &vm->vm_map; |
| 899 | } |
| 900 | |
| 901 | if (frame->tf_err & PGEX_W) |
| 902 | ftype = VM_PROT_WRITE; |
| 903 | else |
| 904 | ftype = VM_PROT_READ; |
| 905 | |
| 906 | if (map != kernel_map) { |
| 907 | /* |
| 908 | * Keep swapout from messing with us during this |
| 909 | * critical time. |
| 910 | */ |
| 911 | ++p->p_lock; |
| 912 | |
| 913 | /* |
| 914 | * Grow the stack if necessary |
| 915 | */ |
| 916 | /* grow_stack returns false only if va falls into |
| 917 | * a growable stack region and the stack growth |
| 918 | * fails. It returns true if va was not within |
| 919 | * a growable stack region, or if the stack |
| 920 | * growth succeeded. |
| 921 | */ |
| 922 | if (!grow_stack (p, va)) { |
| 923 | rv = KERN_FAILURE; |
| 924 | --p->p_lock; |
| 925 | goto nogo; |
| 926 | } |
| 927 | |
| 928 | /* Fault in the user page: */ |
| 929 | rv = vm_fault(map, va, ftype, |
| 930 | (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY |
| 931 | : VM_FAULT_NORMAL); |
| 932 | |
| 933 | --p->p_lock; |
| 934 | } else { |
| 935 | /* |
| 936 | * Don't have to worry about process locking or stacks in the kernel. |
| 937 | */ |
| 938 | rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); |
| 939 | } |
| 940 | |
| 941 | if (rv == KERN_SUCCESS) |
| 942 | return (0); |
| 943 | nogo: |
| 944 | if (!usermode) { |
| 945 | if (mycpu->gd_intr_nesting_level == 0 && curthread->td_pcb->pcb_onfault) { |
| 946 | frame->tf_eip = (int)curthread->td_pcb->pcb_onfault; |
| 947 | return (0); |
| 948 | } |
| 949 | trap_fatal(frame, eva); |
| 950 | return (-1); |
| 951 | } |
| 952 | |
| 953 | /* kludge to pass faulting virtual address to sendsig */ |
| 954 | frame->tf_err = eva; |
| 955 | |
| 956 | return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); |
| 957 | } |
| 958 | |
| 959 | static void |
| 960 | trap_fatal(frame, eva) |
| 961 | struct trapframe *frame; |
| 962 | vm_offset_t eva; |
| 963 | { |
| 964 | int code, type, ss, esp; |
| 965 | struct soft_segment_descriptor softseg; |
| 966 | |
| 967 | code = frame->tf_err; |
| 968 | type = frame->tf_trapno; |
| 969 | sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg); |
| 970 | |
| 971 | if (type <= MAX_TRAP_MSG) |
| 972 | printf("\n\nFatal trap %d: %s while in %s mode\n", |
| 973 | type, trap_msg[type], |
| 974 | frame->tf_eflags & PSL_VM ? "vm86" : |
| 975 | ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); |
| 976 | #ifdef SMP |
| 977 | /* three seperate prints in case of a trap on an unmapped page */ |
| 978 | printf("mp_lock = %08x; ", mp_lock); |
| 979 | printf("cpuid = %d; ", mycpu->gd_cpuid); |
| 980 | printf("lapic.id = %08x\n", lapic.id); |
| 981 | #endif |
| 982 | if (type == T_PAGEFLT) { |
| 983 | printf("fault virtual address = 0x%x\n", eva); |
| 984 | printf("fault code = %s %s, %s\n", |
| 985 | code & PGEX_U ? "user" : "supervisor", |
| 986 | code & PGEX_W ? "write" : "read", |
| 987 | code & PGEX_P ? "protection violation" : "page not present"); |
| 988 | } |
| 989 | printf("instruction pointer = 0x%x:0x%x\n", |
| 990 | frame->tf_cs & 0xffff, frame->tf_eip); |
| 991 | if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) { |
| 992 | ss = frame->tf_ss & 0xffff; |
| 993 | esp = frame->tf_esp; |
| 994 | } else { |
| 995 | ss = GSEL(GDATA_SEL, SEL_KPL); |
| 996 | esp = (int)&frame->tf_esp; |
| 997 | } |
| 998 | printf("stack pointer = 0x%x:0x%x\n", ss, esp); |
| 999 | printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp); |
| 1000 | printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n", |
| 1001 | softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); |
| 1002 | printf(" = DPL %d, pres %d, def32 %d, gran %d\n", |
| 1003 | softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32, |
| 1004 | softseg.ssd_gran); |
| 1005 | printf("processor eflags = "); |
| 1006 | if (frame->tf_eflags & PSL_T) |
| 1007 | printf("trace trap, "); |
| 1008 | if (frame->tf_eflags & PSL_I) |
| 1009 | printf("interrupt enabled, "); |
| 1010 | if (frame->tf_eflags & PSL_NT) |
| 1011 | printf("nested task, "); |
| 1012 | if (frame->tf_eflags & PSL_RF) |
| 1013 | printf("resume, "); |
| 1014 | if (frame->tf_eflags & PSL_VM) |
| 1015 | printf("vm86, "); |
| 1016 | printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12); |
| 1017 | printf("current process = "); |
| 1018 | if (curproc) { |
| 1019 | printf("%lu (%s)\n", |
| 1020 | (u_long)curproc->p_pid, curproc->p_comm ? |
| 1021 | curproc->p_comm : ""); |
| 1022 | } else { |
| 1023 | printf("Idle\n"); |
| 1024 | } |
| 1025 | printf("current thread = pri %d ", curthread->td_pri); |
| 1026 | if (curthread->td_pri >= TDPRI_CRIT) |
| 1027 | printf("(CRIT)"); |
| 1028 | printf("\n"); |
| 1029 | printf("interrupt mask = "); |
| 1030 | if ((curthread->td_cpl & net_imask) == net_imask) |
| 1031 | printf("net "); |
| 1032 | if ((curthread->td_cpl & tty_imask) == tty_imask) |
| 1033 | printf("tty "); |
| 1034 | if ((curthread->td_cpl & bio_imask) == bio_imask) |
| 1035 | printf("bio "); |
| 1036 | if ((curthread->td_cpl & cam_imask) == cam_imask) |
| 1037 | printf("cam "); |
| 1038 | if (curthread->td_cpl == 0) |
| 1039 | printf("none"); |
| 1040 | #ifdef SMP |
| 1041 | /** |
| 1042 | * XXX FIXME: |
| 1043 | * we probably SHOULD have stopped the other CPUs before now! |
| 1044 | * another CPU COULD have been touching cpl at this moment... |
| 1045 | */ |
| 1046 | printf(" <- SMP: XXX"); |
| 1047 | #endif |
| 1048 | printf("\n"); |
| 1049 | |
| 1050 | #ifdef KDB |
| 1051 | if (kdb_trap(&psl)) |
| 1052 | return; |
| 1053 | #endif |
| 1054 | #ifdef DDB |
| 1055 | if ((debugger_on_panic || db_active) && kdb_trap(type, 0, frame)) |
| 1056 | return; |
| 1057 | #endif |
| 1058 | printf("trap number = %d\n", type); |
| 1059 | if (type <= MAX_TRAP_MSG) |
| 1060 | panic("%s", trap_msg[type]); |
| 1061 | else |
| 1062 | panic("unknown/reserved trap"); |
| 1063 | } |
| 1064 | |
| 1065 | /* |
| 1066 | * Double fault handler. Called when a fault occurs while writing |
| 1067 | * a frame for a trap/exception onto the stack. This usually occurs |
| 1068 | * when the stack overflows (such is the case with infinite recursion, |
| 1069 | * for example). |
| 1070 | * |
| 1071 | * XXX Note that the current PTD gets replaced by IdlePTD when the |
| 1072 | * task switch occurs. This means that the stack that was active at |
| 1073 | * the time of the double fault is not available at <kstack> unless |
| 1074 | * the machine was idle when the double fault occurred. The downside |
| 1075 | * of this is that "trace <ebp>" in ddb won't work. |
| 1076 | */ |
| 1077 | void |
| 1078 | dblfault_handler() |
| 1079 | { |
| 1080 | struct mdglobaldata *gd = mdcpu; |
| 1081 | |
| 1082 | printf("\nFatal double fault:\n"); |
| 1083 | printf("eip = 0x%x\n", gd->gd_common_tss.tss_eip); |
| 1084 | printf("esp = 0x%x\n", gd->gd_common_tss.tss_esp); |
| 1085 | printf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp); |
| 1086 | #ifdef SMP |
| 1087 | /* three seperate prints in case of a trap on an unmapped page */ |
| 1088 | printf("mp_lock = %08x; ", mp_lock); |
| 1089 | printf("cpuid = %d; ", mycpu->gd_cpuid); |
| 1090 | printf("lapic.id = %08x\n", lapic.id); |
| 1091 | #endif |
| 1092 | panic("double fault"); |
| 1093 | } |
| 1094 | |
| 1095 | /* |
| 1096 | * Compensate for 386 brain damage (missing URKR). |
| 1097 | * This is a little simpler than the pagefault handler in trap() because |
| 1098 | * it the page tables have already been faulted in and high addresses |
| 1099 | * are thrown out early for other reasons. |
| 1100 | */ |
| 1101 | int trapwrite(addr) |
| 1102 | unsigned addr; |
| 1103 | { |
| 1104 | struct proc *p; |
| 1105 | vm_offset_t va; |
| 1106 | struct vmspace *vm; |
| 1107 | int rv; |
| 1108 | |
| 1109 | va = trunc_page((vm_offset_t)addr); |
| 1110 | /* |
| 1111 | * XXX - MAX is END. Changed > to >= for temp. fix. |
| 1112 | */ |
| 1113 | if (va >= VM_MAXUSER_ADDRESS) |
| 1114 | return (1); |
| 1115 | |
| 1116 | p = curproc; |
| 1117 | vm = p->p_vmspace; |
| 1118 | |
| 1119 | ++p->p_lock; |
| 1120 | |
| 1121 | if (!grow_stack (p, va)) { |
| 1122 | --p->p_lock; |
| 1123 | return (1); |
| 1124 | } |
| 1125 | |
| 1126 | /* |
| 1127 | * fault the data page |
| 1128 | */ |
| 1129 | rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY); |
| 1130 | |
| 1131 | --p->p_lock; |
| 1132 | |
| 1133 | if (rv != KERN_SUCCESS) |
| 1134 | return 1; |
| 1135 | |
| 1136 | return (0); |
| 1137 | } |
| 1138 | |
| 1139 | /* |
| 1140 | * syscall2 - MP aware system call request C handler |
| 1141 | * |
| 1142 | * A system call is essentially treated as a trap except that the |
| 1143 | * MP lock is not held on entry or return. We are responsible for |
| 1144 | * obtaining the MP lock if necessary and for handling ASTs |
| 1145 | * (e.g. a task switch) prior to return. |
| 1146 | * |
| 1147 | * In general, only simple access and manipulation of curproc and |
| 1148 | * the current stack is allowed without having to hold MP lock. |
| 1149 | */ |
| 1150 | void |
| 1151 | syscall2(frame) |
| 1152 | struct trapframe frame; |
| 1153 | { |
| 1154 | struct thread *td = curthread; |
| 1155 | struct proc *p = td->td_proc; |
| 1156 | caddr_t params; |
| 1157 | int i; |
| 1158 | struct sysent *callp; |
| 1159 | register_t orig_tf_eflags; |
| 1160 | u_quad_t sticks; |
| 1161 | int error; |
| 1162 | int narg; |
| 1163 | int args[8]; |
| 1164 | u_int code; |
| 1165 | |
| 1166 | #ifdef DIAGNOSTIC |
| 1167 | if (ISPL(frame.tf_cs) != SEL_UPL) { |
| 1168 | get_mplock(); |
| 1169 | panic("syscall"); |
| 1170 | /* NOT REACHED */ |
| 1171 | } |
| 1172 | #endif |
| 1173 | |
| 1174 | #ifdef SMP |
| 1175 | KASSERT(curthread->td_mpcount == 0, ("badmpcount syscall from %p", (void *)frame.tf_eip)); |
| 1176 | get_mplock(); |
| 1177 | #endif |
| 1178 | /* |
| 1179 | * access non-atomic field from critical section. p_sticks is |
| 1180 | * updated by the clock interrupt. Also use this opportunity |
| 1181 | * to lazy-raise our LWKT priority. |
| 1182 | */ |
| 1183 | crit_enter(); |
| 1184 | userenter(); |
| 1185 | sticks = curthread->td_sticks; |
| 1186 | crit_exit(); |
| 1187 | |
| 1188 | p->p_md.md_regs = &frame; |
| 1189 | params = (caddr_t)frame.tf_esp + sizeof(int); |
| 1190 | code = frame.tf_eax; |
| 1191 | orig_tf_eflags = frame.tf_eflags; |
| 1192 | |
| 1193 | if (p->p_sysent->sv_prepsyscall) { |
| 1194 | /* |
| 1195 | * The prep code is not MP aware. |
| 1196 | */ |
| 1197 | (*p->p_sysent->sv_prepsyscall)(&frame, args, &code, ¶ms); |
| 1198 | } else { |
| 1199 | /* |
| 1200 | * Need to check if this is a 32 bit or 64 bit syscall. |
| 1201 | * fuword is MP aware. |
| 1202 | */ |
| 1203 | if (code == SYS_syscall) { |
| 1204 | /* |
| 1205 | * Code is first argument, followed by actual args. |
| 1206 | */ |
| 1207 | code = fuword(params); |
| 1208 | params += sizeof(int); |
| 1209 | } else if (code == SYS___syscall) { |
| 1210 | /* |
| 1211 | * Like syscall, but code is a quad, so as to maintain |
| 1212 | * quad alignment for the rest of the arguments. |
| 1213 | */ |
| 1214 | code = fuword(params); |
| 1215 | params += sizeof(quad_t); |
| 1216 | } |
| 1217 | } |
| 1218 | |
| 1219 | if (p->p_sysent->sv_mask) |
| 1220 | code &= p->p_sysent->sv_mask; |
| 1221 | |
| 1222 | if (code >= p->p_sysent->sv_size) |
| 1223 | callp = &p->p_sysent->sv_table[0]; |
| 1224 | else |
| 1225 | callp = &p->p_sysent->sv_table[code]; |
| 1226 | |
| 1227 | narg = callp->sy_narg & SYF_ARGMASK; |
| 1228 | |
| 1229 | /* |
| 1230 | * copyin is MP aware, but the tracing code is not |
| 1231 | */ |
| 1232 | if (params && (i = narg * sizeof(int)) && |
| 1233 | (error = copyin(params, (caddr_t)args, (u_int)i))) { |
| 1234 | #ifdef KTRACE |
| 1235 | if (KTRPOINT(td, KTR_SYSCALL)) |
| 1236 | ktrsyscall(p->p_tracep, code, narg, args); |
| 1237 | #endif |
| 1238 | goto bad; |
| 1239 | } |
| 1240 | |
| 1241 | #if 0 |
| 1242 | /* |
| 1243 | * Try to run the syscall without the MP lock if the syscall |
| 1244 | * is MP safe. We have to obtain the MP lock no matter what if |
| 1245 | * we are ktracing |
| 1246 | */ |
| 1247 | if ((callp->sy_narg & SYF_MPSAFE) == 0) { |
| 1248 | get_mplock(); |
| 1249 | have_mplock = 1; |
| 1250 | } |
| 1251 | #endif |
| 1252 | |
| 1253 | #ifdef KTRACE |
| 1254 | if (KTRPOINT(td, KTR_SYSCALL)) { |
| 1255 | ktrsyscall(p->p_tracep, code, narg, args); |
| 1256 | } |
| 1257 | #endif |
| 1258 | p->p_retval[0] = 0; |
| 1259 | p->p_retval[1] = frame.tf_edx; |
| 1260 | |
| 1261 | STOPEVENT(p, S_SCE, narg); /* MP aware */ |
| 1262 | |
| 1263 | error = (*callp->sy_call)(args); |
| 1264 | |
| 1265 | /* |
| 1266 | * MP SAFE (we may or may not have the MP lock at this point) |
| 1267 | */ |
| 1268 | switch (error) { |
| 1269 | case 0: |
| 1270 | /* |
| 1271 | * Reinitialize proc pointer `p' as it may be different |
| 1272 | * if this is a child returning from fork syscall. |
| 1273 | */ |
| 1274 | p = curproc; |
| 1275 | frame.tf_eax = p->p_retval[0]; |
| 1276 | frame.tf_edx = p->p_retval[1]; |
| 1277 | frame.tf_eflags &= ~PSL_C; |
| 1278 | break; |
| 1279 | |
| 1280 | case ERESTART: |
| 1281 | /* |
| 1282 | * Reconstruct pc, assuming lcall $X,y is 7 bytes, |
| 1283 | * int 0x80 is 2 bytes. We saved this in tf_err. |
| 1284 | */ |
| 1285 | frame.tf_eip -= frame.tf_err; |
| 1286 | break; |
| 1287 | |
| 1288 | case EJUSTRETURN: |
| 1289 | break; |
| 1290 | |
| 1291 | default: |
| 1292 | bad: |
| 1293 | if (p->p_sysent->sv_errsize) { |
| 1294 | if (error >= p->p_sysent->sv_errsize) |
| 1295 | error = -1; /* XXX */ |
| 1296 | else |
| 1297 | error = p->p_sysent->sv_errtbl[error]; |
| 1298 | } |
| 1299 | frame.tf_eax = error; |
| 1300 | frame.tf_eflags |= PSL_C; |
| 1301 | break; |
| 1302 | } |
| 1303 | |
| 1304 | /* |
| 1305 | * Traced syscall. trapsignal() is not MP aware. |
| 1306 | */ |
| 1307 | if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) { |
| 1308 | frame.tf_eflags &= ~PSL_T; |
| 1309 | trapsignal(p, SIGTRAP, 0); |
| 1310 | } |
| 1311 | |
| 1312 | /* |
| 1313 | * Handle reschedule and other end-of-syscall issues |
| 1314 | */ |
| 1315 | userret(p, &frame, sticks); |
| 1316 | |
| 1317 | #ifdef KTRACE |
| 1318 | if (KTRPOINT(td, KTR_SYSRET)) { |
| 1319 | ktrsysret(p->p_tracep, code, error, p->p_retval[0]); |
| 1320 | } |
| 1321 | #endif |
| 1322 | |
| 1323 | /* |
| 1324 | * This works because errno is findable through the |
| 1325 | * register set. If we ever support an emulation where this |
| 1326 | * is not the case, this code will need to be revisited. |
| 1327 | */ |
| 1328 | STOPEVENT(p, S_SCX, code); |
| 1329 | |
| 1330 | #ifdef SMP |
| 1331 | /* |
| 1332 | * Release the MP lock if we had to get it |
| 1333 | */ |
| 1334 | KASSERT(curthread->td_mpcount == 1, ("badmpcount syscall from %p", (void *)frame.tf_eip)); |
| 1335 | rel_mplock(); |
| 1336 | #endif |
| 1337 | } |
| 1338 | |
| 1339 | /* |
| 1340 | * Simplified back end of syscall(), used when returning from fork() |
| 1341 | * directly into user mode. MP lock is held on entry and should be |
| 1342 | * released on return. This code will return back into the fork |
| 1343 | * trampoline code which then runs doreti. |
| 1344 | */ |
| 1345 | void |
| 1346 | fork_return(p, frame) |
| 1347 | struct proc *p; |
| 1348 | struct trapframe frame; |
| 1349 | { |
| 1350 | frame.tf_eax = 0; /* Child returns zero */ |
| 1351 | frame.tf_eflags &= ~PSL_C; /* success */ |
| 1352 | frame.tf_edx = 1; |
| 1353 | |
| 1354 | userret(p, &frame, 0); |
| 1355 | #ifdef KTRACE |
| 1356 | if (KTRPOINT(p->p_thread, KTR_SYSRET)) |
| 1357 | ktrsysret(p->p_tracep, SYS_fork, 0, 0); |
| 1358 | #endif |
| 1359 | #ifdef SMP |
| 1360 | KKASSERT(curthread->td_mpcount == 1); |
| 1361 | rel_mplock(); |
| 1362 | #endif |
| 1363 | } |
| 1364 | |