1 /*- 2 * Copyright (c) 1990, 1993 3 * The Regents of the University of California. All rights reserved. 4 * Copyright (C) 1994, David Greenman 5 * Copyright (c) 2008 The DragonFly Project. 6 * Copyright (c) 2008 Jordan Gordeev. 7 * 8 * This code is derived from software contributed to Berkeley by 9 * the University of Utah, and William Jolitz. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. All advertising materials mentioning features or use of this software 20 * must display the following acknowledgement: 21 * This product includes software developed by the University of 22 * California, Berkeley and its contributors. 23 * 4. Neither the name of the University nor the names of its contributors 24 * may be used to endorse or promote products derived from this software 25 * without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 37 * SUCH DAMAGE. 38 * 39 * from: @(#)trap.c 7.4 (Berkeley) 5/13/91 40 * $FreeBSD: src/sys/i386/i386/trap.c,v 1.147.2.11 2003/02/27 19:09:59 luoqi Exp $ 41 */ 42 43 /* 44 * x86_64 Trap and System call handling 45 */ 46 47 #include "use_isa.h" 48 49 #include "opt_ddb.h" 50 #include "opt_ktrace.h" 51 52 #include <machine/frame.h> 53 #include <sys/param.h> 54 #include <sys/systm.h> 55 #include <sys/kernel.h> 56 #include <sys/kerneldump.h> 57 #include <sys/proc.h> 58 #include <sys/pioctl.h> 59 #include <sys/types.h> 60 #include <sys/signal2.h> 61 #include <sys/syscall.h> 62 #include <sys/sysctl.h> 63 #include <sys/sysent.h> 64 #ifdef KTRACE 65 #include <sys/ktrace.h> 66 #endif 67 #include <sys/ktr.h> 68 #include <sys/sysmsg.h> 69 #include <sys/sysproto.h> 70 #include <sys/sysunion.h> 71 72 #include <vm/pmap.h> 73 #include <vm/vm.h> 74 #include <vm/vm_extern.h> 75 #include <vm/vm_kern.h> 76 #include <vm/vm_param.h> 77 #include <machine/cpu.h> 78 #include <machine/pcb.h> 79 #include <machine/smp.h> 80 #include <machine/thread.h> 81 #include <machine/clock.h> 82 #include <machine/vmparam.h> 83 #include <machine/md_var.h> 84 #include <machine_base/isa/isa_intr.h> 85 #include <machine_base/apic/lapic.h> 86 87 #include <ddb/ddb.h> 88 89 #include <sys/thread2.h> 90 #include <sys/spinlock2.h> 91 92 extern void trap(struct trapframe *frame); 93 94 static int trap_pfault(struct trapframe *, int); 95 static void trap_fatal(struct trapframe *, vm_offset_t); 96 void dblfault_handler(struct trapframe *frame); 97 98 #define MAX_TRAP_MSG 30 99 static char *trap_msg[] = { 100 "", /* 0 unused */ 101 "privileged instruction fault", /* 1 T_PRIVINFLT */ 102 "", /* 2 unused */ 103 "breakpoint instruction fault", /* 3 T_BPTFLT */ 104 "", /* 4 unused */ 105 "", /* 5 unused */ 106 "arithmetic trap", /* 6 T_ARITHTRAP */ 107 "system forced exception", /* 7 T_ASTFLT */ 108 "", /* 8 unused */ 109 "general protection fault", /* 9 T_PROTFLT */ 110 "trace trap", /* 10 T_TRCTRAP */ 111 "", /* 11 unused */ 112 "page fault", /* 12 T_PAGEFLT */ 113 "", /* 13 unused */ 114 "alignment fault", /* 14 T_ALIGNFLT */ 115 "", /* 15 unused */ 116 "", /* 16 unused */ 117 "", /* 17 unused */ 118 "integer divide fault", /* 18 T_DIVIDE */ 119 "non-maskable interrupt trap", /* 19 T_NMI */ 120 "overflow trap", /* 20 T_OFLOW */ 121 "FPU bounds check fault", /* 21 T_BOUND */ 122 "FPU device not available", /* 22 T_DNA */ 123 "double fault", /* 23 T_DOUBLEFLT */ 124 "FPU operand fetch fault", /* 24 T_FPOPFLT */ 125 "invalid TSS fault", /* 25 T_TSSFLT */ 126 "segment not present fault", /* 26 T_SEGNPFLT */ 127 "stack fault", /* 27 T_STKFLT */ 128 "machine check trap", /* 28 T_MCHK */ 129 "SIMD floating-point exception", /* 29 T_XMMFLT */ 130 "reserved (unknown) fault", /* 30 T_RESERVED */ 131 }; 132 133 #ifdef DDB 134 static int ddb_on_nmi = 1; 135 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW, 136 &ddb_on_nmi, 0, "Go to DDB on NMI"); 137 static int ddb_on_seg_fault = 0; 138 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_seg_fault, CTLFLAG_RW, 139 &ddb_on_seg_fault, 0, "Go to DDB on user seg-fault"); 140 static int freeze_on_seg_fault = 0; 141 SYSCTL_INT(_machdep, OID_AUTO, freeze_on_seg_fault, CTLFLAG_RW, 142 &freeze_on_seg_fault, 0, "Go to DDB on user seg-fault"); 143 #endif 144 static int panic_on_nmi = 1; 145 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW, 146 &panic_on_nmi, 0, "Panic on NMI"); 147 static int fast_release; 148 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW, 149 &fast_release, 0, "Passive Release was optimal"); 150 static int slow_release; 151 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW, 152 &slow_release, 0, "Passive Release was nonoptimal"); 153 154 /* 155 * System call debugging records the worst-case system call 156 * overhead (inclusive of blocking), but may be inaccurate. 157 */ 158 /*#define SYSCALL_DEBUG*/ 159 #ifdef SYSCALL_DEBUG 160 uint64_t SysCallsWorstCase[SYS_MAXSYSCALL]; 161 #endif 162 163 /* 164 * Passively intercepts the thread switch function to increase 165 * the thread priority from a user priority to a kernel priority, reducing 166 * syscall and trap overhead for the case where no switch occurs. 167 * 168 * Synchronizes td_ucred with p_ucred. This is used by system calls, 169 * signal handling, faults, AST traps, and anything else that enters the 170 * kernel from userland and provides the kernel with a stable read-only 171 * copy of the process ucred. 172 * 173 * To avoid races with another thread updating p_ucred we obtain p_spin. 174 * The other thread doing the update will obtain both p_token and p_spin. 175 * In the case where the cached cred pointer matches, we will already have 176 * the ref and we don't have to do one blessed thing. 177 */ 178 static __inline void 179 userenter(struct thread *curtd, struct proc *curp) 180 { 181 struct ucred *ocred; 182 struct ucred *ncred; 183 184 curtd->td_release = lwkt_passive_release; 185 186 if (curtd->td_ucred != curp->p_ucred) { 187 spin_lock(&curp->p_spin); 188 ncred = crhold(curp->p_ucred); 189 spin_unlock(&curp->p_spin); 190 ocred = curtd->td_ucred; 191 curtd->td_ucred = ncred; 192 if (ocred) 193 crfree(ocred); 194 } 195 196 #ifdef DDB 197 /* 198 * Debugging, remove top two user stack pages to catch kernel faults 199 */ 200 if (freeze_on_seg_fault > 1 && curtd->td_lwp) { 201 pmap_remove(vmspace_pmap(curtd->td_lwp->lwp_vmspace), 202 0x00007FFFFFFFD000LU, 203 0x0000800000000000LU); 204 } 205 #endif 206 } 207 208 /* 209 * Handle signals, upcalls, profiling, and other AST's and/or tasks that 210 * must be completed before we can return to or try to return to userland. 211 * 212 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64 213 * arithmatic on the delta calculation so the absolute tick values are 214 * truncated to an integer. 215 */ 216 static void 217 userret(struct lwp *lp, struct trapframe *frame, int sticks) 218 { 219 struct proc *p = lp->lwp_proc; 220 int sig; 221 int ptok; 222 223 /* 224 * Charge system time if profiling. Note: times are in microseconds. 225 * This may do a copyout and block, so do it first even though it 226 * means some system time will be charged as user time. 227 */ 228 if (p->p_flags & P_PROFIL) { 229 addupc_task(p, frame->tf_rip, 230 (u_int)((int)lp->lwp_thread->td_sticks - sticks)); 231 } 232 233 recheck: 234 /* 235 * Specific on-return-to-usermode checks (LWP_MP_WEXIT, 236 * LWP_MP_VNLRU, etc). 237 */ 238 if (lp->lwp_mpflags & LWP_MP_URETMASK) 239 lwpuserret(lp); 240 241 /* 242 * Block here if we are in a stopped state. 243 */ 244 if (STOPLWP(p, lp)) { 245 lwkt_gettoken(&p->p_token); 246 tstop(); 247 lwkt_reltoken(&p->p_token); 248 goto recheck; 249 } 250 while (dump_stop_usertds) { 251 tsleep(&dump_stop_usertds, 0, "dumpstp", 0); 252 } 253 254 /* 255 * Post any pending upcalls. If running a virtual kernel be sure 256 * to restore the virtual kernel's vmspace before posting the upcall. 257 */ 258 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF)) { 259 lwkt_gettoken(&p->p_token); 260 if (p->p_flags & P_SIGVTALRM) { 261 p->p_flags &= ~P_SIGVTALRM; 262 ksignal(p, SIGVTALRM); 263 } 264 if (p->p_flags & P_SIGPROF) { 265 p->p_flags &= ~P_SIGPROF; 266 ksignal(p, SIGPROF); 267 } 268 lwkt_reltoken(&p->p_token); 269 goto recheck; 270 } 271 272 /* 273 * Post any pending signals. If running a virtual kernel be sure 274 * to restore the virtual kernel's vmspace before posting the signal. 275 * 276 * WARNING! postsig() can exit and not return. 277 */ 278 if ((sig = CURSIG_LCK_TRACE(lp, &ptok)) != 0) { 279 postsig(sig, ptok); 280 goto recheck; 281 } 282 283 /* 284 * block here if we are swapped out, but still process signals 285 * (such as SIGKILL). proc0 (the swapin scheduler) is already 286 * aware of our situation, we do not have to wake it up. 287 */ 288 if (p->p_flags & P_SWAPPEDOUT) { 289 lwkt_gettoken(&p->p_token); 290 p->p_flags |= P_SWAPWAIT; 291 swapin_request(); 292 if (p->p_flags & P_SWAPWAIT) 293 tsleep(p, PCATCH, "SWOUT", 0); 294 p->p_flags &= ~P_SWAPWAIT; 295 lwkt_reltoken(&p->p_token); 296 goto recheck; 297 } 298 299 /* 300 * In a multi-threaded program it is possible for a thread to change 301 * signal state during a system call which temporarily changes the 302 * signal mask. In this case postsig() might not be run and we 303 * have to restore the mask ourselves. 304 */ 305 if (lp->lwp_flags & LWP_OLDMASK) { 306 lp->lwp_flags &= ~LWP_OLDMASK; 307 lp->lwp_sigmask = lp->lwp_oldsigmask; 308 goto recheck; 309 } 310 } 311 312 /* 313 * Cleanup from userenter and any passive release that might have occured. 314 * We must reclaim the current-process designation before we can return 315 * to usermode. We also handle both LWKT and USER reschedule requests. 316 */ 317 static __inline void 318 userexit(struct lwp *lp) 319 { 320 struct thread *td = lp->lwp_thread; 321 /* globaldata_t gd = td->td_gd; */ 322 323 /* 324 * Handle stop requests at kernel priority. Any requests queued 325 * after this loop will generate another AST. 326 */ 327 while (STOPLWP(lp->lwp_proc, lp)) { 328 lwkt_gettoken(&lp->lwp_proc->p_token); 329 tstop(); 330 lwkt_reltoken(&lp->lwp_proc->p_token); 331 } 332 333 /* 334 * Reduce our priority in preparation for a return to userland. If 335 * our passive release function was still in place, our priority was 336 * never raised and does not need to be reduced. 337 */ 338 lwkt_passive_recover(td); 339 340 /* WARNING: we may have migrated cpu's */ 341 /* gd = td->td_gd; */ 342 343 /* 344 * Become the current user scheduled process if we aren't already, 345 * and deal with reschedule requests and other factors. 346 */ 347 lp->lwp_proc->p_usched->acquire_curproc(lp); 348 } 349 350 #if !defined(KTR_KERNENTRY) 351 #define KTR_KERNENTRY KTR_ALL 352 #endif 353 KTR_INFO_MASTER(kernentry); 354 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, 355 "TRAP(pid %d, tid %d, trapno %ld, eva %lu)", 356 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva); 357 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %d, tid %d)", 358 pid_t pid, lwpid_t tid); 359 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %d, tid %d, nr %ld)", 360 pid_t pid, lwpid_t tid, register_t trapno); 361 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %d, tid %d, err %d)", 362 pid_t pid, lwpid_t tid, int err); 363 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %d, tid %d)", 364 pid_t pid, lwpid_t tid); 365 366 /* 367 * Exception, fault, and trap interface to the kernel. 368 * This common code is called from assembly language IDT gate entry 369 * routines that prepare a suitable stack frame, and restore this 370 * frame after the exception has been processed. 371 * 372 * This function is also called from doreti in an interlock to handle ASTs. 373 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap 374 * 375 * NOTE! We have to retrieve the fault address prior to potentially 376 * blocking, including blocking on any token. 377 * 378 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing 379 * if an attempt is made to switch from a fast interrupt or IPI. 380 */ 381 void 382 trap(struct trapframe *frame) 383 { 384 struct globaldata *gd = mycpu; 385 struct thread *td = gd->gd_curthread; 386 struct lwp *lp = td->td_lwp; 387 struct proc *p; 388 int sticks = 0; 389 int i = 0, ucode = 0, type, code; 390 #ifdef INVARIANTS 391 int crit_count = td->td_critcount; 392 lwkt_tokref_t curstop = td->td_toks_stop; 393 #endif 394 vm_offset_t eva; 395 396 p = td->td_proc; 397 clear_quickret(); 398 399 #ifdef DDB 400 /* 401 * We need to allow T_DNA faults when the debugger is active since 402 * some dumping paths do large bcopy() which use the floating 403 * point registers for faster copying. 404 */ 405 if (db_active && frame->tf_trapno != T_DNA) { 406 eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0); 407 ++gd->gd_trap_nesting_level; 408 trap_fatal(frame, eva); 409 --gd->gd_trap_nesting_level; 410 goto out2; 411 } 412 #endif 413 414 eva = 0; 415 416 if ((frame->tf_rflags & PSL_I) == 0) { 417 /* 418 * Buggy application or kernel code has disabled interrupts 419 * and then trapped. Enabling interrupts now is wrong, but 420 * it is better than running with interrupts disabled until 421 * they are accidentally enabled later. 422 */ 423 424 type = frame->tf_trapno; 425 if (ISPL(frame->tf_cs) == SEL_UPL) { 426 /* JG curproc can be NULL */ 427 kprintf( 428 "pid %ld (%s): trap %d with interrupts disabled\n", 429 (long)curproc->p_pid, curproc->p_comm, type); 430 } else if ((type == T_STKFLT || type == T_PROTFLT || 431 type == T_SEGNPFLT) && 432 frame->tf_rip == (long)doreti_iret) { 433 /* 434 * iretq fault from kernel mode during return to 435 * userland. 436 * 437 * This situation is expected, don't complain. 438 */ 439 } else if (type != T_NMI && type != T_BPTFLT && 440 type != T_TRCTRAP) { 441 /* 442 * XXX not quite right, since this may be for a 443 * multiple fault in user mode. 444 */ 445 kprintf("kernel trap %d (%s @ 0x%016jx) with " 446 "interrupts disabled\n", 447 type, 448 td->td_comm, 449 frame->tf_rip); 450 } 451 cpu_enable_intr(); 452 } 453 454 type = frame->tf_trapno; 455 code = frame->tf_err; 456 457 if (ISPL(frame->tf_cs) == SEL_UPL) { 458 /* user trap */ 459 460 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid, 461 frame->tf_trapno, eva); 462 463 userenter(td, p); 464 465 sticks = (int)td->td_sticks; 466 KASSERT(lp->lwp_md.md_regs == frame, 467 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame)); 468 469 switch (type) { 470 case T_PRIVINFLT: /* privileged instruction fault */ 471 i = SIGILL; 472 ucode = ILL_PRVOPC; 473 break; 474 475 case T_BPTFLT: /* bpt instruction fault */ 476 case T_TRCTRAP: /* trace trap */ 477 frame->tf_rflags &= ~PSL_T; 478 i = SIGTRAP; 479 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT); 480 break; 481 482 case T_ARITHTRAP: /* arithmetic trap */ 483 ucode = code; 484 i = SIGFPE; 485 break; 486 487 case T_ASTFLT: /* Allow process switch */ 488 mycpu->gd_cnt.v_soft++; 489 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) { 490 atomic_clear_int(&mycpu->gd_reqflags, 491 RQF_AST_OWEUPC); 492 addupc_task(p, p->p_prof.pr_addr, 493 p->p_prof.pr_ticks); 494 } 495 goto out; 496 497 case T_PROTFLT: /* general protection fault */ 498 i = SIGBUS; 499 ucode = BUS_OBJERR; 500 break; 501 case T_STKFLT: /* stack fault */ 502 case T_SEGNPFLT: /* segment not present fault */ 503 i = SIGBUS; 504 ucode = BUS_ADRERR; 505 break; 506 case T_TSSFLT: /* invalid TSS fault */ 507 case T_DOUBLEFLT: /* double fault */ 508 default: 509 i = SIGBUS; 510 ucode = BUS_OBJERR; 511 break; 512 513 case T_PAGEFLT: /* page fault */ 514 i = trap_pfault(frame, TRUE); 515 #ifdef DDB 516 if (frame->tf_rip == 0) { 517 /* used for kernel debugging only */ 518 while (freeze_on_seg_fault) 519 tsleep(p, 0, "freeze", hz * 20); 520 } 521 #endif 522 if (i == -1 || i == 0) 523 goto out; 524 if (i == SIGSEGV) { 525 ucode = SEGV_MAPERR; 526 } else { 527 i = SIGSEGV; 528 ucode = SEGV_ACCERR; 529 } 530 break; 531 532 case T_DIVIDE: /* integer divide fault */ 533 ucode = FPE_INTDIV; 534 i = SIGFPE; 535 break; 536 537 #if NISA > 0 538 case T_NMI: 539 /* machine/parity/power fail/"kitchen sink" faults */ 540 if (isa_nmi(code) == 0) { 541 #ifdef DDB 542 /* 543 * NMI can be hooked up to a pushbutton 544 * for debugging. 545 */ 546 if (ddb_on_nmi) { 547 kprintf ("NMI ... going to debugger\n"); 548 kdb_trap(type, 0, frame); 549 } 550 #endif /* DDB */ 551 goto out2; 552 } else if (panic_on_nmi) 553 panic("NMI indicates hardware failure"); 554 break; 555 #endif /* NISA > 0 */ 556 557 case T_OFLOW: /* integer overflow fault */ 558 ucode = FPE_INTOVF; 559 i = SIGFPE; 560 break; 561 562 case T_BOUND: /* bounds check fault */ 563 ucode = FPE_FLTSUB; 564 i = SIGFPE; 565 break; 566 567 case T_DNA: 568 /* 569 * Virtual kernel intercept - pass the DNA exception 570 * to the virtual kernel if it asked to handle it. 571 * This occurs when the virtual kernel is holding 572 * onto the FP context for a different emulated 573 * process then the one currently running. 574 * 575 * We must still call npxdna() since we may have 576 * saved FP state that the virtual kernel needs 577 * to hand over to a different emulated process. 578 */ 579 if (lp->lwp_vkernel && lp->lwp_vkernel->ve && 580 (td->td_pcb->pcb_flags & FP_VIRTFP) 581 ) { 582 npxdna(); 583 break; 584 } 585 586 /* 587 * The kernel may have switched out the FP unit's 588 * state, causing the user process to take a fault 589 * when it tries to use the FP unit. Restore the 590 * state here 591 */ 592 if (npxdna()) 593 goto out; 594 i = SIGFPE; 595 ucode = FPE_FPU_NP_TRAP; 596 break; 597 598 case T_FPOPFLT: /* FPU operand fetch fault */ 599 ucode = ILL_COPROC; 600 i = SIGILL; 601 break; 602 603 case T_XMMFLT: /* SIMD floating-point exception */ 604 ucode = 0; /* XXX */ 605 i = SIGFPE; 606 break; 607 } 608 } else { 609 /* kernel trap */ 610 611 switch (type) { 612 case T_PAGEFLT: /* page fault */ 613 trap_pfault(frame, FALSE); 614 goto out2; 615 616 case T_DNA: 617 /* 618 * The kernel is apparently using fpu for copying. 619 * XXX this should be fatal unless the kernel has 620 * registered such use. 621 */ 622 if (npxdna()) 623 goto out2; 624 break; 625 626 case T_STKFLT: /* stack fault */ 627 case T_PROTFLT: /* general protection fault */ 628 case T_SEGNPFLT: /* segment not present fault */ 629 /* 630 * Invalid segment selectors and out of bounds 631 * %rip's and %rsp's can be set up in user mode. 632 * This causes a fault in kernel mode when the 633 * kernel tries to return to user mode. We want 634 * to get this fault so that we can fix the 635 * problem here and not have to check all the 636 * selectors and pointers when the user changes 637 * them. 638 */ 639 if (mycpu->gd_intr_nesting_level == 0) { 640 /* 641 * NOTE: in 64-bit mode traps push rsp/ss 642 * even if no ring change occurs. 643 */ 644 if (td->td_pcb->pcb_onfault && 645 td->td_pcb->pcb_onfault_sp == 646 frame->tf_rsp) { 647 frame->tf_rip = (register_t) 648 td->td_pcb->pcb_onfault; 649 goto out2; 650 } 651 652 /* 653 * If the iretq in doreti faults during 654 * return to user, it will be special-cased 655 * in IDTVEC(prot) to get here. We want 656 * to 'return' to doreti_iret_fault in 657 * ipl.s in approximately the same state we 658 * were in at the iretq. 659 */ 660 if (frame->tf_rip == (long)doreti_iret) { 661 frame->tf_rip = (long)doreti_iret_fault; 662 goto out2; 663 } 664 } 665 break; 666 667 case T_TSSFLT: 668 /* 669 * PSL_NT can be set in user mode and isn't cleared 670 * automatically when the kernel is entered. This 671 * causes a TSS fault when the kernel attempts to 672 * `iret' because the TSS link is uninitialized. We 673 * want to get this fault so that we can fix the 674 * problem here and not every time the kernel is 675 * entered. 676 */ 677 if (frame->tf_rflags & PSL_NT) { 678 frame->tf_rflags &= ~PSL_NT; 679 #if 0 680 /* do we need this? */ 681 if (frame->tf_rip == (long)doreti_iret) 682 frame->tf_rip = (long)doreti_iret_fault; 683 #endif 684 goto out2; 685 } 686 break; 687 688 case T_TRCTRAP: /* trace trap */ 689 #if 0 690 if (frame->tf_rip == (int)IDTVEC(syscall)) { 691 /* 692 * We've just entered system mode via the 693 * syscall lcall. Continue single stepping 694 * silently until the syscall handler has 695 * saved the flags. 696 */ 697 goto out2; 698 } 699 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) { 700 /* 701 * The syscall handler has now saved the 702 * flags. Stop single stepping it. 703 */ 704 frame->tf_rflags &= ~PSL_T; 705 goto out2; 706 } 707 #endif 708 709 /* 710 * Ignore debug register trace traps due to 711 * accesses in the user's address space, which 712 * can happen under several conditions such as 713 * if a user sets a watchpoint on a buffer and 714 * then passes that buffer to a system call. 715 * We still want to get TRCTRAPS for addresses 716 * in kernel space because that is useful when 717 * debugging the kernel. 718 */ 719 #if 0 /* JG */ 720 if (user_dbreg_trap()) { 721 /* 722 * Reset breakpoint bits because the 723 * processor doesn't 724 */ 725 /* XXX check upper bits here */ 726 load_dr6(rdr6() & 0xfffffff0); 727 goto out2; 728 } 729 #endif 730 /* 731 * FALLTHROUGH (TRCTRAP kernel mode, kernel address) 732 */ 733 case T_BPTFLT: 734 /* 735 * If DDB is enabled, let it handle the debugger trap. 736 * Otherwise, debugger traps "can't happen". 737 */ 738 ucode = TRAP_BRKPT; 739 #ifdef DDB 740 if (kdb_trap(type, 0, frame)) 741 goto out2; 742 #endif 743 break; 744 745 #if NISA > 0 746 case T_NMI: 747 /* machine/parity/power fail/"kitchen sink" faults */ 748 if (isa_nmi(code) == 0) { 749 #ifdef DDB 750 /* 751 * NMI can be hooked up to a pushbutton 752 * for debugging. 753 */ 754 if (ddb_on_nmi) { 755 kprintf ("NMI ... going to debugger\n"); 756 kdb_trap(type, 0, frame); 757 } 758 #endif /* DDB */ 759 goto out2; 760 } else if (panic_on_nmi == 0) 761 goto out2; 762 /* FALL THROUGH */ 763 #endif /* NISA > 0 */ 764 } 765 trap_fatal(frame, 0); 766 goto out2; 767 } 768 769 /* 770 * Fault from user mode, virtual kernel interecept. 771 * 772 * If the fault is directly related to a VM context managed by a 773 * virtual kernel then let the virtual kernel handle it. 774 */ 775 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 776 vkernel_trap(lp, frame); 777 goto out; 778 } 779 780 /* Translate fault for emulators (e.g. Linux) */ 781 if (*p->p_sysent->sv_transtrap) 782 i = (*p->p_sysent->sv_transtrap)(i, type); 783 784 trapsignal(lp, i, ucode); 785 786 #ifdef DEBUG 787 if (type <= MAX_TRAP_MSG) { 788 uprintf("fatal process exception: %s", 789 trap_msg[type]); 790 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 791 uprintf(", fault VA = 0x%lx", frame->tf_addr); 792 uprintf("\n"); 793 } 794 #endif 795 796 out: 797 userret(lp, frame, sticks); 798 userexit(lp); 799 out2: ; 800 if (p != NULL && lp != NULL) 801 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid); 802 #ifdef INVARIANTS 803 KASSERT(crit_count == td->td_critcount, 804 ("trap: critical section count mismatch! %d/%d", 805 crit_count, td->td_pri)); 806 KASSERT(curstop == td->td_toks_stop, 807 ("trap: extra tokens held after trap! %ld/%ld", 808 curstop - &td->td_toks_base, 809 td->td_toks_stop - &td->td_toks_base)); 810 #endif 811 } 812 813 void 814 trap_handle_userenter(struct thread *td) 815 { 816 userenter(td, td->td_proc); 817 } 818 819 void 820 trap_handle_userexit(struct trapframe *frame, int sticks) 821 { 822 struct lwp *lp = curthread->td_lwp; 823 824 if (lp) { 825 userret(lp, frame, sticks); 826 userexit(lp); 827 } 828 } 829 830 static int 831 trap_pfault(struct trapframe *frame, int usermode) 832 { 833 vm_offset_t va; 834 struct vmspace *vm = NULL; 835 vm_map_t map; 836 int rv = 0; 837 int fault_flags; 838 vm_prot_t ftype; 839 thread_t td = curthread; 840 struct lwp *lp = td->td_lwp; 841 struct proc *p; 842 843 va = trunc_page(frame->tf_addr); 844 if (va >= VM_MIN_KERNEL_ADDRESS) { 845 /* 846 * Don't allow user-mode faults in kernel address space. 847 */ 848 if (usermode) { 849 fault_flags = -1; 850 ftype = -1; 851 goto nogo; 852 } 853 854 map = &kernel_map; 855 } else { 856 /* 857 * This is a fault on non-kernel virtual memory. 858 * vm is initialized above to NULL. If curproc is NULL 859 * or curproc->p_vmspace is NULL the fault is fatal. 860 */ 861 if (lp != NULL) 862 vm = lp->lwp_vmspace; 863 864 if (vm == NULL) { 865 fault_flags = -1; 866 ftype = -1; 867 goto nogo; 868 } 869 870 /* 871 * Debugging, try to catch kernel faults on the user address 872 * space when not inside on onfault (e.g. copyin/copyout) 873 * routine. 874 */ 875 if (usermode == 0 && (td->td_pcb == NULL || 876 td->td_pcb->pcb_onfault == NULL)) { 877 #ifdef DDB 878 if (freeze_on_seg_fault) { 879 kprintf("trap_pfault: user address fault from kernel mode " 880 "%016lx\n", (long)frame->tf_addr); 881 while (freeze_on_seg_fault) 882 tsleep(&freeze_on_seg_fault, 0, "frzseg", hz * 20); 883 } 884 #endif 885 } 886 map = &vm->vm_map; 887 } 888 889 /* 890 * PGEX_I is defined only if the execute disable bit capability is 891 * supported and enabled. 892 */ 893 if (frame->tf_err & PGEX_W) 894 ftype = VM_PROT_WRITE; 895 else if (frame->tf_err & PGEX_I) 896 ftype = VM_PROT_EXECUTE; 897 else 898 ftype = VM_PROT_READ; 899 900 if (map != &kernel_map) { 901 /* 902 * Keep swapout from messing with us during this 903 * critical time. 904 */ 905 PHOLD(lp->lwp_proc); 906 907 /* 908 * Issue fault 909 */ 910 fault_flags = 0; 911 if (usermode) 912 fault_flags |= VM_FAULT_BURST | VM_FAULT_USERMODE; 913 if (ftype & VM_PROT_WRITE) 914 fault_flags |= VM_FAULT_DIRTY; 915 else 916 fault_flags |= VM_FAULT_NORMAL; 917 rv = vm_fault(map, va, ftype, fault_flags); 918 919 PRELE(lp->lwp_proc); 920 } else { 921 /* 922 * Don't have to worry about process locking or stacks in the 923 * kernel. 924 */ 925 fault_flags = VM_FAULT_NORMAL; 926 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 927 } 928 if (rv == KERN_SUCCESS) 929 return (0); 930 nogo: 931 if (!usermode) { 932 /* 933 * NOTE: in 64-bit mode traps push rsp/ss 934 * even if no ring change occurs. 935 */ 936 if (td->td_pcb->pcb_onfault && 937 td->td_pcb->pcb_onfault_sp == frame->tf_rsp && 938 td->td_gd->gd_intr_nesting_level == 0) { 939 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault; 940 return (0); 941 } 942 trap_fatal(frame, frame->tf_addr); 943 return (-1); 944 } 945 946 /* 947 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no 948 * kludge is needed to pass the fault address to signal handlers. 949 */ 950 p = td->td_proc; 951 #ifdef DDB 952 if (td->td_lwp->lwp_vkernel == NULL) { 953 while (freeze_on_seg_fault) { 954 tsleep(p, 0, "freeze", hz * 20); 955 } 956 if (ddb_on_seg_fault) 957 Debugger("ddb_on_seg_fault"); 958 } 959 #endif 960 961 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 962 } 963 964 static void 965 trap_fatal(struct trapframe *frame, vm_offset_t eva) 966 { 967 int code, ss; 968 u_int type; 969 long rsp; 970 struct soft_segment_descriptor softseg; 971 char *msg; 972 973 code = frame->tf_err; 974 type = frame->tf_trapno; 975 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg); 976 977 if (type <= MAX_TRAP_MSG) 978 msg = trap_msg[type]; 979 else 980 msg = "UNKNOWN"; 981 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg, 982 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 983 /* three separate prints in case of a trap on an unmapped page */ 984 kprintf("cpuid = %d; ", mycpu->gd_cpuid); 985 if (lapic) 986 kprintf("lapic->id = %08x\n", lapic->id); 987 if (type == T_PAGEFLT) { 988 kprintf("fault virtual address = 0x%lx\n", eva); 989 kprintf("fault code = %s %s %s, %s\n", 990 code & PGEX_U ? "user" : "supervisor", 991 code & PGEX_W ? "write" : "read", 992 code & PGEX_I ? "instruction" : "data", 993 code & PGEX_P ? "protection violation" : "page not present"); 994 } 995 kprintf("instruction pointer = 0x%lx:0x%lx\n", 996 frame->tf_cs & 0xffff, frame->tf_rip); 997 if (ISPL(frame->tf_cs) == SEL_UPL) { 998 ss = frame->tf_ss & 0xffff; 999 rsp = frame->tf_rsp; 1000 } else { 1001 /* 1002 * NOTE: in 64-bit mode traps push rsp/ss even if no ring 1003 * change occurs. 1004 */ 1005 ss = GSEL(GDATA_SEL, SEL_KPL); 1006 rsp = frame->tf_rsp; 1007 } 1008 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp); 1009 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp); 1010 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n", 1011 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); 1012 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n", 1013 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32, 1014 softseg.ssd_gran); 1015 kprintf("processor eflags = "); 1016 if (frame->tf_rflags & PSL_T) 1017 kprintf("trace trap, "); 1018 if (frame->tf_rflags & PSL_I) 1019 kprintf("interrupt enabled, "); 1020 if (frame->tf_rflags & PSL_NT) 1021 kprintf("nested task, "); 1022 if (frame->tf_rflags & PSL_RF) 1023 kprintf("resume, "); 1024 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12); 1025 kprintf("current process = "); 1026 if (curproc) { 1027 kprintf("%lu\n", 1028 (u_long)curproc->p_pid); 1029 } else { 1030 kprintf("Idle\n"); 1031 } 1032 kprintf("current thread = pri %d ", curthread->td_pri); 1033 if (curthread->td_critcount) 1034 kprintf("(CRIT)"); 1035 kprintf("\n"); 1036 1037 #ifdef DDB 1038 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame)) 1039 return; 1040 #endif 1041 kprintf("trap number = %d\n", type); 1042 if (type <= MAX_TRAP_MSG) 1043 panic("%s", trap_msg[type]); 1044 else 1045 panic("unknown/reserved trap"); 1046 } 1047 1048 /* 1049 * Double fault handler. Called when a fault occurs while writing 1050 * a frame for a trap/exception onto the stack. This usually occurs 1051 * when the stack overflows (such is the case with infinite recursion, 1052 * for example). 1053 */ 1054 static __inline 1055 int 1056 in_kstack_guard(register_t rptr) 1057 { 1058 thread_t td = curthread; 1059 1060 if ((char *)rptr >= td->td_kstack && 1061 (char *)rptr < td->td_kstack + PAGE_SIZE) { 1062 return 1; 1063 } 1064 return 0; 1065 } 1066 1067 void 1068 dblfault_handler(struct trapframe *frame) 1069 { 1070 thread_t td = curthread; 1071 1072 if (in_kstack_guard(frame->tf_rsp) || in_kstack_guard(frame->tf_rbp)) { 1073 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n"); 1074 if (in_kstack_guard(frame->tf_rsp)) 1075 frame->tf_rsp = (register_t)(td->td_kstack + PAGE_SIZE); 1076 if (in_kstack_guard(frame->tf_rbp)) 1077 frame->tf_rbp = (register_t)(td->td_kstack + PAGE_SIZE); 1078 } else { 1079 kprintf("DOUBLE FAULT\n"); 1080 } 1081 kprintf("\nFatal double fault\n"); 1082 kprintf("rip = 0x%lx\n", frame->tf_rip); 1083 kprintf("rsp = 0x%lx\n", frame->tf_rsp); 1084 kprintf("rbp = 0x%lx\n", frame->tf_rbp); 1085 /* three separate prints in case of a trap on an unmapped page */ 1086 kprintf("cpuid = %d; ", mycpu->gd_cpuid); 1087 kprintf("lapic->id = %08x\n", lapic->id); 1088 panic("double fault"); 1089 } 1090 1091 /* 1092 * syscall2 - MP aware system call request C handler 1093 * 1094 * A system call is essentially treated as a trap except that the 1095 * MP lock is not held on entry or return. We are responsible for 1096 * obtaining the MP lock if necessary and for handling ASTs 1097 * (e.g. a task switch) prior to return. 1098 * 1099 * MPSAFE 1100 */ 1101 void 1102 syscall2(struct trapframe *frame) 1103 { 1104 struct thread *td = curthread; 1105 struct proc *p = td->td_proc; 1106 struct lwp *lp = td->td_lwp; 1107 caddr_t params; 1108 struct sysent *callp; 1109 register_t orig_tf_rflags; 1110 int sticks; 1111 int error; 1112 int narg; 1113 #ifdef INVARIANTS 1114 int crit_count = td->td_critcount; 1115 #endif 1116 register_t *argp; 1117 u_int code; 1118 int reg, regcnt; 1119 union sysunion args; 1120 register_t *argsdst; 1121 1122 mycpu->gd_cnt.v_syscall++; 1123 1124 #ifdef DIAGNOSTIC 1125 if (ISPL(frame->tf_cs) != SEL_UPL) { 1126 panic("syscall"); 1127 /* NOT REACHED */ 1128 } 1129 #endif 1130 1131 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid, 1132 frame->tf_rax); 1133 1134 userenter(td, p); /* lazy raise our priority */ 1135 1136 reg = 0; 1137 regcnt = 6; 1138 /* 1139 * Misc 1140 */ 1141 sticks = (int)td->td_sticks; 1142 orig_tf_rflags = frame->tf_rflags; 1143 1144 /* 1145 * Virtual kernel intercept - if a VM context managed by a virtual 1146 * kernel issues a system call the virtual kernel handles it, not us. 1147 * Restore the virtual kernel context and return from its system 1148 * call. The current frame is copied out to the virtual kernel. 1149 */ 1150 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 1151 vkernel_trap(lp, frame); 1152 error = EJUSTRETURN; 1153 callp = NULL; 1154 goto out; 1155 } 1156 1157 /* 1158 * Get the system call parameters and account for time 1159 */ 1160 KASSERT(lp->lwp_md.md_regs == frame, 1161 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame)); 1162 params = (caddr_t)frame->tf_rsp + sizeof(register_t); 1163 code = frame->tf_rax; 1164 1165 if (p->p_sysent->sv_prepsyscall) { 1166 (*p->p_sysent->sv_prepsyscall)( 1167 frame, (int *)(&args.nosys.sysmsg + 1), 1168 &code, ¶ms); 1169 } else { 1170 if (code == SYS_syscall || code == SYS___syscall) { 1171 code = frame->tf_rdi; 1172 reg++; 1173 regcnt--; 1174 } 1175 } 1176 1177 if (p->p_sysent->sv_mask) 1178 code &= p->p_sysent->sv_mask; 1179 1180 if (code >= p->p_sysent->sv_size) 1181 callp = &p->p_sysent->sv_table[0]; 1182 else 1183 callp = &p->p_sysent->sv_table[code]; 1184 1185 narg = callp->sy_narg & SYF_ARGMASK; 1186 1187 /* 1188 * On x86_64 we get up to six arguments in registers. The rest are 1189 * on the stack. The first six members of 'struct trapframe' happen 1190 * to be the registers used to pass arguments, in exactly the right 1191 * order. 1192 */ 1193 argp = &frame->tf_rdi; 1194 argp += reg; 1195 argsdst = (register_t *)(&args.nosys.sysmsg + 1); 1196 /* 1197 * JG can we overflow the space pointed to by 'argsdst' 1198 * either with 'bcopy' or with 'copyin'? 1199 */ 1200 bcopy(argp, argsdst, sizeof(register_t) * regcnt); 1201 /* 1202 * copyin is MP aware, but the tracing code is not 1203 */ 1204 if (narg > regcnt) { 1205 KASSERT(params != NULL, ("copyin args with no params!")); 1206 error = copyin(params, &argsdst[regcnt], 1207 (narg - regcnt) * sizeof(register_t)); 1208 if (error) { 1209 #ifdef KTRACE 1210 if (KTRPOINT(td, KTR_SYSCALL)) { 1211 ktrsyscall(lp, code, narg, 1212 (void *)(&args.nosys.sysmsg + 1)); 1213 } 1214 #endif 1215 goto bad; 1216 } 1217 } 1218 1219 #ifdef KTRACE 1220 if (KTRPOINT(td, KTR_SYSCALL)) { 1221 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1)); 1222 } 1223 #endif 1224 1225 /* 1226 * Default return value is 0 (will be copied to %rax). Double-value 1227 * returns use %rax and %rdx. %rdx is left unchanged for system 1228 * calls which return only one result. 1229 */ 1230 args.sysmsg_fds[0] = 0; 1231 args.sysmsg_fds[1] = frame->tf_rdx; 1232 1233 /* 1234 * The syscall might manipulate the trap frame. If it does it 1235 * will probably return EJUSTRETURN. 1236 */ 1237 args.sysmsg_frame = frame; 1238 1239 STOPEVENT(p, S_SCE, narg); /* MP aware */ 1240 1241 /* 1242 * NOTE: All system calls run MPSAFE now. The system call itself 1243 * is responsible for getting the MP lock. 1244 */ 1245 #ifdef SYSCALL_DEBUG 1246 tsc_uclock_t tscval = rdtsc(); 1247 #endif 1248 error = (*callp->sy_call)(&args); 1249 #ifdef SYSCALL_DEBUG 1250 tscval = rdtsc() - tscval; 1251 tscval = tscval * 1000000 / tsc_frequency; 1252 if (SysCallsWorstCase[code] < tscval) 1253 SysCallsWorstCase[code] = tscval; 1254 #endif 1255 1256 out: 1257 /* 1258 * MP SAFE (we may or may not have the MP lock at this point) 1259 */ 1260 //kprintf("SYSMSG %d ", error); 1261 switch (error) { 1262 case 0: 1263 /* 1264 * Reinitialize proc pointer `p' as it may be different 1265 * if this is a child returning from fork syscall. 1266 */ 1267 p = curproc; 1268 lp = curthread->td_lwp; 1269 frame->tf_rax = args.sysmsg_fds[0]; 1270 frame->tf_rdx = args.sysmsg_fds[1]; 1271 frame->tf_rflags &= ~PSL_C; 1272 break; 1273 case ERESTART: 1274 /* 1275 * Reconstruct pc, we know that 'syscall' is 2 bytes. 1276 * We have to do a full context restore so that %r10 1277 * (which was holding the value of %rcx) is restored for 1278 * the next iteration. 1279 */ 1280 if (frame->tf_err != 0 && frame->tf_err != 2) 1281 kprintf("lp %s:%d frame->tf_err is weird %ld\n", 1282 td->td_comm, lp->lwp_proc->p_pid, frame->tf_err); 1283 frame->tf_rip -= frame->tf_err; 1284 frame->tf_r10 = frame->tf_rcx; 1285 break; 1286 case EJUSTRETURN: 1287 break; 1288 case EASYNC: 1289 panic("Unexpected EASYNC return value (for now)"); 1290 default: 1291 bad: 1292 if (p->p_sysent->sv_errsize) { 1293 if (error >= p->p_sysent->sv_errsize) 1294 error = -1; /* XXX */ 1295 else 1296 error = p->p_sysent->sv_errtbl[error]; 1297 } 1298 frame->tf_rax = error; 1299 frame->tf_rflags |= PSL_C; 1300 break; 1301 } 1302 1303 /* 1304 * Traced syscall. trapsignal() should now be MP aware 1305 */ 1306 if (orig_tf_rflags & PSL_T) { 1307 frame->tf_rflags &= ~PSL_T; 1308 trapsignal(lp, SIGTRAP, TRAP_TRACE); 1309 } 1310 1311 /* 1312 * Handle reschedule and other end-of-syscall issues 1313 */ 1314 userret(lp, frame, sticks); 1315 1316 #ifdef KTRACE 1317 if (KTRPOINT(td, KTR_SYSRET)) { 1318 ktrsysret(lp, code, error, args.sysmsg_result); 1319 } 1320 #endif 1321 1322 /* 1323 * This works because errno is findable through the 1324 * register set. If we ever support an emulation where this 1325 * is not the case, this code will need to be revisited. 1326 */ 1327 STOPEVENT(p, S_SCX, code); 1328 1329 userexit(lp); 1330 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error); 1331 #ifdef INVARIANTS 1332 KASSERT(crit_count == td->td_critcount, 1333 ("syscall: critical section count mismatch! %d/%d", 1334 crit_count, td->td_pri)); 1335 KASSERT(&td->td_toks_base == td->td_toks_stop, 1336 ("syscall: %ld extra tokens held after trap! syscall %p", 1337 td->td_toks_stop - &td->td_toks_base, 1338 callp->sy_call)); 1339 #endif 1340 } 1341 1342 void 1343 fork_return(struct lwp *lp, struct trapframe *frame) 1344 { 1345 frame->tf_rax = 0; /* Child returns zero */ 1346 frame->tf_rflags &= ~PSL_C; /* success */ 1347 frame->tf_rdx = 1; 1348 1349 generic_lwp_return(lp, frame); 1350 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid); 1351 } 1352 1353 /* 1354 * Simplified back end of syscall(), used when returning from fork() 1355 * directly into user mode. 1356 * 1357 * This code will return back into the fork trampoline code which then 1358 * runs doreti. 1359 */ 1360 void 1361 generic_lwp_return(struct lwp *lp, struct trapframe *frame) 1362 { 1363 struct proc *p = lp->lwp_proc; 1364 1365 /* 1366 * Check for exit-race. If one lwp exits the process concurrent with 1367 * another lwp creating a new thread, the two operations may cross 1368 * each other resulting in the newly-created lwp not receiving a 1369 * KILL signal. 1370 */ 1371 if (p->p_flags & P_WEXIT) { 1372 lwpsignal(p, lp, SIGKILL); 1373 } 1374 1375 /* 1376 * Newly forked processes are given a kernel priority. We have to 1377 * adjust the priority to a normal user priority and fake entry 1378 * into the kernel (call userenter()) to install a passive release 1379 * function just in case userret() decides to stop the process. This 1380 * can occur when ^Z races a fork. If we do not install the passive 1381 * release function the current process designation will not be 1382 * released when the thread goes to sleep. 1383 */ 1384 lwkt_setpri_self(TDPRI_USER_NORM); 1385 userenter(lp->lwp_thread, p); 1386 userret(lp, frame, 0); 1387 #ifdef KTRACE 1388 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET)) 1389 ktrsysret(lp, SYS_fork, 0, 0); 1390 #endif 1391 lp->lwp_flags |= LWP_PASSIVE_ACQ; 1392 userexit(lp); 1393 lp->lwp_flags &= ~LWP_PASSIVE_ACQ; 1394 } 1395 1396 /* 1397 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA 1398 * fault (which is then passed back to the virtual kernel) if an attempt is 1399 * made to use the FP unit. 1400 * 1401 * XXX this is a fairly big hack. 1402 */ 1403 void 1404 set_vkernel_fp(struct trapframe *frame) 1405 { 1406 struct thread *td = curthread; 1407 1408 if (frame->tf_xflags & PGEX_FPFAULT) { 1409 td->td_pcb->pcb_flags |= FP_VIRTFP; 1410 if (mdcpu->gd_npxthread == td) 1411 npxexit(); 1412 } else { 1413 td->td_pcb->pcb_flags &= ~FP_VIRTFP; 1414 } 1415 } 1416 1417 /* 1418 * Called from vkernel_trap() to fixup the vkernel's syscall 1419 * frame for vmspace_ctl() return. 1420 */ 1421 void 1422 cpu_vkernel_trap(struct trapframe *frame, int error) 1423 { 1424 frame->tf_rax = error; 1425 if (error) 1426 frame->tf_rflags |= PSL_C; 1427 else 1428 frame->tf_rflags &= ~PSL_C; 1429 } 1430