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 type = frame->tf_trapno; 424 if (ISPL(frame->tf_cs) == SEL_UPL) { 425 /* JG curproc can be NULL */ 426 kprintf( 427 "pid %ld (%s): trap %d with interrupts disabled\n", 428 (long)curproc->p_pid, curproc->p_comm, type); 429 } else if (type != T_NMI && type != T_BPTFLT && 430 type != T_TRCTRAP) { 431 /* 432 * XXX not quite right, since this may be for a 433 * multiple fault in user mode. 434 */ 435 kprintf("kernel trap %d (%s @ 0x%016jx) with " 436 "interrupts disabled\n", 437 type, 438 td->td_comm, 439 frame->tf_rip); 440 } 441 cpu_enable_intr(); 442 } 443 444 type = frame->tf_trapno; 445 code = frame->tf_err; 446 447 if (ISPL(frame->tf_cs) == SEL_UPL) { 448 /* user trap */ 449 450 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid, 451 frame->tf_trapno, eva); 452 453 userenter(td, p); 454 455 sticks = (int)td->td_sticks; 456 KASSERT(lp->lwp_md.md_regs == frame, 457 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame)); 458 459 switch (type) { 460 case T_PRIVINFLT: /* privileged instruction fault */ 461 i = SIGILL; 462 ucode = ILL_PRVOPC; 463 break; 464 465 case T_BPTFLT: /* bpt instruction fault */ 466 case T_TRCTRAP: /* trace trap */ 467 frame->tf_rflags &= ~PSL_T; 468 i = SIGTRAP; 469 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT); 470 break; 471 472 case T_ARITHTRAP: /* arithmetic trap */ 473 ucode = code; 474 i = SIGFPE; 475 break; 476 477 case T_ASTFLT: /* Allow process switch */ 478 mycpu->gd_cnt.v_soft++; 479 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) { 480 atomic_clear_int(&mycpu->gd_reqflags, 481 RQF_AST_OWEUPC); 482 addupc_task(p, p->p_prof.pr_addr, 483 p->p_prof.pr_ticks); 484 } 485 goto out; 486 487 case T_PROTFLT: /* general protection fault */ 488 i = SIGBUS; 489 ucode = BUS_OBJERR; 490 break; 491 case T_STKFLT: /* stack fault */ 492 case T_SEGNPFLT: /* segment not present fault */ 493 i = SIGBUS; 494 ucode = BUS_ADRERR; 495 break; 496 case T_TSSFLT: /* invalid TSS fault */ 497 case T_DOUBLEFLT: /* double fault */ 498 default: 499 i = SIGBUS; 500 ucode = BUS_OBJERR; 501 break; 502 503 case T_PAGEFLT: /* page fault */ 504 i = trap_pfault(frame, TRUE); 505 #ifdef DDB 506 if (frame->tf_rip == 0) { 507 /* used for kernel debugging only */ 508 while (freeze_on_seg_fault) 509 tsleep(p, 0, "freeze", hz * 20); 510 } 511 #endif 512 if (i == -1 || i == 0) 513 goto out; 514 if (i == SIGSEGV) { 515 ucode = SEGV_MAPERR; 516 } else { 517 i = SIGSEGV; 518 ucode = SEGV_ACCERR; 519 } 520 break; 521 522 case T_DIVIDE: /* integer divide fault */ 523 ucode = FPE_INTDIV; 524 i = SIGFPE; 525 break; 526 527 #if NISA > 0 528 case T_NMI: 529 /* machine/parity/power fail/"kitchen sink" faults */ 530 if (isa_nmi(code) == 0) { 531 #ifdef DDB 532 /* 533 * NMI can be hooked up to a pushbutton 534 * for debugging. 535 */ 536 if (ddb_on_nmi) { 537 kprintf ("NMI ... going to debugger\n"); 538 kdb_trap(type, 0, frame); 539 } 540 #endif /* DDB */ 541 goto out2; 542 } else if (panic_on_nmi) 543 panic("NMI indicates hardware failure"); 544 break; 545 #endif /* NISA > 0 */ 546 547 case T_OFLOW: /* integer overflow fault */ 548 ucode = FPE_INTOVF; 549 i = SIGFPE; 550 break; 551 552 case T_BOUND: /* bounds check fault */ 553 ucode = FPE_FLTSUB; 554 i = SIGFPE; 555 break; 556 557 case T_DNA: 558 /* 559 * Virtual kernel intercept - pass the DNA exception 560 * to the virtual kernel if it asked to handle it. 561 * This occurs when the virtual kernel is holding 562 * onto the FP context for a different emulated 563 * process then the one currently running. 564 * 565 * We must still call npxdna() since we may have 566 * saved FP state that the virtual kernel needs 567 * to hand over to a different emulated process. 568 */ 569 if (lp->lwp_vkernel && lp->lwp_vkernel->ve && 570 (td->td_pcb->pcb_flags & FP_VIRTFP) 571 ) { 572 npxdna(); 573 break; 574 } 575 576 /* 577 * The kernel may have switched out the FP unit's 578 * state, causing the user process to take a fault 579 * when it tries to use the FP unit. Restore the 580 * state here 581 */ 582 if (npxdna()) 583 goto out; 584 i = SIGFPE; 585 ucode = FPE_FPU_NP_TRAP; 586 break; 587 588 case T_FPOPFLT: /* FPU operand fetch fault */ 589 ucode = ILL_COPROC; 590 i = SIGILL; 591 break; 592 593 case T_XMMFLT: /* SIMD floating-point exception */ 594 ucode = 0; /* XXX */ 595 i = SIGFPE; 596 break; 597 } 598 } else { 599 /* kernel trap */ 600 601 switch (type) { 602 case T_PAGEFLT: /* page fault */ 603 trap_pfault(frame, FALSE); 604 goto out2; 605 606 case T_DNA: 607 /* 608 * The kernel is apparently using fpu for copying. 609 * XXX this should be fatal unless the kernel has 610 * registered such use. 611 */ 612 if (npxdna()) 613 goto out2; 614 break; 615 616 case T_STKFLT: /* stack fault */ 617 case T_PROTFLT: /* general protection fault */ 618 case T_SEGNPFLT: /* segment not present fault */ 619 /* 620 * Invalid segment selectors and out of bounds 621 * %rip's and %rsp's can be set up in user mode. 622 * This causes a fault in kernel mode when the 623 * kernel tries to return to user mode. We want 624 * to get this fault so that we can fix the 625 * problem here and not have to check all the 626 * selectors and pointers when the user changes 627 * them. 628 */ 629 if (mycpu->gd_intr_nesting_level == 0) { 630 /* 631 * NOTE: in 64-bit mode traps push rsp/ss 632 * even if no ring change occurs. 633 */ 634 if (td->td_pcb->pcb_onfault && 635 td->td_pcb->pcb_onfault_sp == 636 frame->tf_rsp) { 637 frame->tf_rip = (register_t) 638 td->td_pcb->pcb_onfault; 639 goto out2; 640 } 641 if (frame->tf_rip == (long)doreti_iret) { 642 frame->tf_rip = (long)doreti_iret_fault; 643 goto out2; 644 } 645 } 646 break; 647 648 case T_TSSFLT: 649 /* 650 * PSL_NT can be set in user mode and isn't cleared 651 * automatically when the kernel is entered. This 652 * causes a TSS fault when the kernel attempts to 653 * `iret' because the TSS link is uninitialized. We 654 * want to get this fault so that we can fix the 655 * problem here and not every time the kernel is 656 * entered. 657 */ 658 if (frame->tf_rflags & PSL_NT) { 659 frame->tf_rflags &= ~PSL_NT; 660 #if 0 661 /* do we need this? */ 662 if (frame->tf_rip == (long)doreti_iret) 663 frame->tf_rip = (long)doreti_iret_fault; 664 #endif 665 goto out2; 666 } 667 break; 668 669 case T_TRCTRAP: /* trace trap */ 670 #if 0 671 if (frame->tf_rip == (int)IDTVEC(syscall)) { 672 /* 673 * We've just entered system mode via the 674 * syscall lcall. Continue single stepping 675 * silently until the syscall handler has 676 * saved the flags. 677 */ 678 goto out2; 679 } 680 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) { 681 /* 682 * The syscall handler has now saved the 683 * flags. Stop single stepping it. 684 */ 685 frame->tf_rflags &= ~PSL_T; 686 goto out2; 687 } 688 #endif 689 690 /* 691 * Ignore debug register trace traps due to 692 * accesses in the user's address space, which 693 * can happen under several conditions such as 694 * if a user sets a watchpoint on a buffer and 695 * then passes that buffer to a system call. 696 * We still want to get TRCTRAPS for addresses 697 * in kernel space because that is useful when 698 * debugging the kernel. 699 */ 700 #if 0 /* JG */ 701 if (user_dbreg_trap()) { 702 /* 703 * Reset breakpoint bits because the 704 * processor doesn't 705 */ 706 /* XXX check upper bits here */ 707 load_dr6(rdr6() & 0xfffffff0); 708 goto out2; 709 } 710 #endif 711 /* 712 * FALLTHROUGH (TRCTRAP kernel mode, kernel address) 713 */ 714 case T_BPTFLT: 715 /* 716 * If DDB is enabled, let it handle the debugger trap. 717 * Otherwise, debugger traps "can't happen". 718 */ 719 ucode = TRAP_BRKPT; 720 #ifdef DDB 721 if (kdb_trap(type, 0, frame)) 722 goto out2; 723 #endif 724 break; 725 726 #if NISA > 0 727 case T_NMI: 728 /* machine/parity/power fail/"kitchen sink" faults */ 729 if (isa_nmi(code) == 0) { 730 #ifdef DDB 731 /* 732 * NMI can be hooked up to a pushbutton 733 * for debugging. 734 */ 735 if (ddb_on_nmi) { 736 kprintf ("NMI ... going to debugger\n"); 737 kdb_trap(type, 0, frame); 738 } 739 #endif /* DDB */ 740 goto out2; 741 } else if (panic_on_nmi == 0) 742 goto out2; 743 /* FALL THROUGH */ 744 #endif /* NISA > 0 */ 745 } 746 trap_fatal(frame, 0); 747 goto out2; 748 } 749 750 /* 751 * Fault from user mode, virtual kernel interecept. 752 * 753 * If the fault is directly related to a VM context managed by a 754 * virtual kernel then let the virtual kernel handle it. 755 */ 756 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 757 vkernel_trap(lp, frame); 758 goto out; 759 } 760 761 /* Translate fault for emulators (e.g. Linux) */ 762 if (*p->p_sysent->sv_transtrap) 763 i = (*p->p_sysent->sv_transtrap)(i, type); 764 765 trapsignal(lp, i, ucode); 766 767 #ifdef DEBUG 768 if (type <= MAX_TRAP_MSG) { 769 uprintf("fatal process exception: %s", 770 trap_msg[type]); 771 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 772 uprintf(", fault VA = 0x%lx", frame->tf_addr); 773 uprintf("\n"); 774 } 775 #endif 776 777 out: 778 userret(lp, frame, sticks); 779 userexit(lp); 780 out2: ; 781 if (p != NULL && lp != NULL) 782 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid); 783 #ifdef INVARIANTS 784 KASSERT(crit_count == td->td_critcount, 785 ("trap: critical section count mismatch! %d/%d", 786 crit_count, td->td_pri)); 787 KASSERT(curstop == td->td_toks_stop, 788 ("trap: extra tokens held after trap! %ld/%ld", 789 curstop - &td->td_toks_base, 790 td->td_toks_stop - &td->td_toks_base)); 791 #endif 792 } 793 794 void 795 trap_handle_userenter(struct thread *td) 796 { 797 userenter(td, td->td_proc); 798 } 799 800 void 801 trap_handle_userexit(struct trapframe *frame, int sticks) 802 { 803 struct lwp *lp = curthread->td_lwp; 804 805 if (lp) { 806 userret(lp, frame, sticks); 807 userexit(lp); 808 } 809 } 810 811 static int 812 trap_pfault(struct trapframe *frame, int usermode) 813 { 814 vm_offset_t va; 815 struct vmspace *vm = NULL; 816 vm_map_t map; 817 int rv = 0; 818 int fault_flags; 819 vm_prot_t ftype; 820 thread_t td = curthread; 821 struct lwp *lp = td->td_lwp; 822 struct proc *p; 823 824 va = trunc_page(frame->tf_addr); 825 if (va >= VM_MIN_KERNEL_ADDRESS) { 826 /* 827 * Don't allow user-mode faults in kernel address space. 828 */ 829 if (usermode) { 830 fault_flags = -1; 831 ftype = -1; 832 goto nogo; 833 } 834 835 map = &kernel_map; 836 } else { 837 /* 838 * This is a fault on non-kernel virtual memory. 839 * vm is initialized above to NULL. If curproc is NULL 840 * or curproc->p_vmspace is NULL the fault is fatal. 841 */ 842 if (lp != NULL) 843 vm = lp->lwp_vmspace; 844 845 if (vm == NULL) { 846 fault_flags = -1; 847 ftype = -1; 848 goto nogo; 849 } 850 851 /* 852 * Debugging, try to catch kernel faults on the user address 853 * space when not inside on onfault (e.g. copyin/copyout) 854 * routine. 855 */ 856 if (usermode == 0 && (td->td_pcb == NULL || 857 td->td_pcb->pcb_onfault == NULL)) { 858 #ifdef DDB 859 if (freeze_on_seg_fault) { 860 kprintf("trap_pfault: user address fault from kernel mode " 861 "%016lx\n", (long)frame->tf_addr); 862 while (freeze_on_seg_fault) 863 tsleep(&freeze_on_seg_fault, 0, "frzseg", hz * 20); 864 } 865 #endif 866 } 867 map = &vm->vm_map; 868 } 869 870 /* 871 * PGEX_I is defined only if the execute disable bit capability is 872 * supported and enabled. 873 */ 874 if (frame->tf_err & PGEX_W) 875 ftype = VM_PROT_WRITE; 876 else if (frame->tf_err & PGEX_I) 877 ftype = VM_PROT_EXECUTE; 878 else 879 ftype = VM_PROT_READ; 880 881 if (map != &kernel_map) { 882 /* 883 * Keep swapout from messing with us during this 884 * critical time. 885 */ 886 PHOLD(lp->lwp_proc); 887 888 /* 889 * Issue fault 890 */ 891 fault_flags = 0; 892 if (usermode) 893 fault_flags |= VM_FAULT_BURST | VM_FAULT_USERMODE; 894 if (ftype & VM_PROT_WRITE) 895 fault_flags |= VM_FAULT_DIRTY; 896 else 897 fault_flags |= VM_FAULT_NORMAL; 898 rv = vm_fault(map, va, ftype, fault_flags); 899 900 PRELE(lp->lwp_proc); 901 } else { 902 /* 903 * Don't have to worry about process locking or stacks in the 904 * kernel. 905 */ 906 fault_flags = VM_FAULT_NORMAL; 907 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 908 } 909 if (rv == KERN_SUCCESS) 910 return (0); 911 nogo: 912 if (!usermode) { 913 /* 914 * NOTE: in 64-bit mode traps push rsp/ss 915 * even if no ring change occurs. 916 */ 917 if (td->td_pcb->pcb_onfault && 918 td->td_pcb->pcb_onfault_sp == frame->tf_rsp && 919 td->td_gd->gd_intr_nesting_level == 0) { 920 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault; 921 return (0); 922 } 923 trap_fatal(frame, frame->tf_addr); 924 return (-1); 925 } 926 927 /* 928 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no 929 * kludge is needed to pass the fault address to signal handlers. 930 */ 931 p = td->td_proc; 932 #ifdef DDB 933 if (td->td_lwp->lwp_vkernel == NULL) { 934 while (freeze_on_seg_fault) { 935 tsleep(p, 0, "freeze", hz * 20); 936 } 937 if (ddb_on_seg_fault) 938 Debugger("ddb_on_seg_fault"); 939 } 940 #endif 941 942 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 943 } 944 945 static void 946 trap_fatal(struct trapframe *frame, vm_offset_t eva) 947 { 948 int code, ss; 949 u_int type; 950 long rsp; 951 struct soft_segment_descriptor softseg; 952 char *msg; 953 954 code = frame->tf_err; 955 type = frame->tf_trapno; 956 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg); 957 958 if (type <= MAX_TRAP_MSG) 959 msg = trap_msg[type]; 960 else 961 msg = "UNKNOWN"; 962 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg, 963 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 964 /* three separate prints in case of a trap on an unmapped page */ 965 kprintf("cpuid = %d; ", mycpu->gd_cpuid); 966 if (lapic) 967 kprintf("lapic->id = %08x\n", lapic->id); 968 if (type == T_PAGEFLT) { 969 kprintf("fault virtual address = 0x%lx\n", eva); 970 kprintf("fault code = %s %s %s, %s\n", 971 code & PGEX_U ? "user" : "supervisor", 972 code & PGEX_W ? "write" : "read", 973 code & PGEX_I ? "instruction" : "data", 974 code & PGEX_P ? "protection violation" : "page not present"); 975 } 976 kprintf("instruction pointer = 0x%lx:0x%lx\n", 977 frame->tf_cs & 0xffff, frame->tf_rip); 978 if (ISPL(frame->tf_cs) == SEL_UPL) { 979 ss = frame->tf_ss & 0xffff; 980 rsp = frame->tf_rsp; 981 } else { 982 /* 983 * NOTE: in 64-bit mode traps push rsp/ss even if no ring 984 * change occurs. 985 */ 986 ss = GSEL(GDATA_SEL, SEL_KPL); 987 rsp = frame->tf_rsp; 988 } 989 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp); 990 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp); 991 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n", 992 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); 993 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n", 994 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32, 995 softseg.ssd_gran); 996 kprintf("processor eflags = "); 997 if (frame->tf_rflags & PSL_T) 998 kprintf("trace trap, "); 999 if (frame->tf_rflags & PSL_I) 1000 kprintf("interrupt enabled, "); 1001 if (frame->tf_rflags & PSL_NT) 1002 kprintf("nested task, "); 1003 if (frame->tf_rflags & PSL_RF) 1004 kprintf("resume, "); 1005 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12); 1006 kprintf("current process = "); 1007 if (curproc) { 1008 kprintf("%lu\n", 1009 (u_long)curproc->p_pid); 1010 } else { 1011 kprintf("Idle\n"); 1012 } 1013 kprintf("current thread = pri %d ", curthread->td_pri); 1014 if (curthread->td_critcount) 1015 kprintf("(CRIT)"); 1016 kprintf("\n"); 1017 1018 #ifdef DDB 1019 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame)) 1020 return; 1021 #endif 1022 kprintf("trap number = %d\n", type); 1023 if (type <= MAX_TRAP_MSG) 1024 panic("%s", trap_msg[type]); 1025 else 1026 panic("unknown/reserved trap"); 1027 } 1028 1029 /* 1030 * Double fault handler. Called when a fault occurs while writing 1031 * a frame for a trap/exception onto the stack. This usually occurs 1032 * when the stack overflows (such is the case with infinite recursion, 1033 * for example). 1034 */ 1035 static __inline 1036 int 1037 in_kstack_guard(register_t rptr) 1038 { 1039 thread_t td = curthread; 1040 1041 if ((char *)rptr >= td->td_kstack && 1042 (char *)rptr < td->td_kstack + PAGE_SIZE) { 1043 return 1; 1044 } 1045 return 0; 1046 } 1047 1048 void 1049 dblfault_handler(struct trapframe *frame) 1050 { 1051 thread_t td = curthread; 1052 1053 if (in_kstack_guard(frame->tf_rsp) || in_kstack_guard(frame->tf_rbp)) { 1054 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n"); 1055 if (in_kstack_guard(frame->tf_rsp)) 1056 frame->tf_rsp = (register_t)(td->td_kstack + PAGE_SIZE); 1057 if (in_kstack_guard(frame->tf_rbp)) 1058 frame->tf_rbp = (register_t)(td->td_kstack + PAGE_SIZE); 1059 } else { 1060 kprintf("DOUBLE FAULT\n"); 1061 } 1062 kprintf("\nFatal double fault\n"); 1063 kprintf("rip = 0x%lx\n", frame->tf_rip); 1064 kprintf("rsp = 0x%lx\n", frame->tf_rsp); 1065 kprintf("rbp = 0x%lx\n", frame->tf_rbp); 1066 /* three separate prints in case of a trap on an unmapped page */ 1067 kprintf("cpuid = %d; ", mycpu->gd_cpuid); 1068 kprintf("lapic->id = %08x\n", lapic->id); 1069 panic("double fault"); 1070 } 1071 1072 /* 1073 * syscall2 - MP aware system call request C handler 1074 * 1075 * A system call is essentially treated as a trap except that the 1076 * MP lock is not held on entry or return. We are responsible for 1077 * obtaining the MP lock if necessary and for handling ASTs 1078 * (e.g. a task switch) prior to return. 1079 * 1080 * MPSAFE 1081 */ 1082 void 1083 syscall2(struct trapframe *frame) 1084 { 1085 struct thread *td = curthread; 1086 struct proc *p = td->td_proc; 1087 struct lwp *lp = td->td_lwp; 1088 caddr_t params; 1089 struct sysent *callp; 1090 register_t orig_tf_rflags; 1091 int sticks; 1092 int error; 1093 int narg; 1094 #ifdef INVARIANTS 1095 int crit_count = td->td_critcount; 1096 #endif 1097 register_t *argp; 1098 u_int code; 1099 int reg, regcnt; 1100 union sysunion args; 1101 register_t *argsdst; 1102 1103 mycpu->gd_cnt.v_syscall++; 1104 1105 #ifdef DIAGNOSTIC 1106 if (ISPL(frame->tf_cs) != SEL_UPL) { 1107 panic("syscall"); 1108 /* NOT REACHED */ 1109 } 1110 #endif 1111 1112 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid, 1113 frame->tf_rax); 1114 1115 userenter(td, p); /* lazy raise our priority */ 1116 1117 reg = 0; 1118 regcnt = 6; 1119 /* 1120 * Misc 1121 */ 1122 sticks = (int)td->td_sticks; 1123 orig_tf_rflags = frame->tf_rflags; 1124 1125 /* 1126 * Virtual kernel intercept - if a VM context managed by a virtual 1127 * kernel issues a system call the virtual kernel handles it, not us. 1128 * Restore the virtual kernel context and return from its system 1129 * call. The current frame is copied out to the virtual kernel. 1130 */ 1131 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 1132 vkernel_trap(lp, frame); 1133 error = EJUSTRETURN; 1134 callp = NULL; 1135 goto out; 1136 } 1137 1138 /* 1139 * Get the system call parameters and account for time 1140 */ 1141 KASSERT(lp->lwp_md.md_regs == frame, 1142 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame)); 1143 params = (caddr_t)frame->tf_rsp + sizeof(register_t); 1144 code = frame->tf_rax; 1145 1146 if (p->p_sysent->sv_prepsyscall) { 1147 (*p->p_sysent->sv_prepsyscall)( 1148 frame, (int *)(&args.nosys.sysmsg + 1), 1149 &code, ¶ms); 1150 } else { 1151 if (code == SYS_syscall || code == SYS___syscall) { 1152 code = frame->tf_rdi; 1153 reg++; 1154 regcnt--; 1155 } 1156 } 1157 1158 if (p->p_sysent->sv_mask) 1159 code &= p->p_sysent->sv_mask; 1160 1161 if (code >= p->p_sysent->sv_size) 1162 callp = &p->p_sysent->sv_table[0]; 1163 else 1164 callp = &p->p_sysent->sv_table[code]; 1165 1166 narg = callp->sy_narg & SYF_ARGMASK; 1167 1168 /* 1169 * On x86_64 we get up to six arguments in registers. The rest are 1170 * on the stack. The first six members of 'struct trapframe' happen 1171 * to be the registers used to pass arguments, in exactly the right 1172 * order. 1173 */ 1174 argp = &frame->tf_rdi; 1175 argp += reg; 1176 argsdst = (register_t *)(&args.nosys.sysmsg + 1); 1177 /* 1178 * JG can we overflow the space pointed to by 'argsdst' 1179 * either with 'bcopy' or with 'copyin'? 1180 */ 1181 bcopy(argp, argsdst, sizeof(register_t) * regcnt); 1182 /* 1183 * copyin is MP aware, but the tracing code is not 1184 */ 1185 if (narg > regcnt) { 1186 KASSERT(params != NULL, ("copyin args with no params!")); 1187 error = copyin(params, &argsdst[regcnt], 1188 (narg - regcnt) * sizeof(register_t)); 1189 if (error) { 1190 #ifdef KTRACE 1191 if (KTRPOINT(td, KTR_SYSCALL)) { 1192 ktrsyscall(lp, code, narg, 1193 (void *)(&args.nosys.sysmsg + 1)); 1194 } 1195 #endif 1196 goto bad; 1197 } 1198 } 1199 1200 #ifdef KTRACE 1201 if (KTRPOINT(td, KTR_SYSCALL)) { 1202 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1)); 1203 } 1204 #endif 1205 1206 /* 1207 * Default return value is 0 (will be copied to %rax). Double-value 1208 * returns use %rax and %rdx. %rdx is left unchanged for system 1209 * calls which return only one result. 1210 */ 1211 args.sysmsg_fds[0] = 0; 1212 args.sysmsg_fds[1] = frame->tf_rdx; 1213 1214 /* 1215 * The syscall might manipulate the trap frame. If it does it 1216 * will probably return EJUSTRETURN. 1217 */ 1218 args.sysmsg_frame = frame; 1219 1220 STOPEVENT(p, S_SCE, narg); /* MP aware */ 1221 1222 /* 1223 * NOTE: All system calls run MPSAFE now. The system call itself 1224 * is responsible for getting the MP lock. 1225 */ 1226 #ifdef SYSCALL_DEBUG 1227 tsc_uclock_t tscval = rdtsc(); 1228 #endif 1229 error = (*callp->sy_call)(&args); 1230 #ifdef SYSCALL_DEBUG 1231 tscval = rdtsc() - tscval; 1232 tscval = tscval * 1000000 / tsc_frequency; 1233 if (SysCallsWorstCase[code] < tscval) 1234 SysCallsWorstCase[code] = tscval; 1235 #endif 1236 1237 out: 1238 /* 1239 * MP SAFE (we may or may not have the MP lock at this point) 1240 */ 1241 //kprintf("SYSMSG %d ", error); 1242 switch (error) { 1243 case 0: 1244 /* 1245 * Reinitialize proc pointer `p' as it may be different 1246 * if this is a child returning from fork syscall. 1247 */ 1248 p = curproc; 1249 lp = curthread->td_lwp; 1250 frame->tf_rax = args.sysmsg_fds[0]; 1251 frame->tf_rdx = args.sysmsg_fds[1]; 1252 frame->tf_rflags &= ~PSL_C; 1253 break; 1254 case ERESTART: 1255 /* 1256 * Reconstruct pc, we know that 'syscall' is 2 bytes. 1257 * We have to do a full context restore so that %r10 1258 * (which was holding the value of %rcx) is restored for 1259 * the next iteration. 1260 */ 1261 if (frame->tf_err != 0 && frame->tf_err != 2) 1262 kprintf("lp %s:%d frame->tf_err is weird %ld\n", 1263 td->td_comm, lp->lwp_proc->p_pid, frame->tf_err); 1264 frame->tf_rip -= frame->tf_err; 1265 frame->tf_r10 = frame->tf_rcx; 1266 break; 1267 case EJUSTRETURN: 1268 break; 1269 case EASYNC: 1270 panic("Unexpected EASYNC return value (for now)"); 1271 default: 1272 bad: 1273 if (p->p_sysent->sv_errsize) { 1274 if (error >= p->p_sysent->sv_errsize) 1275 error = -1; /* XXX */ 1276 else 1277 error = p->p_sysent->sv_errtbl[error]; 1278 } 1279 frame->tf_rax = error; 1280 frame->tf_rflags |= PSL_C; 1281 break; 1282 } 1283 1284 /* 1285 * Traced syscall. trapsignal() should now be MP aware 1286 */ 1287 if (orig_tf_rflags & PSL_T) { 1288 frame->tf_rflags &= ~PSL_T; 1289 trapsignal(lp, SIGTRAP, TRAP_TRACE); 1290 } 1291 1292 /* 1293 * Handle reschedule and other end-of-syscall issues 1294 */ 1295 userret(lp, frame, sticks); 1296 1297 #ifdef KTRACE 1298 if (KTRPOINT(td, KTR_SYSRET)) { 1299 ktrsysret(lp, code, error, args.sysmsg_result); 1300 } 1301 #endif 1302 1303 /* 1304 * This works because errno is findable through the 1305 * register set. If we ever support an emulation where this 1306 * is not the case, this code will need to be revisited. 1307 */ 1308 STOPEVENT(p, S_SCX, code); 1309 1310 userexit(lp); 1311 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error); 1312 #ifdef INVARIANTS 1313 KASSERT(crit_count == td->td_critcount, 1314 ("syscall: critical section count mismatch! %d/%d", 1315 crit_count, td->td_pri)); 1316 KASSERT(&td->td_toks_base == td->td_toks_stop, 1317 ("syscall: %ld extra tokens held after trap! syscall %p", 1318 td->td_toks_stop - &td->td_toks_base, 1319 callp->sy_call)); 1320 #endif 1321 } 1322 1323 void 1324 fork_return(struct lwp *lp, struct trapframe *frame) 1325 { 1326 frame->tf_rax = 0; /* Child returns zero */ 1327 frame->tf_rflags &= ~PSL_C; /* success */ 1328 frame->tf_rdx = 1; 1329 1330 generic_lwp_return(lp, frame); 1331 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid); 1332 } 1333 1334 /* 1335 * Simplified back end of syscall(), used when returning from fork() 1336 * directly into user mode. 1337 * 1338 * This code will return back into the fork trampoline code which then 1339 * runs doreti. 1340 */ 1341 void 1342 generic_lwp_return(struct lwp *lp, struct trapframe *frame) 1343 { 1344 struct proc *p = lp->lwp_proc; 1345 1346 /* 1347 * Check for exit-race. If one lwp exits the process concurrent with 1348 * another lwp creating a new thread, the two operations may cross 1349 * each other resulting in the newly-created lwp not receiving a 1350 * KILL signal. 1351 */ 1352 if (p->p_flags & P_WEXIT) { 1353 lwpsignal(p, lp, SIGKILL); 1354 } 1355 1356 /* 1357 * Newly forked processes are given a kernel priority. We have to 1358 * adjust the priority to a normal user priority and fake entry 1359 * into the kernel (call userenter()) to install a passive release 1360 * function just in case userret() decides to stop the process. This 1361 * can occur when ^Z races a fork. If we do not install the passive 1362 * release function the current process designation will not be 1363 * released when the thread goes to sleep. 1364 */ 1365 lwkt_setpri_self(TDPRI_USER_NORM); 1366 userenter(lp->lwp_thread, p); 1367 userret(lp, frame, 0); 1368 #ifdef KTRACE 1369 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET)) 1370 ktrsysret(lp, SYS_fork, 0, 0); 1371 #endif 1372 lp->lwp_flags |= LWP_PASSIVE_ACQ; 1373 userexit(lp); 1374 lp->lwp_flags &= ~LWP_PASSIVE_ACQ; 1375 } 1376 1377 /* 1378 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA 1379 * fault (which is then passed back to the virtual kernel) if an attempt is 1380 * made to use the FP unit. 1381 * 1382 * XXX this is a fairly big hack. 1383 */ 1384 void 1385 set_vkernel_fp(struct trapframe *frame) 1386 { 1387 struct thread *td = curthread; 1388 1389 if (frame->tf_xflags & PGEX_FPFAULT) { 1390 td->td_pcb->pcb_flags |= FP_VIRTFP; 1391 if (mdcpu->gd_npxthread == td) 1392 npxexit(); 1393 } else { 1394 td->td_pcb->pcb_flags &= ~FP_VIRTFP; 1395 } 1396 } 1397 1398 /* 1399 * Called from vkernel_trap() to fixup the vkernel's syscall 1400 * frame for vmspace_ctl() return. 1401 */ 1402 void 1403 cpu_vkernel_trap(struct trapframe *frame, int error) 1404 { 1405 frame->tf_rax = error; 1406 if (error) 1407 frame->tf_rflags |= PSL_C; 1408 else 1409 frame->tf_rflags &= ~PSL_C; 1410 } 1411