| 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/platform/vkernel/i386/trap.c,v 1.35 2008/09/09 04:06:19 dillon Exp $ |
| 40 | */ |
| 41 | |
| 42 | /* |
| 43 | * 386 Trap and System call handling |
| 44 | */ |
| 45 | |
| 46 | #include "use_isa.h" |
| 47 | #include "use_npx.h" |
| 48 | |
| 49 | #include "opt_ddb.h" |
| 50 | #include "opt_ktrace.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/signal2.h> |
| 60 | #include <sys/syscall.h> |
| 61 | #include <sys/sysctl.h> |
| 62 | #include <sys/sysent.h> |
| 63 | #include <sys/uio.h> |
| 64 | #include <sys/vmmeter.h> |
| 65 | #include <sys/malloc.h> |
| 66 | #ifdef KTRACE |
| 67 | #include <sys/ktrace.h> |
| 68 | #endif |
| 69 | #include <sys/ktr.h> |
| 70 | #include <sys/upcall.h> |
| 71 | #include <sys/vkernel.h> |
| 72 | #include <sys/sysproto.h> |
| 73 | #include <sys/sysunion.h> |
| 74 | #include <sys/vmspace.h> |
| 75 | |
| 76 | #include <vm/vm.h> |
| 77 | #include <vm/vm_param.h> |
| 78 | #include <sys/lock.h> |
| 79 | #include <vm/pmap.h> |
| 80 | #include <vm/vm_kern.h> |
| 81 | #include <vm/vm_map.h> |
| 82 | #include <vm/vm_page.h> |
| 83 | #include <vm/vm_extern.h> |
| 84 | |
| 85 | #include <machine/cpu.h> |
| 86 | #include <machine/md_var.h> |
| 87 | #include <machine/pcb.h> |
| 88 | #include <machine/smp.h> |
| 89 | #include <machine/tss.h> |
| 90 | #include <machine/globaldata.h> |
| 91 | |
| 92 | #include <machine/vm86.h> |
| 93 | |
| 94 | #include <ddb/ddb.h> |
| 95 | |
| 96 | #include <sys/msgport2.h> |
| 97 | #include <sys/thread2.h> |
| 98 | #include <sys/mplock2.h> |
| 99 | |
| 100 | #ifdef SMP |
| 101 | |
| 102 | #define MAKEMPSAFE(have_mplock) \ |
| 103 | if (have_mplock == 0) { \ |
| 104 | get_mplock(); \ |
| 105 | have_mplock = 1; \ |
| 106 | } |
| 107 | |
| 108 | #else |
| 109 | |
| 110 | #define MAKEMPSAFE(have_mplock) |
| 111 | |
| 112 | #endif |
| 113 | |
| 114 | int (*pmath_emulate) (struct trapframe *); |
| 115 | |
| 116 | static int trap_pfault (struct trapframe *, int, vm_offset_t); |
| 117 | static void trap_fatal (struct trapframe *, int, vm_offset_t); |
| 118 | void dblfault_handler (void); |
| 119 | |
| 120 | #if 0 |
| 121 | extern inthand_t IDTVEC(syscall); |
| 122 | #endif |
| 123 | |
| 124 | #define MAX_TRAP_MSG 28 |
| 125 | static char *trap_msg[] = { |
| 126 | "", /* 0 unused */ |
| 127 | "privileged instruction fault", /* 1 T_PRIVINFLT */ |
| 128 | "", /* 2 unused */ |
| 129 | "breakpoint instruction fault", /* 3 T_BPTFLT */ |
| 130 | "", /* 4 unused */ |
| 131 | "", /* 5 unused */ |
| 132 | "arithmetic trap", /* 6 T_ARITHTRAP */ |
| 133 | "system forced exception", /* 7 T_ASTFLT */ |
| 134 | "", /* 8 unused */ |
| 135 | "general protection fault", /* 9 T_PROTFLT */ |
| 136 | "trace trap", /* 10 T_TRCTRAP */ |
| 137 | "", /* 11 unused */ |
| 138 | "page fault", /* 12 T_PAGEFLT */ |
| 139 | "", /* 13 unused */ |
| 140 | "alignment fault", /* 14 T_ALIGNFLT */ |
| 141 | "", /* 15 unused */ |
| 142 | "", /* 16 unused */ |
| 143 | "", /* 17 unused */ |
| 144 | "integer divide fault", /* 18 T_DIVIDE */ |
| 145 | "non-maskable interrupt trap", /* 19 T_NMI */ |
| 146 | "overflow trap", /* 20 T_OFLOW */ |
| 147 | "FPU bounds check fault", /* 21 T_BOUND */ |
| 148 | "FPU device not available", /* 22 T_DNA */ |
| 149 | "double fault", /* 23 T_DOUBLEFLT */ |
| 150 | "FPU operand fetch fault", /* 24 T_FPOPFLT */ |
| 151 | "invalid TSS fault", /* 25 T_TSSFLT */ |
| 152 | "segment not present fault", /* 26 T_SEGNPFLT */ |
| 153 | "stack fault", /* 27 T_STKFLT */ |
| 154 | "machine check trap", /* 28 T_MCHK */ |
| 155 | }; |
| 156 | |
| 157 | #ifdef DDB |
| 158 | static int ddb_on_nmi = 1; |
| 159 | SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW, |
| 160 | &ddb_on_nmi, 0, "Go to DDB on NMI"); |
| 161 | #endif |
| 162 | static int panic_on_nmi = 1; |
| 163 | SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW, |
| 164 | &panic_on_nmi, 0, "Panic on NMI"); |
| 165 | static int fast_release; |
| 166 | SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW, |
| 167 | &fast_release, 0, "Passive Release was optimal"); |
| 168 | static int slow_release; |
| 169 | SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW, |
| 170 | &slow_release, 0, "Passive Release was nonoptimal"); |
| 171 | |
| 172 | MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure"); |
| 173 | extern int max_sysmsg; |
| 174 | |
| 175 | /* |
| 176 | * Passively intercepts the thread switch function to increase |
| 177 | * the thread priority from a user priority to a kernel priority, reducing |
| 178 | * syscall and trap overhead for the case where no switch occurs. |
| 179 | * |
| 180 | * Synchronizes td_ucred with p_ucred. This is used by system calls, |
| 181 | * signal handling, faults, AST traps, and anything else that enters the |
| 182 | * kernel from userland and provides the kernel with a stable read-only |
| 183 | * copy of the process ucred. |
| 184 | */ |
| 185 | static __inline void |
| 186 | userenter(struct thread *curtd, struct proc *curp) |
| 187 | { |
| 188 | struct ucred *ocred; |
| 189 | struct ucred *ncred; |
| 190 | |
| 191 | curtd->td_release = lwkt_passive_release; |
| 192 | |
| 193 | if (curtd->td_ucred != curp->p_ucred) { |
| 194 | ncred = crhold(curp->p_ucred); |
| 195 | ocred = curtd->td_ucred; |
| 196 | curtd->td_ucred = ncred; |
| 197 | if (ocred) |
| 198 | crfree(ocred); |
| 199 | } |
| 200 | } |
| 201 | |
| 202 | /* |
| 203 | * Handle signals, upcalls, profiling, and other AST's and/or tasks that |
| 204 | * must be completed before we can return to or try to return to userland. |
| 205 | * |
| 206 | * Note that td_sticks is a 64 bit quantity, but there's no point doing 64 |
| 207 | * arithmatic on the delta calculation so the absolute tick values are |
| 208 | * truncated to an integer. |
| 209 | */ |
| 210 | static void |
| 211 | userret(struct lwp *lp, struct trapframe *frame, int sticks) |
| 212 | { |
| 213 | struct proc *p = lp->lwp_proc; |
| 214 | int sig; |
| 215 | |
| 216 | /* |
| 217 | * Charge system time if profiling. Note: times are in microseconds. |
| 218 | * This may do a copyout and block, so do it first even though it |
| 219 | * means some system time will be charged as user time. |
| 220 | */ |
| 221 | if (p->p_flags & P_PROFIL) { |
| 222 | addupc_task(p, frame->tf_eip, |
| 223 | (u_int)((int)lp->lwp_thread->td_sticks - sticks)); |
| 224 | } |
| 225 | |
| 226 | recheck: |
| 227 | /* |
| 228 | * If the jungle wants us dead, so be it. |
| 229 | */ |
| 230 | if (lp->lwp_mpflags & LWP_MP_WEXIT) { |
| 231 | lwkt_gettoken(&p->p_token); |
| 232 | lwp_exit(0); |
| 233 | lwkt_reltoken(&p->p_token); /* NOT REACHED */ |
| 234 | } |
| 235 | |
| 236 | /* |
| 237 | * Block here if we are in a stopped state. |
| 238 | */ |
| 239 | if (p->p_stat == SSTOP) { |
| 240 | lwkt_gettoken(&p->p_token); |
| 241 | tstop(); |
| 242 | lwkt_reltoken(&p->p_token); |
| 243 | goto recheck; |
| 244 | } |
| 245 | |
| 246 | /* |
| 247 | * Post any pending upcalls. If running a virtual kernel be sure |
| 248 | * to restore the virtual kernel's vmspace before posting the upcall. |
| 249 | */ |
| 250 | if (p->p_flags & (P_SIGVTALRM | P_SIGPROF | P_UPCALLPEND)) { |
| 251 | lwkt_gettoken(&p->p_token); |
| 252 | if (p->p_flags & P_SIGVTALRM) { |
| 253 | p->p_flags &= ~P_SIGVTALRM; |
| 254 | ksignal(p, SIGVTALRM); |
| 255 | } |
| 256 | if (p->p_flags & P_SIGPROF) { |
| 257 | p->p_flags &= ~P_SIGPROF; |
| 258 | ksignal(p, SIGPROF); |
| 259 | } |
| 260 | if (p->p_flags & P_UPCALLPEND) { |
| 261 | p->p_flags &= ~P_UPCALLPEND; |
| 262 | postupcall(lp); |
| 263 | } |
| 264 | lwkt_reltoken(&p->p_token); |
| 265 | goto recheck; |
| 266 | } |
| 267 | |
| 268 | /* |
| 269 | * Post any pending signals |
| 270 | * |
| 271 | * WARNING! postsig() can exit and not return. |
| 272 | */ |
| 273 | if ((sig = CURSIG_TRACE(lp)) != 0) { |
| 274 | lwkt_gettoken(&p->p_token); |
| 275 | postsig(sig); |
| 276 | lwkt_reltoken(&p->p_token); |
| 277 | goto recheck; |
| 278 | } |
| 279 | |
| 280 | /* |
| 281 | * block here if we are swapped out, but still process signals |
| 282 | * (such as SIGKILL). proc0 (the swapin scheduler) is already |
| 283 | * aware of our situation, we do not have to wake it up. |
| 284 | */ |
| 285 | if (p->p_flags & P_SWAPPEDOUT) { |
| 286 | lwkt_gettoken(&p->p_token); |
| 287 | get_mplock(); |
| 288 | p->p_flags |= P_SWAPWAIT; |
| 289 | swapin_request(); |
| 290 | if (p->p_flags & P_SWAPWAIT) |
| 291 | tsleep(p, PCATCH, "SWOUT", 0); |
| 292 | p->p_flags &= ~P_SWAPWAIT; |
| 293 | rel_mplock(); |
| 294 | lwkt_reltoken(&p->p_token); |
| 295 | goto recheck; |
| 296 | } |
| 297 | |
| 298 | /* |
| 299 | * Make sure postsig() handled request to restore old signal mask after |
| 300 | * running signal handler. |
| 301 | */ |
| 302 | KKASSERT((lp->lwp_flags & LWP_OLDMASK) == 0); |
| 303 | } |
| 304 | |
| 305 | /* |
| 306 | * Cleanup from userenter and any passive release that might have occured. |
| 307 | * We must reclaim the current-process designation before we can return |
| 308 | * to usermode. We also handle both LWKT and USER reschedule requests. |
| 309 | */ |
| 310 | static __inline void |
| 311 | userexit(struct lwp *lp) |
| 312 | { |
| 313 | struct thread *td = lp->lwp_thread; |
| 314 | /* globaldata_t gd = td->td_gd; */ |
| 315 | |
| 316 | /* |
| 317 | * Handle stop requests at kernel priority. Any requests queued |
| 318 | * after this loop will generate another AST. |
| 319 | */ |
| 320 | while (lp->lwp_proc->p_stat == SSTOP) { |
| 321 | lwkt_gettoken(&lp->lwp_proc->p_token); |
| 322 | tstop(); |
| 323 | lwkt_reltoken(&lp->lwp_proc->p_token); |
| 324 | } |
| 325 | |
| 326 | /* |
| 327 | * Reduce our priority in preparation for a return to userland. If |
| 328 | * our passive release function was still in place, our priority was |
| 329 | * never raised and does not need to be reduced. |
| 330 | */ |
| 331 | lwkt_passive_recover(td); |
| 332 | |
| 333 | /* |
| 334 | * Become the current user scheduled process if we aren't already, |
| 335 | * and deal with reschedule requests and other factors. |
| 336 | */ |
| 337 | lp->lwp_proc->p_usched->acquire_curproc(lp); |
| 338 | /* WARNING: we may have migrated cpu's */ |
| 339 | /* gd = td->td_gd; */ |
| 340 | } |
| 341 | |
| 342 | #if !defined(KTR_KERNENTRY) |
| 343 | #define KTR_KERNENTRY KTR_ALL |
| 344 | #endif |
| 345 | KTR_INFO_MASTER(kernentry); |
| 346 | KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p", |
| 347 | sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t)); |
| 348 | KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d", |
| 349 | sizeof(int) + sizeof(int)); |
| 350 | KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d", |
| 351 | sizeof(int) + sizeof(int) + sizeof(int)); |
| 352 | KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d", |
| 353 | sizeof(int) + sizeof(int) + sizeof(int)); |
| 354 | KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d", |
| 355 | sizeof(int) + sizeof(int)); |
| 356 | |
| 357 | /* |
| 358 | * Exception, fault, and trap interface to the kernel. |
| 359 | * This common code is called from assembly language IDT gate entry |
| 360 | * routines that prepare a suitable stack frame, and restore this |
| 361 | * frame after the exception has been processed. |
| 362 | * |
| 363 | * This function is also called from doreti in an interlock to handle ASTs. |
| 364 | * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap |
| 365 | * |
| 366 | * NOTE! We have to retrieve the fault address prior to obtaining the |
| 367 | * MP lock because get_mplock() may switch out. YYY cr2 really ought |
| 368 | * to be retrieved by the assembly code, not here. |
| 369 | * |
| 370 | * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing |
| 371 | * if an attempt is made to switch from a fast interrupt or IPI. This is |
| 372 | * necessary to properly take fatal kernel traps on SMP machines if |
| 373 | * get_mplock() has to block. |
| 374 | */ |
| 375 | |
| 376 | void |
| 377 | user_trap(struct trapframe *frame) |
| 378 | { |
| 379 | struct globaldata *gd = mycpu; |
| 380 | struct thread *td = gd->gd_curthread; |
| 381 | struct lwp *lp = td->td_lwp; |
| 382 | struct proc *p; |
| 383 | int sticks = 0; |
| 384 | int i = 0, ucode = 0, type, code; |
| 385 | #ifdef SMP |
| 386 | int have_mplock = 0; |
| 387 | #endif |
| 388 | #ifdef INVARIANTS |
| 389 | int crit_count = td->td_critcount; |
| 390 | lwkt_tokref_t curstop = td->td_toks_stop; |
| 391 | #endif |
| 392 | vm_offset_t eva; |
| 393 | |
| 394 | p = td->td_proc; |
| 395 | |
| 396 | /* |
| 397 | * This is a bad kludge to avoid changing the various trapframe |
| 398 | * structures. Because we are enabled as a virtual kernel, |
| 399 | * the original tf_err field will be passed to us shifted 16 |
| 400 | * over in the tf_trapno field for T_PAGEFLT. |
| 401 | */ |
| 402 | if (frame->tf_trapno == T_PAGEFLT) |
| 403 | eva = frame->tf_err; |
| 404 | else |
| 405 | eva = 0; |
| 406 | #if 0 |
| 407 | kprintf("USER_TRAP AT %08x xflags %d trapno %d eva %08x\n", |
| 408 | frame->tf_eip, frame->tf_xflags, frame->tf_trapno, eva); |
| 409 | #endif |
| 410 | |
| 411 | /* |
| 412 | * Everything coming from user mode runs through user_trap, |
| 413 | * including system calls. |
| 414 | */ |
| 415 | if (frame->tf_trapno == T_SYSCALL80) { |
| 416 | syscall2(frame); |
| 417 | return; |
| 418 | } |
| 419 | |
| 420 | KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid, |
| 421 | frame->tf_trapno, eva); |
| 422 | |
| 423 | #ifdef DDB |
| 424 | if (db_active) { |
| 425 | eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0); |
| 426 | ++gd->gd_trap_nesting_level; |
| 427 | MAKEMPSAFE(have_mplock); |
| 428 | trap_fatal(frame, TRUE, eva); |
| 429 | --gd->gd_trap_nesting_level; |
| 430 | goto out2; |
| 431 | } |
| 432 | #endif |
| 433 | |
| 434 | #if defined(I586_CPU) && !defined(NO_F00F_HACK) |
| 435 | restart: |
| 436 | #endif |
| 437 | type = frame->tf_trapno; |
| 438 | code = frame->tf_err; |
| 439 | |
| 440 | userenter(td, p); |
| 441 | |
| 442 | sticks = (int)td->td_sticks; |
| 443 | lp->lwp_md.md_regs = frame; |
| 444 | |
| 445 | switch (type) { |
| 446 | case T_PRIVINFLT: /* privileged instruction fault */ |
| 447 | ucode = ILL_PRVOPC; |
| 448 | i = SIGILL; |
| 449 | break; |
| 450 | |
| 451 | case T_BPTFLT: /* bpt instruction fault */ |
| 452 | case T_TRCTRAP: /* trace trap */ |
| 453 | frame->tf_eflags &= ~PSL_T; |
| 454 | ucode = TRAP_TRACE; |
| 455 | i = SIGTRAP; |
| 456 | break; |
| 457 | |
| 458 | case T_ARITHTRAP: /* arithmetic trap */ |
| 459 | ucode = code; |
| 460 | i = SIGFPE; |
| 461 | break; |
| 462 | |
| 463 | case T_ASTFLT: /* Allow process switch */ |
| 464 | mycpu->gd_cnt.v_soft++; |
| 465 | if (mycpu->gd_reqflags & RQF_AST_OWEUPC) { |
| 466 | atomic_clear_int(&mycpu->gd_reqflags, |
| 467 | RQF_AST_OWEUPC); |
| 468 | addupc_task(p, p->p_prof.pr_addr, |
| 469 | p->p_prof.pr_ticks); |
| 470 | } |
| 471 | goto out; |
| 472 | |
| 473 | /* |
| 474 | * The following two traps can happen in |
| 475 | * vm86 mode, and, if so, we want to handle |
| 476 | * them specially. |
| 477 | */ |
| 478 | case T_PROTFLT: /* general protection fault */ |
| 479 | case T_STKFLT: /* stack fault */ |
| 480 | #if 0 |
| 481 | if (frame->tf_eflags & PSL_VM) { |
| 482 | i = vm86_emulate((struct vm86frame *)frame); |
| 483 | if (i == 0) |
| 484 | goto out; |
| 485 | break; |
| 486 | } |
| 487 | #endif |
| 488 | i = SIGBUS; |
| 489 | ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR; |
| 490 | break; |
| 491 | case T_SEGNPFLT: /* segment not present fault */ |
| 492 | i = SIGBUS; |
| 493 | ucode = BUS_ADRERR; |
| 494 | break; |
| 495 | case T_TSSFLT: /* invalid TSS fault */ |
| 496 | case T_DOUBLEFLT: /* double fault */ |
| 497 | i = SIGBUS; |
| 498 | ucode = BUS_OBJERR; |
| 499 | default: |
| 500 | #if 0 |
| 501 | ucode = code + BUS_SEGM_FAULT ; /* XXX: ???*/ |
| 502 | #endif |
| 503 | ucode = BUS_OBJERR; |
| 504 | i = SIGBUS; |
| 505 | break; |
| 506 | |
| 507 | case T_PAGEFLT: /* page fault */ |
| 508 | MAKEMPSAFE(have_mplock); |
| 509 | i = trap_pfault(frame, TRUE, eva); |
| 510 | if (i == -1) |
| 511 | goto out; |
| 512 | #if defined(I586_CPU) && !defined(NO_F00F_HACK) |
| 513 | if (i == -2) |
| 514 | goto restart; |
| 515 | #endif |
| 516 | if (i == 0) |
| 517 | goto out; |
| 518 | |
| 519 | #if 0 |
| 520 | ucode = T_PAGEFLT; |
| 521 | #endif |
| 522 | if (i == SIGSEGV) |
| 523 | ucode = SEGV_MAPERR; |
| 524 | else |
| 525 | ucode = BUS_ADRERR; |
| 526 | break; |
| 527 | |
| 528 | case T_DIVIDE: /* integer divide fault */ |
| 529 | ucode = FPE_INTDIV; |
| 530 | i = SIGFPE; |
| 531 | break; |
| 532 | |
| 533 | #if NISA > 0 |
| 534 | case T_NMI: |
| 535 | MAKEMPSAFE(have_mplock); |
| 536 | /* machine/parity/power fail/"kitchen sink" faults */ |
| 537 | if (isa_nmi(code) == 0) { |
| 538 | #ifdef DDB |
| 539 | /* |
| 540 | * NMI can be hooked up to a pushbutton |
| 541 | * for debugging. |
| 542 | */ |
| 543 | if (ddb_on_nmi) { |
| 544 | kprintf ("NMI ... going to debugger\n"); |
| 545 | kdb_trap (type, 0, frame); |
| 546 | } |
| 547 | #endif /* DDB */ |
| 548 | goto out2; |
| 549 | } else if (panic_on_nmi) |
| 550 | panic("NMI indicates hardware failure"); |
| 551 | break; |
| 552 | #endif /* NISA > 0 */ |
| 553 | |
| 554 | case T_OFLOW: /* integer overflow fault */ |
| 555 | ucode = FPE_INTOVF; |
| 556 | i = SIGFPE; |
| 557 | break; |
| 558 | |
| 559 | case T_BOUND: /* bounds check fault */ |
| 560 | ucode = FPE_FLTSUB; |
| 561 | i = SIGFPE; |
| 562 | break; |
| 563 | |
| 564 | case T_DNA: |
| 565 | /* |
| 566 | * Virtual kernel intercept - pass the DNA exception |
| 567 | * to the (emulated) virtual kernel if it asked to handle |
| 568 | * it. This occurs when the virtual kernel is holding |
| 569 | * onto the FP context for a different emulated |
| 570 | * process then the one currently running. |
| 571 | * |
| 572 | * We must still call npxdna() since we may have |
| 573 | * saved FP state that the (emulated) virtual kernel |
| 574 | * needs to hand over to a different emulated process. |
| 575 | */ |
| 576 | if (lp->lwp_vkernel && lp->lwp_vkernel->ve && |
| 577 | (td->td_pcb->pcb_flags & FP_VIRTFP) |
| 578 | ) { |
| 579 | npxdna(frame); |
| 580 | break; |
| 581 | } |
| 582 | #if NNPX > 0 |
| 583 | /* |
| 584 | * The kernel may have switched out the FP unit's |
| 585 | * state, causing the user process to take a fault |
| 586 | * when it tries to use the FP unit. Restore the |
| 587 | * state here |
| 588 | */ |
| 589 | if (npxdna(frame)) |
| 590 | goto out; |
| 591 | #endif |
| 592 | if (!pmath_emulate) { |
| 593 | i = SIGFPE; |
| 594 | ucode = FPE_FPU_NP_TRAP; |
| 595 | break; |
| 596 | } |
| 597 | i = (*pmath_emulate)(frame); |
| 598 | if (i == 0) { |
| 599 | if (!(frame->tf_eflags & PSL_T)) |
| 600 | goto out2; |
| 601 | frame->tf_eflags &= ~PSL_T; |
| 602 | i = SIGTRAP; |
| 603 | } |
| 604 | /* else ucode = emulator_only_knows() XXX */ |
| 605 | break; |
| 606 | |
| 607 | case T_FPOPFLT: /* FPU operand fetch fault */ |
| 608 | ucode = ILL_COPROC; |
| 609 | i = SIGILL; |
| 610 | break; |
| 611 | |
| 612 | case T_XMMFLT: /* SIMD floating-point exception */ |
| 613 | ucode = 0; /* XXX */ |
| 614 | i = SIGFPE; |
| 615 | break; |
| 616 | } |
| 617 | |
| 618 | /* |
| 619 | * Virtual kernel intercept - if the fault is directly related to a |
| 620 | * VM context managed by a virtual kernel then let the virtual kernel |
| 621 | * handle it. |
| 622 | */ |
| 623 | if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { |
| 624 | vkernel_trap(lp, frame); |
| 625 | goto out; |
| 626 | } |
| 627 | |
| 628 | /* |
| 629 | * Translate fault for emulators (e.g. Linux) |
| 630 | */ |
| 631 | if (*p->p_sysent->sv_transtrap) |
| 632 | i = (*p->p_sysent->sv_transtrap)(i, type); |
| 633 | |
| 634 | MAKEMPSAFE(have_mplock); |
| 635 | trapsignal(lp, i, ucode); |
| 636 | |
| 637 | #ifdef DEBUG |
| 638 | if (type <= MAX_TRAP_MSG) { |
| 639 | uprintf("fatal process exception: %s", |
| 640 | trap_msg[type]); |
| 641 | if ((type == T_PAGEFLT) || (type == T_PROTFLT)) |
| 642 | uprintf(", fault VA = 0x%lx", (u_long)eva); |
| 643 | uprintf("\n"); |
| 644 | } |
| 645 | #endif |
| 646 | |
| 647 | out: |
| 648 | userret(lp, frame, sticks); |
| 649 | userexit(lp); |
| 650 | out2: ; |
| 651 | #ifdef SMP |
| 652 | if (have_mplock) |
| 653 | rel_mplock(); |
| 654 | #endif |
| 655 | KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid); |
| 656 | #ifdef INVARIANTS |
| 657 | KASSERT(crit_count == td->td_critcount, |
| 658 | ("trap: critical section count mismatch! %d/%d", |
| 659 | crit_count, td->td_pri)); |
| 660 | KASSERT(curstop == td->td_toks_stop, |
| 661 | ("trap: extra tokens held after trap! %zd/%zd", |
| 662 | curstop - &td->td_toks_base, |
| 663 | td->td_toks_stop - &td->td_toks_base)); |
| 664 | #endif |
| 665 | } |
| 666 | |
| 667 | void |
| 668 | kern_trap(struct trapframe *frame) |
| 669 | { |
| 670 | struct globaldata *gd = mycpu; |
| 671 | struct thread *td = gd->gd_curthread; |
| 672 | struct lwp *lp; |
| 673 | struct proc *p; |
| 674 | int i = 0, ucode = 0, type, code; |
| 675 | #ifdef SMP |
| 676 | int have_mplock = 0; |
| 677 | #endif |
| 678 | #ifdef INVARIANTS |
| 679 | int crit_count = td->td_critcount; |
| 680 | lwkt_tokref_t curstop = td->td_toks_stop; |
| 681 | #endif |
| 682 | vm_offset_t eva; |
| 683 | |
| 684 | lp = td->td_lwp; |
| 685 | p = td->td_proc; |
| 686 | |
| 687 | if (frame->tf_trapno == T_PAGEFLT) |
| 688 | eva = frame->tf_err; |
| 689 | else |
| 690 | eva = 0; |
| 691 | |
| 692 | #ifdef DDB |
| 693 | if (db_active) { |
| 694 | ++gd->gd_trap_nesting_level; |
| 695 | MAKEMPSAFE(have_mplock); |
| 696 | trap_fatal(frame, FALSE, eva); |
| 697 | --gd->gd_trap_nesting_level; |
| 698 | goto out2; |
| 699 | } |
| 700 | #endif |
| 701 | type = frame->tf_trapno; |
| 702 | code = frame->tf_err; |
| 703 | |
| 704 | #if 0 |
| 705 | kernel_trap: |
| 706 | #endif |
| 707 | /* kernel trap */ |
| 708 | |
| 709 | switch (type) { |
| 710 | case T_PAGEFLT: /* page fault */ |
| 711 | MAKEMPSAFE(have_mplock); |
| 712 | trap_pfault(frame, FALSE, eva); |
| 713 | goto out2; |
| 714 | |
| 715 | case T_DNA: |
| 716 | #if NNPX > 0 |
| 717 | /* |
| 718 | * The kernel may be using npx for copying or other |
| 719 | * purposes. |
| 720 | */ |
| 721 | panic("kernel NPX should not happen"); |
| 722 | if (npxdna(frame)) |
| 723 | goto out2; |
| 724 | #endif |
| 725 | break; |
| 726 | |
| 727 | case T_PROTFLT: /* general protection fault */ |
| 728 | case T_SEGNPFLT: /* segment not present fault */ |
| 729 | /* |
| 730 | * Invalid segment selectors and out of bounds |
| 731 | * %eip's and %esp's can be set up in user mode. |
| 732 | * This causes a fault in kernel mode when the |
| 733 | * kernel tries to return to user mode. We want |
| 734 | * to get this fault so that we can fix the |
| 735 | * problem here and not have to check all the |
| 736 | * selectors and pointers when the user changes |
| 737 | * them. |
| 738 | */ |
| 739 | if (mycpu->gd_intr_nesting_level == 0) { |
| 740 | if (td->td_pcb->pcb_onfault) { |
| 741 | frame->tf_eip = |
| 742 | (register_t)td->td_pcb->pcb_onfault; |
| 743 | goto out2; |
| 744 | } |
| 745 | } |
| 746 | break; |
| 747 | |
| 748 | case T_TSSFLT: |
| 749 | /* |
| 750 | * PSL_NT can be set in user mode and isn't cleared |
| 751 | * automatically when the kernel is entered. This |
| 752 | * causes a TSS fault when the kernel attempts to |
| 753 | * `iret' because the TSS link is uninitialized. We |
| 754 | * want to get this fault so that we can fix the |
| 755 | * problem here and not every time the kernel is |
| 756 | * entered. |
| 757 | */ |
| 758 | if (frame->tf_eflags & PSL_NT) { |
| 759 | frame->tf_eflags &= ~PSL_NT; |
| 760 | goto out2; |
| 761 | } |
| 762 | break; |
| 763 | |
| 764 | case T_TRCTRAP: /* trace trap */ |
| 765 | #if 0 |
| 766 | if (frame->tf_eip == (int)IDTVEC(syscall)) { |
| 767 | /* |
| 768 | * We've just entered system mode via the |
| 769 | * syscall lcall. Continue single stepping |
| 770 | * silently until the syscall handler has |
| 771 | * saved the flags. |
| 772 | */ |
| 773 | goto out2; |
| 774 | } |
| 775 | if (frame->tf_eip == (int)IDTVEC(syscall) + 1) { |
| 776 | /* |
| 777 | * The syscall handler has now saved the |
| 778 | * flags. Stop single stepping it. |
| 779 | */ |
| 780 | frame->tf_eflags &= ~PSL_T; |
| 781 | goto out2; |
| 782 | } |
| 783 | #endif |
| 784 | #if 0 |
| 785 | /* |
| 786 | * Ignore debug register trace traps due to |
| 787 | * accesses in the user's address space, which |
| 788 | * can happen under several conditions such as |
| 789 | * if a user sets a watchpoint on a buffer and |
| 790 | * then passes that buffer to a system call. |
| 791 | * We still want to get TRCTRAPS for addresses |
| 792 | * in kernel space because that is useful when |
| 793 | * debugging the kernel. |
| 794 | */ |
| 795 | if (user_dbreg_trap()) { |
| 796 | /* |
| 797 | * Reset breakpoint bits because the |
| 798 | * processor doesn't |
| 799 | */ |
| 800 | load_dr6(rdr6() & 0xfffffff0); |
| 801 | goto out2; |
| 802 | } |
| 803 | #endif |
| 804 | /* |
| 805 | * Fall through (TRCTRAP kernel mode, kernel address) |
| 806 | */ |
| 807 | case T_BPTFLT: |
| 808 | /* |
| 809 | * If DDB is enabled, let it handle the debugger trap. |
| 810 | * Otherwise, debugger traps "can't happen". |
| 811 | */ |
| 812 | #ifdef DDB |
| 813 | MAKEMPSAFE(have_mplock); |
| 814 | if (kdb_trap (type, 0, frame)) |
| 815 | goto out2; |
| 816 | #endif |
| 817 | break; |
| 818 | case T_DIVIDE: |
| 819 | MAKEMPSAFE(have_mplock); |
| 820 | trap_fatal(frame, FALSE, eva); |
| 821 | goto out2; |
| 822 | case T_NMI: |
| 823 | MAKEMPSAFE(have_mplock); |
| 824 | trap_fatal(frame, FALSE, eva); |
| 825 | goto out2; |
| 826 | case T_SYSCALL80: |
| 827 | /* |
| 828 | * Ignore this trap generated from a spurious SIGTRAP. |
| 829 | * |
| 830 | * single stepping in / syscalls leads to spurious / SIGTRAP |
| 831 | * so ignore |
| 832 | * |
| 833 | * Haiku (c) 2007 Simon 'corecode' Schubert |
| 834 | */ |
| 835 | goto out2; |
| 836 | } |
| 837 | |
| 838 | /* |
| 839 | * Translate fault for emulators (e.g. Linux) |
| 840 | */ |
| 841 | if (*p->p_sysent->sv_transtrap) |
| 842 | i = (*p->p_sysent->sv_transtrap)(i, type); |
| 843 | |
| 844 | MAKEMPSAFE(have_mplock); |
| 845 | trapsignal(lp, i, ucode); |
| 846 | |
| 847 | #ifdef DEBUG |
| 848 | if (type <= MAX_TRAP_MSG) { |
| 849 | uprintf("fatal process exception: %s", |
| 850 | trap_msg[type]); |
| 851 | if ((type == T_PAGEFLT) || (type == T_PROTFLT)) |
| 852 | uprintf(", fault VA = 0x%lx", (u_long)eva); |
| 853 | uprintf("\n"); |
| 854 | } |
| 855 | #endif |
| 856 | |
| 857 | out2: |
| 858 | ; |
| 859 | #ifdef SMP |
| 860 | if (have_mplock) |
| 861 | rel_mplock(); |
| 862 | #endif |
| 863 | #ifdef INVARIANTS |
| 864 | KASSERT(crit_count == td->td_critcount, |
| 865 | ("trap: critical section count mismatch! %d/%d", |
| 866 | crit_count, td->td_pri)); |
| 867 | KASSERT(curstop == td->td_toks_stop, |
| 868 | ("trap: extra tokens held after trap! %zd/%zd", |
| 869 | curstop - &td->td_toks_base, |
| 870 | td->td_toks_stop - &td->td_toks_base)); |
| 871 | #endif |
| 872 | } |
| 873 | |
| 874 | int |
| 875 | trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva) |
| 876 | { |
| 877 | vm_offset_t va; |
| 878 | struct vmspace *vm = NULL; |
| 879 | vm_map_t map = 0; |
| 880 | int rv = 0; |
| 881 | int fault_flags; |
| 882 | vm_prot_t ftype; |
| 883 | thread_t td = curthread; |
| 884 | struct lwp *lp = td->td_lwp; |
| 885 | |
| 886 | va = trunc_page(eva); |
| 887 | if (usermode == FALSE) { |
| 888 | /* |
| 889 | * This is a fault on kernel virtual memory. |
| 890 | */ |
| 891 | map = &kernel_map; |
| 892 | } else { |
| 893 | /* |
| 894 | * This is a fault on non-kernel virtual memory. |
| 895 | * vm is initialized above to NULL. If curproc is NULL |
| 896 | * or curproc->p_vmspace is NULL the fault is fatal. |
| 897 | */ |
| 898 | if (lp != NULL) |
| 899 | vm = lp->lwp_vmspace; |
| 900 | |
| 901 | if (vm == NULL) |
| 902 | goto nogo; |
| 903 | |
| 904 | map = &vm->vm_map; |
| 905 | } |
| 906 | |
| 907 | if (frame->tf_xflags & PGEX_W) |
| 908 | ftype = VM_PROT_READ | VM_PROT_WRITE; |
| 909 | else |
| 910 | ftype = VM_PROT_READ; |
| 911 | |
| 912 | if (map != &kernel_map) { |
| 913 | /* |
| 914 | * Keep swapout from messing with us during this |
| 915 | * critical time. |
| 916 | */ |
| 917 | PHOLD(lp->lwp_proc); |
| 918 | |
| 919 | /* |
| 920 | * Issue fault |
| 921 | */ |
| 922 | fault_flags = 0; |
| 923 | if (usermode) |
| 924 | fault_flags |= VM_FAULT_BURST; |
| 925 | if (ftype & VM_PROT_WRITE) |
| 926 | fault_flags |= VM_FAULT_DIRTY; |
| 927 | else |
| 928 | fault_flags |= VM_FAULT_NORMAL; |
| 929 | rv = vm_fault(map, va, ftype, fault_flags); |
| 930 | |
| 931 | PRELE(lp->lwp_proc); |
| 932 | } else { |
| 933 | /* |
| 934 | * Don't have to worry about process locking or stacks in the kernel. |
| 935 | */ |
| 936 | rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); |
| 937 | } |
| 938 | |
| 939 | if (rv == KERN_SUCCESS) |
| 940 | return (0); |
| 941 | nogo: |
| 942 | if (!usermode) { |
| 943 | if (td->td_gd->gd_intr_nesting_level == 0 && |
| 944 | td->td_pcb->pcb_onfault) { |
| 945 | frame->tf_eip = (register_t)td->td_pcb->pcb_onfault; |
| 946 | return (0); |
| 947 | } |
| 948 | trap_fatal(frame, usermode, eva); |
| 949 | return (-1); |
| 950 | } |
| 951 | return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); |
| 952 | } |
| 953 | |
| 954 | static void |
| 955 | trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva) |
| 956 | { |
| 957 | int code, type, ss, esp; |
| 958 | |
| 959 | code = frame->tf_xflags; |
| 960 | type = frame->tf_trapno; |
| 961 | |
| 962 | if (type <= MAX_TRAP_MSG) { |
| 963 | kprintf("\n\nFatal trap %d: %s while in %s mode\n", |
| 964 | type, trap_msg[type], |
| 965 | (usermode ? "user" : "kernel")); |
| 966 | } |
| 967 | #ifdef SMP |
| 968 | /* two separate prints in case of a trap on an unmapped page */ |
| 969 | kprintf("cpuid = %d\n", mycpu->gd_cpuid); |
| 970 | #endif |
| 971 | if (type == T_PAGEFLT) { |
| 972 | kprintf("fault virtual address = %p\n", (void *)eva); |
| 973 | kprintf("fault code = %s %s, %s\n", |
| 974 | usermode ? "user" : "supervisor", |
| 975 | code & PGEX_W ? "write" : "read", |
| 976 | code & PGEX_P ? "protection violation" : "page not present"); |
| 977 | } |
| 978 | kprintf("instruction pointer = 0x%x:0x%x\n", |
| 979 | frame->tf_cs & 0xffff, frame->tf_eip); |
| 980 | if (usermode) { |
| 981 | ss = frame->tf_ss & 0xffff; |
| 982 | esp = frame->tf_esp; |
| 983 | } else { |
| 984 | ss = GSEL(GDATA_SEL, SEL_KPL); |
| 985 | esp = (int)&frame->tf_esp; |
| 986 | } |
| 987 | kprintf("stack pointer = 0x%x:0x%x\n", ss, esp); |
| 988 | kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp); |
| 989 | kprintf("processor eflags = "); |
| 990 | if (frame->tf_eflags & PSL_T) |
| 991 | kprintf("trace trap, "); |
| 992 | if (frame->tf_eflags & PSL_I) |
| 993 | kprintf("interrupt enabled, "); |
| 994 | if (frame->tf_eflags & PSL_NT) |
| 995 | kprintf("nested task, "); |
| 996 | if (frame->tf_eflags & PSL_RF) |
| 997 | kprintf("resume, "); |
| 998 | #if 0 |
| 999 | if (frame->tf_eflags & PSL_VM) |
| 1000 | kprintf("vm86, "); |
| 1001 | #endif |
| 1002 | kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12); |
| 1003 | kprintf("current process = "); |
| 1004 | if (curproc) { |
| 1005 | kprintf("%lu (%s)\n", |
| 1006 | (u_long)curproc->p_pid, curproc->p_comm ? |
| 1007 | curproc->p_comm : ""); |
| 1008 | } else { |
| 1009 | kprintf("Idle\n"); |
| 1010 | } |
| 1011 | kprintf("current thread = pri %d ", curthread->td_pri); |
| 1012 | if (curthread->td_critcount) |
| 1013 | kprintf("(CRIT)"); |
| 1014 | kprintf("\n"); |
| 1015 | #ifdef SMP |
| 1016 | /** |
| 1017 | * XXX FIXME: |
| 1018 | * we probably SHOULD have stopped the other CPUs before now! |
| 1019 | * another CPU COULD have been touching cpl at this moment... |
| 1020 | */ |
| 1021 | kprintf(" <- SMP: XXX"); |
| 1022 | #endif |
| 1023 | kprintf("\n"); |
| 1024 | |
| 1025 | #ifdef KDB |
| 1026 | if (kdb_trap(&psl)) |
| 1027 | return; |
| 1028 | #endif |
| 1029 | #ifdef DDB |
| 1030 | if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame)) |
| 1031 | return; |
| 1032 | #endif |
| 1033 | kprintf("trap number = %d\n", type); |
| 1034 | if (type <= MAX_TRAP_MSG) |
| 1035 | panic("%s", trap_msg[type]); |
| 1036 | else |
| 1037 | panic("unknown/reserved trap"); |
| 1038 | } |
| 1039 | |
| 1040 | /* |
| 1041 | * Double fault handler. Called when a fault occurs while writing |
| 1042 | * a frame for a trap/exception onto the stack. This usually occurs |
| 1043 | * when the stack overflows (such is the case with infinite recursion, |
| 1044 | * for example). |
| 1045 | * |
| 1046 | * XXX Note that the current PTD gets replaced by IdlePTD when the |
| 1047 | * task switch occurs. This means that the stack that was active at |
| 1048 | * the time of the double fault is not available at <kstack> unless |
| 1049 | * the machine was idle when the double fault occurred. The downside |
| 1050 | * of this is that "trace <ebp>" in ddb won't work. |
| 1051 | */ |
| 1052 | void |
| 1053 | dblfault_handler(void) |
| 1054 | { |
| 1055 | struct mdglobaldata *gd = mdcpu; |
| 1056 | |
| 1057 | kprintf("\nFatal double fault:\n"); |
| 1058 | kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip); |
| 1059 | kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp); |
| 1060 | kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp); |
| 1061 | #ifdef SMP |
| 1062 | /* two separate prints in case of a trap on an unmapped page */ |
| 1063 | kprintf("cpuid = %d\n", mycpu->gd_cpuid); |
| 1064 | #endif |
| 1065 | panic("double fault"); |
| 1066 | } |
| 1067 | |
| 1068 | /* |
| 1069 | * syscall2 - MP aware system call request C handler |
| 1070 | * |
| 1071 | * A system call is essentially treated as a trap except that the |
| 1072 | * MP lock is not held on entry or return. We are responsible for |
| 1073 | * obtaining the MP lock if necessary and for handling ASTs |
| 1074 | * (e.g. a task switch) prior to return. |
| 1075 | * |
| 1076 | * MPSAFE |
| 1077 | */ |
| 1078 | void |
| 1079 | syscall2(struct trapframe *frame) |
| 1080 | { |
| 1081 | struct thread *td = curthread; |
| 1082 | struct proc *p = td->td_proc; |
| 1083 | struct lwp *lp = td->td_lwp; |
| 1084 | caddr_t params; |
| 1085 | struct sysent *callp; |
| 1086 | register_t orig_tf_eflags; |
| 1087 | int sticks; |
| 1088 | int error; |
| 1089 | int narg; |
| 1090 | #ifdef INVARIANTS |
| 1091 | int crit_count = td->td_critcount; |
| 1092 | #endif |
| 1093 | #ifdef SMP |
| 1094 | int have_mplock = 0; |
| 1095 | #endif |
| 1096 | u_int code; |
| 1097 | union sysunion args; |
| 1098 | |
| 1099 | KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid, |
| 1100 | frame->tf_eax); |
| 1101 | |
| 1102 | userenter(td, p); /* lazy raise our priority */ |
| 1103 | |
| 1104 | /* |
| 1105 | * Misc |
| 1106 | */ |
| 1107 | sticks = (int)td->td_sticks; |
| 1108 | orig_tf_eflags = frame->tf_eflags; |
| 1109 | |
| 1110 | /* |
| 1111 | * Virtual kernel intercept - if a VM context managed by a virtual |
| 1112 | * kernel issues a system call the virtual kernel handles it, not us. |
| 1113 | * Restore the virtual kernel context and return from its system |
| 1114 | * call. The current frame is copied out to the virtual kernel. |
| 1115 | */ |
| 1116 | if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { |
| 1117 | vkernel_trap(lp, frame); |
| 1118 | error = EJUSTRETURN; |
| 1119 | goto out; |
| 1120 | } |
| 1121 | |
| 1122 | /* |
| 1123 | * Get the system call parameters and account for time |
| 1124 | */ |
| 1125 | lp->lwp_md.md_regs = frame; |
| 1126 | params = (caddr_t)frame->tf_esp + sizeof(int); |
| 1127 | code = frame->tf_eax; |
| 1128 | |
| 1129 | if (p->p_sysent->sv_prepsyscall) { |
| 1130 | (*p->p_sysent->sv_prepsyscall)( |
| 1131 | frame, (int *)(&args.nosys.sysmsg + 1), |
| 1132 | &code, ¶ms); |
| 1133 | } else { |
| 1134 | /* |
| 1135 | * Need to check if this is a 32 bit or 64 bit syscall. |
| 1136 | * fuword is MP aware. |
| 1137 | */ |
| 1138 | if (code == SYS_syscall) { |
| 1139 | /* |
| 1140 | * Code is first argument, followed by actual args. |
| 1141 | */ |
| 1142 | code = fuword(params); |
| 1143 | params += sizeof(int); |
| 1144 | } else if (code == SYS___syscall) { |
| 1145 | /* |
| 1146 | * Like syscall, but code is a quad, so as to maintain |
| 1147 | * quad alignment for the rest of the arguments. |
| 1148 | */ |
| 1149 | code = fuword(params); |
| 1150 | params += sizeof(quad_t); |
| 1151 | } |
| 1152 | } |
| 1153 | |
| 1154 | code &= p->p_sysent->sv_mask; |
| 1155 | if (code >= p->p_sysent->sv_size) |
| 1156 | callp = &p->p_sysent->sv_table[0]; |
| 1157 | else |
| 1158 | callp = &p->p_sysent->sv_table[code]; |
| 1159 | |
| 1160 | narg = callp->sy_narg & SYF_ARGMASK; |
| 1161 | |
| 1162 | /* |
| 1163 | * copyin is MP aware, but the tracing code is not |
| 1164 | */ |
| 1165 | if (narg && params) { |
| 1166 | error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1), |
| 1167 | narg * sizeof(register_t)); |
| 1168 | if (error) { |
| 1169 | #ifdef KTRACE |
| 1170 | if (KTRPOINT(td, KTR_SYSCALL)) { |
| 1171 | MAKEMPSAFE(have_mplock); |
| 1172 | |
| 1173 | ktrsyscall(lp, code, narg, |
| 1174 | (void *)(&args.nosys.sysmsg + 1)); |
| 1175 | } |
| 1176 | #endif |
| 1177 | goto bad; |
| 1178 | } |
| 1179 | } |
| 1180 | |
| 1181 | #ifdef KTRACE |
| 1182 | if (KTRPOINT(td, KTR_SYSCALL)) { |
| 1183 | MAKEMPSAFE(have_mplock); |
| 1184 | ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1)); |
| 1185 | } |
| 1186 | #endif |
| 1187 | |
| 1188 | /* |
| 1189 | * For traditional syscall code edx is left untouched when 32 bit |
| 1190 | * results are returned. Since edx is loaded from fds[1] when the |
| 1191 | * system call returns we pre-set it here. |
| 1192 | */ |
| 1193 | args.sysmsg_fds[0] = 0; |
| 1194 | args.sysmsg_fds[1] = frame->tf_edx; |
| 1195 | |
| 1196 | /* |
| 1197 | * The syscall might manipulate the trap frame. If it does it |
| 1198 | * will probably return EJUSTRETURN. |
| 1199 | */ |
| 1200 | args.sysmsg_frame = frame; |
| 1201 | |
| 1202 | STOPEVENT(p, S_SCE, narg); /* MP aware */ |
| 1203 | |
| 1204 | /* |
| 1205 | * NOTE: All system calls run MPSAFE now. The system call itself |
| 1206 | * is responsible for getting the MP lock. |
| 1207 | */ |
| 1208 | error = (*callp->sy_call)(&args); |
| 1209 | |
| 1210 | #if 0 |
| 1211 | kprintf("system call %d returned %d\n", code, error); |
| 1212 | #endif |
| 1213 | |
| 1214 | out: |
| 1215 | /* |
| 1216 | * MP SAFE (we may or may not have the MP lock at this point) |
| 1217 | */ |
| 1218 | switch (error) { |
| 1219 | case 0: |
| 1220 | /* |
| 1221 | * Reinitialize proc pointer `p' as it may be different |
| 1222 | * if this is a child returning from fork syscall. |
| 1223 | */ |
| 1224 | p = curproc; |
| 1225 | lp = curthread->td_lwp; |
| 1226 | frame->tf_eax = args.sysmsg_fds[0]; |
| 1227 | frame->tf_edx = args.sysmsg_fds[1]; |
| 1228 | frame->tf_eflags &= ~PSL_C; |
| 1229 | break; |
| 1230 | case ERESTART: |
| 1231 | /* |
| 1232 | * Reconstruct pc, assuming lcall $X,y is 7 bytes, |
| 1233 | * int 0x80 is 2 bytes. We saved this in tf_err. |
| 1234 | */ |
| 1235 | frame->tf_eip -= frame->tf_err; |
| 1236 | break; |
| 1237 | case EJUSTRETURN: |
| 1238 | break; |
| 1239 | case EASYNC: |
| 1240 | panic("Unexpected EASYNC return value (for now)"); |
| 1241 | default: |
| 1242 | bad: |
| 1243 | if (p->p_sysent->sv_errsize) { |
| 1244 | if (error >= p->p_sysent->sv_errsize) |
| 1245 | error = -1; /* XXX */ |
| 1246 | else |
| 1247 | error = p->p_sysent->sv_errtbl[error]; |
| 1248 | } |
| 1249 | frame->tf_eax = error; |
| 1250 | frame->tf_eflags |= PSL_C; |
| 1251 | break; |
| 1252 | } |
| 1253 | |
| 1254 | /* |
| 1255 | * Traced syscall. trapsignal() is not MP aware. |
| 1256 | */ |
| 1257 | if ((orig_tf_eflags & PSL_T) /*&& !(orig_tf_eflags & PSL_VM)*/) { |
| 1258 | MAKEMPSAFE(have_mplock); |
| 1259 | frame->tf_eflags &= ~PSL_T; |
| 1260 | trapsignal(lp, SIGTRAP, TRAP_TRACE); |
| 1261 | } |
| 1262 | |
| 1263 | /* |
| 1264 | * Handle reschedule and other end-of-syscall issues |
| 1265 | */ |
| 1266 | userret(lp, frame, sticks); |
| 1267 | |
| 1268 | #ifdef KTRACE |
| 1269 | if (KTRPOINT(td, KTR_SYSRET)) { |
| 1270 | MAKEMPSAFE(have_mplock); |
| 1271 | ktrsysret(lp, code, error, args.sysmsg_result); |
| 1272 | } |
| 1273 | #endif |
| 1274 | |
| 1275 | /* |
| 1276 | * This works because errno is findable through the |
| 1277 | * register set. If we ever support an emulation where this |
| 1278 | * is not the case, this code will need to be revisited. |
| 1279 | */ |
| 1280 | STOPEVENT(p, S_SCX, code); |
| 1281 | |
| 1282 | userexit(lp); |
| 1283 | #ifdef SMP |
| 1284 | /* |
| 1285 | * Release the MP lock if we had to get it |
| 1286 | */ |
| 1287 | if (have_mplock) |
| 1288 | rel_mplock(); |
| 1289 | #endif |
| 1290 | KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error); |
| 1291 | #ifdef INVARIANTS |
| 1292 | KASSERT(crit_count == td->td_critcount, |
| 1293 | ("syscall: critical section count mismatch! %d/%d", |
| 1294 | crit_count, td->td_pri)); |
| 1295 | KASSERT(&td->td_toks_base == td->td_toks_stop, |
| 1296 | ("syscall: extra tokens held after trap! %zd", |
| 1297 | td->td_toks_stop - &td->td_toks_base)); |
| 1298 | #endif |
| 1299 | } |
| 1300 | |
| 1301 | /* |
| 1302 | * NOTE: mplock not held at any point |
| 1303 | */ |
| 1304 | void |
| 1305 | fork_return(struct lwp *lp, struct trapframe *frame) |
| 1306 | { |
| 1307 | frame->tf_eax = 0; /* Child returns zero */ |
| 1308 | frame->tf_eflags &= ~PSL_C; /* success */ |
| 1309 | frame->tf_edx = 1; |
| 1310 | |
| 1311 | generic_lwp_return(lp, frame); |
| 1312 | KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid); |
| 1313 | } |
| 1314 | |
| 1315 | /* |
| 1316 | * Simplified back end of syscall(), used when returning from fork() |
| 1317 | * directly into user mode. |
| 1318 | * |
| 1319 | * This code will return back into the fork trampoline code which then |
| 1320 | * runs doreti. |
| 1321 | * |
| 1322 | * NOTE: The mplock is not held at any point. |
| 1323 | */ |
| 1324 | void |
| 1325 | generic_lwp_return(struct lwp *lp, struct trapframe *frame) |
| 1326 | { |
| 1327 | struct proc *p = lp->lwp_proc; |
| 1328 | |
| 1329 | /* |
| 1330 | * Newly forked processes are given a kernel priority. We have to |
| 1331 | * adjust the priority to a normal user priority and fake entry |
| 1332 | * into the kernel (call userenter()) to install a passive release |
| 1333 | * function just in case userret() decides to stop the process. This |
| 1334 | * can occur when ^Z races a fork. If we do not install the passive |
| 1335 | * release function the current process designation will not be |
| 1336 | * released when the thread goes to sleep. |
| 1337 | */ |
| 1338 | lwkt_setpri_self(TDPRI_USER_NORM); |
| 1339 | userenter(lp->lwp_thread, p); |
| 1340 | userret(lp, frame, 0); |
| 1341 | #ifdef KTRACE |
| 1342 | if (KTRPOINT(lp->lwp_thread, KTR_SYSRET)) |
| 1343 | ktrsysret(lp, SYS_fork, 0, 0); |
| 1344 | #endif |
| 1345 | lp->lwp_flags |= LWP_PASSIVE_ACQ; |
| 1346 | userexit(lp); |
| 1347 | lp->lwp_flags &= ~LWP_PASSIVE_ACQ; |
| 1348 | } |
| 1349 | |
| 1350 | /* |
| 1351 | * doreti has turned into this. The frame is directly on the stack. We |
| 1352 | * pull everything else we need (fpu and tls context) from the current |
| 1353 | * thread. |
| 1354 | * |
| 1355 | * Note on fpu interactions: In a virtual kernel, the fpu context for |
| 1356 | * an emulated user mode process is not shared with the virtual kernel's |
| 1357 | * fpu context, so we only have to 'stack' fpu contexts within the virtual |
| 1358 | * kernel itself, and not even then since the signal() contexts that we care |
| 1359 | * about save and restore the FPU state (I think anyhow). |
| 1360 | * |
| 1361 | * vmspace_ctl() returns an error only if it had problems instaling the |
| 1362 | * context we supplied or problems copying data to/from our VM space. |
| 1363 | */ |
| 1364 | void |
| 1365 | go_user(struct intrframe *frame) |
| 1366 | { |
| 1367 | struct trapframe *tf = (void *)&frame->if_gs; |
| 1368 | int r; |
| 1369 | |
| 1370 | /* |
| 1371 | * Interrupts may be disabled on entry, make sure all signals |
| 1372 | * can be received before beginning our loop. |
| 1373 | */ |
| 1374 | sigsetmask(0); |
| 1375 | |
| 1376 | /* |
| 1377 | * Switch to the current simulated user process, then call |
| 1378 | * user_trap() when we break out of it (usually due to a signal). |
| 1379 | */ |
| 1380 | for (;;) { |
| 1381 | /* |
| 1382 | * Tell the real kernel whether it is ok to use the FP |
| 1383 | * unit or not. |
| 1384 | * |
| 1385 | * The critical section is required to prevent an interrupt |
| 1386 | * from causing a preemptive task switch and changing |
| 1387 | * the FP state. |
| 1388 | */ |
| 1389 | crit_enter(); |
| 1390 | if (mdcpu->gd_npxthread == curthread) { |
| 1391 | tf->tf_xflags &= ~PGEX_FPFAULT; |
| 1392 | } else { |
| 1393 | tf->tf_xflags |= PGEX_FPFAULT; |
| 1394 | } |
| 1395 | |
| 1396 | /* |
| 1397 | * Run emulated user process context. This call interlocks |
| 1398 | * with new mailbox signals. |
| 1399 | * |
| 1400 | * Set PGEX_U unconditionally, indicating a user frame (the |
| 1401 | * bit is normally set only by T_PAGEFLT). |
| 1402 | */ |
| 1403 | r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN, |
| 1404 | tf, &curthread->td_savevext); |
| 1405 | crit_exit(); |
| 1406 | frame->if_xflags |= PGEX_U; |
| 1407 | #if 0 |
| 1408 | kprintf("GO USER %d trap %d EVA %08x EIP %08x ESP %08x XFLAGS %02x/%02x\n", |
| 1409 | r, tf->tf_trapno, tf->tf_err, tf->tf_eip, tf->tf_esp, |
| 1410 | tf->tf_xflags, frame->if_xflags); |
| 1411 | #endif |
| 1412 | if (r < 0) { |
| 1413 | if (errno != EINTR) |
| 1414 | panic("vmspace_ctl failed error %d", errno); |
| 1415 | } else { |
| 1416 | if (tf->tf_trapno) { |
| 1417 | user_trap(tf); |
| 1418 | } |
| 1419 | } |
| 1420 | if (mycpu->gd_reqflags & RQF_AST_MASK) { |
| 1421 | tf->tf_trapno = T_ASTFLT; |
| 1422 | user_trap(tf); |
| 1423 | } |
| 1424 | tf->tf_trapno = 0; |
| 1425 | } |
| 1426 | } |
| 1427 | |
| 1428 | /* |
| 1429 | * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA |
| 1430 | * fault (which is then passed back to the virtual kernel) if an attempt is |
| 1431 | * made to use the FP unit. |
| 1432 | * |
| 1433 | * XXX this is a fairly big hack. |
| 1434 | */ |
| 1435 | void |
| 1436 | set_vkernel_fp(struct trapframe *frame) |
| 1437 | { |
| 1438 | struct thread *td = curthread; |
| 1439 | |
| 1440 | if (frame->tf_xflags & PGEX_FPFAULT) { |
| 1441 | td->td_pcb->pcb_flags |= FP_VIRTFP; |
| 1442 | if (mdcpu->gd_npxthread == td) |
| 1443 | npxexit(); |
| 1444 | } else { |
| 1445 | td->td_pcb->pcb_flags &= ~FP_VIRTFP; |
| 1446 | } |
| 1447 | } |
| 1448 | |
| 1449 | /* |
| 1450 | * Called from vkernel_trap() to fixup the vkernel's syscall |
| 1451 | * frame for vmspace_ctl() return. |
| 1452 | */ |
| 1453 | void |
| 1454 | cpu_vkernel_trap(struct trapframe *frame, int error) |
| 1455 | { |
| 1456 | frame->tf_eax = error; |
| 1457 | if (error) |
| 1458 | frame->tf_eflags |= PSL_C; |
| 1459 | else |
| 1460 | frame->tf_eflags &= ~PSL_C; |
| 1461 | } |