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 #include <sys/systm.h> 65 #ifdef KTRACE 66 #include <sys/ktrace.h> 67 #endif 68 #include <sys/ktr.h> 69 #include <sys/sysmsg.h> 70 #include <sys/sysproto.h> 71 #include <sys/sysunion.h> 72 73 #include <vm/pmap.h> 74 #include <vm/vm.h> 75 #include <vm/vm_extern.h> 76 #include <vm/vm_kern.h> 77 #include <vm/vm_param.h> 78 #include <machine/cpu.h> 79 #include <machine/pcb.h> 80 #include <machine/smp.h> 81 #include <machine/thread.h> 82 #include <machine/clock.h> 83 #include <machine/vmparam.h> 84 #include <machine/md_var.h> 85 #include <machine_base/isa/isa_intr.h> 86 #include <machine_base/apic/lapic.h> 87 88 #include <ddb/ddb.h> 89 90 #include <sys/thread2.h> 91 #include <sys/mplock2.h> 92 93 #ifdef SMP 94 95 #define MAKEMPSAFE(have_mplock) \ 96 if (have_mplock == 0) { \ 97 get_mplock(); \ 98 have_mplock = 1; \ 99 } 100 101 #else 102 103 #define MAKEMPSAFE(have_mplock) 104 105 #endif 106 107 extern void trap(struct trapframe *frame); 108 109 static int trap_pfault(struct trapframe *, int); 110 static void trap_fatal(struct trapframe *, vm_offset_t); 111 void dblfault_handler(struct trapframe *frame); 112 113 #define MAX_TRAP_MSG 30 114 static char *trap_msg[] = { 115 "", /* 0 unused */ 116 "privileged instruction fault", /* 1 T_PRIVINFLT */ 117 "", /* 2 unused */ 118 "breakpoint instruction fault", /* 3 T_BPTFLT */ 119 "", /* 4 unused */ 120 "", /* 5 unused */ 121 "arithmetic trap", /* 6 T_ARITHTRAP */ 122 "system forced exception", /* 7 T_ASTFLT */ 123 "", /* 8 unused */ 124 "general protection fault", /* 9 T_PROTFLT */ 125 "trace trap", /* 10 T_TRCTRAP */ 126 "", /* 11 unused */ 127 "page fault", /* 12 T_PAGEFLT */ 128 "", /* 13 unused */ 129 "alignment fault", /* 14 T_ALIGNFLT */ 130 "", /* 15 unused */ 131 "", /* 16 unused */ 132 "", /* 17 unused */ 133 "integer divide fault", /* 18 T_DIVIDE */ 134 "non-maskable interrupt trap", /* 19 T_NMI */ 135 "overflow trap", /* 20 T_OFLOW */ 136 "FPU bounds check fault", /* 21 T_BOUND */ 137 "FPU device not available", /* 22 T_DNA */ 138 "double fault", /* 23 T_DOUBLEFLT */ 139 "FPU operand fetch fault", /* 24 T_FPOPFLT */ 140 "invalid TSS fault", /* 25 T_TSSFLT */ 141 "segment not present fault", /* 26 T_SEGNPFLT */ 142 "stack fault", /* 27 T_STKFLT */ 143 "machine check trap", /* 28 T_MCHK */ 144 "SIMD floating-point exception", /* 29 T_XMMFLT */ 145 "reserved (unknown) fault", /* 30 T_RESERVED */ 146 }; 147 148 #ifdef DDB 149 static int ddb_on_nmi = 1; 150 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW, 151 &ddb_on_nmi, 0, "Go to DDB on NMI"); 152 static int ddb_on_seg_fault = 0; 153 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_seg_fault, CTLFLAG_RW, 154 &ddb_on_seg_fault, 0, "Go to DDB on user seg-fault"); 155 static int freeze_on_seg_fault = 0; 156 SYSCTL_INT(_machdep, OID_AUTO, freeze_on_seg_fault, CTLFLAG_RW, 157 &freeze_on_seg_fault, 0, "Go to DDB on user seg-fault"); 158 #endif 159 static int panic_on_nmi = 1; 160 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW, 161 &panic_on_nmi, 0, "Panic on NMI"); 162 static int fast_release; 163 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW, 164 &fast_release, 0, "Passive Release was optimal"); 165 static int slow_release; 166 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW, 167 &slow_release, 0, "Passive Release was nonoptimal"); 168 169 /* 170 * System call debugging records the worst-case system call 171 * overhead (inclusive of blocking), but may be inaccurate. 172 */ 173 /*#define SYSCALL_DEBUG*/ 174 #ifdef SYSCALL_DEBUG 175 uint64_t SysCallsWorstCase[SYS_MAXSYSCALL]; 176 #endif 177 178 /* 179 * Passively intercepts the thread switch function to increase 180 * the thread priority from a user priority to a kernel priority, reducing 181 * syscall and trap overhead for the case where no switch occurs. 182 * 183 * Synchronizes td_ucred with p_ucred. This is used by system calls, 184 * signal handling, faults, AST traps, and anything else that enters the 185 * kernel from userland and provides the kernel with a stable read-only 186 * copy of the process ucred. 187 */ 188 static __inline void 189 userenter(struct thread *curtd, struct proc *curp) 190 { 191 struct ucred *ocred; 192 struct ucred *ncred; 193 194 curtd->td_release = lwkt_passive_release; 195 196 if (curtd->td_ucred != curp->p_ucred) { 197 ncred = crhold(curp->p_ucred); 198 ocred = curtd->td_ucred; 199 curtd->td_ucred = ncred; 200 if (ocred) 201 crfree(ocred); 202 } 203 204 /* 205 * Debugging, remove top two user stack pages to catch kernel faults 206 */ 207 if (freeze_on_seg_fault > 1 && curtd->td_lwp) { 208 pmap_remove(vmspace_pmap(curtd->td_lwp->lwp_vmspace), 209 0x00007FFFFFFFD000LU, 210 0x0000800000000000LU); 211 } 212 } 213 214 /* 215 * Handle signals, upcalls, profiling, and other AST's and/or tasks that 216 * must be completed before we can return to or try to return to userland. 217 * 218 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64 219 * arithmatic on the delta calculation so the absolute tick values are 220 * truncated to an integer. 221 */ 222 static void 223 userret(struct lwp *lp, struct trapframe *frame, int sticks) 224 { 225 struct proc *p = lp->lwp_proc; 226 int sig; 227 228 /* 229 * Charge system time if profiling. Note: times are in microseconds. 230 * This may do a copyout and block, so do it first even though it 231 * means some system time will be charged as user time. 232 */ 233 if (p->p_flags & P_PROFIL) { 234 addupc_task(p, frame->tf_rip, 235 (u_int)((int)lp->lwp_thread->td_sticks - sticks)); 236 } 237 238 recheck: 239 /* 240 * If the jungle wants us dead, so be it. 241 */ 242 if (lp->lwp_mpflags & LWP_MP_WEXIT) { 243 lwkt_gettoken(&p->p_token); 244 lwp_exit(0); 245 lwkt_reltoken(&p->p_token); /* NOT REACHED */ 246 } 247 248 /* 249 * Block here if we are in a stopped state. 250 */ 251 if (p->p_stat == SSTOP || dump_stop_usertds) { 252 lwkt_gettoken(&p->p_token); 253 tstop(); 254 lwkt_reltoken(&p->p_token); 255 goto recheck; 256 } 257 258 /* 259 * Post any pending upcalls. If running a virtual kernel be sure 260 * to restore the virtual kernel's vmspace before posting the upcall. 261 */ 262 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF | P_UPCALLPEND)) { 263 lwkt_gettoken(&p->p_token); 264 if (p->p_flags & P_SIGVTALRM) { 265 p->p_flags &= ~P_SIGVTALRM; 266 ksignal(p, SIGVTALRM); 267 } 268 if (p->p_flags & P_SIGPROF) { 269 p->p_flags &= ~P_SIGPROF; 270 ksignal(p, SIGPROF); 271 } 272 if (p->p_flags & P_UPCALLPEND) { 273 p->p_flags &= ~P_UPCALLPEND; 274 postupcall(lp); 275 } 276 lwkt_reltoken(&p->p_token); 277 goto recheck; 278 } 279 280 /* 281 * Post any pending signals. If running a virtual kernel be sure 282 * to restore the virtual kernel's vmspace before posting the signal. 283 * 284 * WARNING! postsig() can exit and not return. 285 */ 286 if ((sig = CURSIG_TRACE(lp)) != 0) { 287 lwkt_gettoken(&p->p_token); 288 postsig(sig); 289 lwkt_reltoken(&p->p_token); 290 goto recheck; 291 } 292 293 /* 294 * block here if we are swapped out, but still process signals 295 * (such as SIGKILL). proc0 (the swapin scheduler) is already 296 * aware of our situation, we do not have to wake it up. 297 */ 298 if (p->p_flags & P_SWAPPEDOUT) { 299 lwkt_gettoken(&p->p_token); 300 get_mplock(); 301 p->p_flags |= P_SWAPWAIT; 302 swapin_request(); 303 if (p->p_flags & P_SWAPWAIT) 304 tsleep(p, PCATCH, "SWOUT", 0); 305 p->p_flags &= ~P_SWAPWAIT; 306 rel_mplock(); 307 lwkt_reltoken(&p->p_token); 308 goto recheck; 309 } 310 311 /* 312 * In a multi-threaded program it is possible for a thread to change 313 * signal state during a system call which temporarily changes the 314 * signal mask. In this case postsig() might not be run and we 315 * have to restore the mask ourselves. 316 */ 317 if (lp->lwp_flags & LWP_OLDMASK) { 318 lp->lwp_flags &= ~LWP_OLDMASK; 319 lp->lwp_sigmask = lp->lwp_oldsigmask; 320 goto recheck; 321 } 322 } 323 324 /* 325 * Cleanup from userenter and any passive release that might have occured. 326 * We must reclaim the current-process designation before we can return 327 * to usermode. We also handle both LWKT and USER reschedule requests. 328 */ 329 static __inline void 330 userexit(struct lwp *lp) 331 { 332 struct thread *td = lp->lwp_thread; 333 /* globaldata_t gd = td->td_gd; */ 334 335 /* 336 * Handle stop requests at kernel priority. Any requests queued 337 * after this loop will generate another AST. 338 */ 339 while (lp->lwp_proc->p_stat == SSTOP) { 340 lwkt_gettoken(&lp->lwp_proc->p_token); 341 tstop(); 342 lwkt_reltoken(&lp->lwp_proc->p_token); 343 } 344 345 /* 346 * Reduce our priority in preparation for a return to userland. If 347 * our passive release function was still in place, our priority was 348 * never raised and does not need to be reduced. 349 */ 350 lwkt_passive_recover(td); 351 352 /* 353 * Become the current user scheduled process if we aren't already, 354 * and deal with reschedule requests and other factors. 355 */ 356 lp->lwp_proc->p_usched->acquire_curproc(lp); 357 /* WARNING: we may have migrated cpu's */ 358 /* gd = td->td_gd; */ 359 } 360 361 #if !defined(KTR_KERNENTRY) 362 #define KTR_KERNENTRY KTR_ALL 363 #endif 364 KTR_INFO_MASTER(kernentry); 365 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, 366 "TRAP(pid %d, tid %d, trapno %ld, eva %lu)", 367 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva); 368 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %d, tid %d)", 369 pid_t pid, lwpid_t tid); 370 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %d, tid %d, nr %ld)", 371 pid_t pid, lwpid_t tid, register_t trapno); 372 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %d, tid %d, err %d)", 373 pid_t pid, lwpid_t tid, int err); 374 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %d, tid %d)", 375 pid_t pid, lwpid_t tid); 376 377 /* 378 * Exception, fault, and trap interface to the kernel. 379 * This common code is called from assembly language IDT gate entry 380 * routines that prepare a suitable stack frame, and restore this 381 * frame after the exception has been processed. 382 * 383 * This function is also called from doreti in an interlock to handle ASTs. 384 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap 385 * 386 * NOTE! We have to retrieve the fault address prior to obtaining the 387 * MP lock because get_mplock() may switch out. YYY cr2 really ought 388 * to be retrieved by the assembly code, not here. 389 * 390 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing 391 * if an attempt is made to switch from a fast interrupt or IPI. This is 392 * necessary to properly take fatal kernel traps on SMP machines if 393 * get_mplock() has to block. 394 */ 395 396 void 397 trap(struct trapframe *frame) 398 { 399 struct globaldata *gd = mycpu; 400 struct thread *td = gd->gd_curthread; 401 struct lwp *lp = td->td_lwp; 402 struct proc *p; 403 int sticks = 0; 404 int i = 0, ucode = 0, type, code; 405 #ifdef SMP 406 int have_mplock = 0; 407 #endif 408 #ifdef INVARIANTS 409 int crit_count = td->td_critcount; 410 lwkt_tokref_t curstop = td->td_toks_stop; 411 #endif 412 vm_offset_t eva; 413 414 p = td->td_proc; 415 clear_quickret(); 416 417 #ifdef DDB 418 /* 419 * We need to allow T_DNA faults when the debugger is active since 420 * some dumping paths do large bcopy() which use the floating 421 * point registers for faster copying. 422 */ 423 if (db_active && frame->tf_trapno != T_DNA) { 424 eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0); 425 ++gd->gd_trap_nesting_level; 426 MAKEMPSAFE(have_mplock); 427 trap_fatal(frame, eva); 428 --gd->gd_trap_nesting_level; 429 goto out2; 430 } 431 #endif 432 433 eva = 0; 434 435 if ((frame->tf_rflags & PSL_I) == 0) { 436 /* 437 * Buggy application or kernel code has disabled interrupts 438 * and then trapped. Enabling interrupts now is wrong, but 439 * it is better than running with interrupts disabled until 440 * they are accidentally enabled later. 441 */ 442 type = frame->tf_trapno; 443 if (ISPL(frame->tf_cs) == SEL_UPL) { 444 MAKEMPSAFE(have_mplock); 445 /* JG curproc can be NULL */ 446 kprintf( 447 "pid %ld (%s): trap %d with interrupts disabled\n", 448 (long)curproc->p_pid, curproc->p_comm, type); 449 } else if (type != T_NMI && type != T_BPTFLT && 450 type != T_TRCTRAP) { 451 /* 452 * XXX not quite right, since this may be for a 453 * multiple fault in user mode. 454 */ 455 MAKEMPSAFE(have_mplock); 456 kprintf("kernel trap %d with interrupts disabled\n", 457 type); 458 } 459 cpu_enable_intr(); 460 } 461 462 type = frame->tf_trapno; 463 code = frame->tf_err; 464 465 if (ISPL(frame->tf_cs) == SEL_UPL) { 466 /* user trap */ 467 468 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid, 469 frame->tf_trapno, eva); 470 471 userenter(td, p); 472 473 sticks = (int)td->td_sticks; 474 KASSERT(lp->lwp_md.md_regs == frame, 475 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame)); 476 477 switch (type) { 478 case T_PRIVINFLT: /* privileged instruction fault */ 479 i = SIGILL; 480 ucode = ILL_PRVOPC; 481 break; 482 483 case T_BPTFLT: /* bpt instruction fault */ 484 case T_TRCTRAP: /* trace trap */ 485 frame->tf_rflags &= ~PSL_T; 486 i = SIGTRAP; 487 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT); 488 break; 489 490 case T_ARITHTRAP: /* arithmetic trap */ 491 ucode = code; 492 i = SIGFPE; 493 break; 494 495 case T_ASTFLT: /* Allow process switch */ 496 mycpu->gd_cnt.v_soft++; 497 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) { 498 atomic_clear_int(&mycpu->gd_reqflags, 499 RQF_AST_OWEUPC); 500 addupc_task(p, p->p_prof.pr_addr, 501 p->p_prof.pr_ticks); 502 } 503 goto out; 504 505 case T_PROTFLT: /* general protection fault */ 506 i = SIGBUS; 507 ucode = BUS_OBJERR; 508 break; 509 case T_STKFLT: /* stack fault */ 510 case T_SEGNPFLT: /* segment not present fault */ 511 i = SIGBUS; 512 ucode = BUS_ADRERR; 513 break; 514 case T_TSSFLT: /* invalid TSS fault */ 515 case T_DOUBLEFLT: /* double fault */ 516 default: 517 i = SIGBUS; 518 ucode = BUS_OBJERR; 519 break; 520 521 case T_PAGEFLT: /* page fault */ 522 i = trap_pfault(frame, TRUE); 523 if (frame->tf_rip == 0) { 524 kprintf("T_PAGEFLT: Warning %%rip == 0!\n"); 525 while (freeze_on_seg_fault) { 526 tsleep(p, 0, "freeze", hz * 20); 527 } 528 } 529 if (i == -1 || i == 0) 530 goto out; 531 532 533 if (i == SIGSEGV) 534 ucode = SEGV_MAPERR; 535 else { 536 i = SIGSEGV; 537 ucode = SEGV_ACCERR; 538 } 539 break; 540 541 case T_DIVIDE: /* integer divide fault */ 542 ucode = FPE_INTDIV; 543 i = SIGFPE; 544 break; 545 546 #if NISA > 0 547 case T_NMI: 548 MAKEMPSAFE(have_mplock); 549 /* machine/parity/power fail/"kitchen sink" faults */ 550 if (isa_nmi(code) == 0) { 551 #ifdef DDB 552 /* 553 * NMI can be hooked up to a pushbutton 554 * for debugging. 555 */ 556 if (ddb_on_nmi) { 557 kprintf ("NMI ... going to debugger\n"); 558 kdb_trap(type, 0, frame); 559 } 560 #endif /* DDB */ 561 goto out2; 562 } else if (panic_on_nmi) 563 panic("NMI indicates hardware failure"); 564 break; 565 #endif /* NISA > 0 */ 566 567 case T_OFLOW: /* integer overflow fault */ 568 ucode = FPE_INTOVF; 569 i = SIGFPE; 570 break; 571 572 case T_BOUND: /* bounds check fault */ 573 ucode = FPE_FLTSUB; 574 i = SIGFPE; 575 break; 576 577 case T_DNA: 578 /* 579 * Virtual kernel intercept - pass the DNA exception 580 * to the virtual kernel if it asked to handle it. 581 * This occurs when the virtual kernel is holding 582 * onto the FP context for a different emulated 583 * process then the one currently running. 584 * 585 * We must still call npxdna() since we may have 586 * saved FP state that the virtual kernel needs 587 * to hand over to a different emulated process. 588 */ 589 if (lp->lwp_vkernel && lp->lwp_vkernel->ve && 590 (td->td_pcb->pcb_flags & FP_VIRTFP) 591 ) { 592 npxdna(); 593 break; 594 } 595 596 /* 597 * The kernel may have switched out the FP unit's 598 * state, causing the user process to take a fault 599 * when it tries to use the FP unit. Restore the 600 * state here 601 */ 602 if (npxdna()) 603 goto out; 604 i = SIGFPE; 605 ucode = FPE_FPU_NP_TRAP; 606 break; 607 608 case T_FPOPFLT: /* FPU operand fetch fault */ 609 ucode = ILL_COPROC; 610 i = SIGILL; 611 break; 612 613 case T_XMMFLT: /* SIMD floating-point exception */ 614 ucode = 0; /* XXX */ 615 i = SIGFPE; 616 break; 617 } 618 } else { 619 /* kernel trap */ 620 621 switch (type) { 622 case T_PAGEFLT: /* page fault */ 623 trap_pfault(frame, FALSE); 624 goto out2; 625 626 case T_DNA: 627 /* 628 * The kernel is apparently using fpu for copying. 629 * XXX this should be fatal unless the kernel has 630 * registered such use. 631 */ 632 if (npxdna()) 633 goto out2; 634 break; 635 636 case T_STKFLT: /* stack fault */ 637 break; 638 639 case T_PROTFLT: /* general protection fault */ 640 case T_SEGNPFLT: /* segment not present fault */ 641 /* 642 * Invalid segment selectors and out of bounds 643 * %rip's and %rsp's can be set up in user mode. 644 * This causes a fault in kernel mode when the 645 * kernel tries to return to user mode. We want 646 * to get this fault so that we can fix the 647 * problem here and not have to check all the 648 * selectors and pointers when the user changes 649 * them. 650 */ 651 if (mycpu->gd_intr_nesting_level == 0) { 652 if (td->td_pcb->pcb_onfault) { 653 frame->tf_rip = (register_t) 654 td->td_pcb->pcb_onfault; 655 goto out2; 656 } 657 if (frame->tf_rip == (long)doreti_iret) { 658 frame->tf_rip = (long)doreti_iret_fault; 659 goto out2; 660 } 661 } 662 break; 663 664 case T_TSSFLT: 665 /* 666 * PSL_NT can be set in user mode and isn't cleared 667 * automatically when the kernel is entered. This 668 * causes a TSS fault when the kernel attempts to 669 * `iret' because the TSS link is uninitialized. We 670 * want to get this fault so that we can fix the 671 * problem here and not every time the kernel is 672 * entered. 673 */ 674 if (frame->tf_rflags & PSL_NT) { 675 frame->tf_rflags &= ~PSL_NT; 676 goto out2; 677 } 678 break; 679 680 case T_TRCTRAP: /* trace trap */ 681 #if 0 682 if (frame->tf_rip == (int)IDTVEC(syscall)) { 683 /* 684 * We've just entered system mode via the 685 * syscall lcall. Continue single stepping 686 * silently until the syscall handler has 687 * saved the flags. 688 */ 689 goto out2; 690 } 691 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) { 692 /* 693 * The syscall handler has now saved the 694 * flags. Stop single stepping it. 695 */ 696 frame->tf_rflags &= ~PSL_T; 697 goto out2; 698 } 699 #endif 700 701 /* 702 * Ignore debug register trace traps due to 703 * accesses in the user's address space, which 704 * can happen under several conditions such as 705 * if a user sets a watchpoint on a buffer and 706 * then passes that buffer to a system call. 707 * We still want to get TRCTRAPS for addresses 708 * in kernel space because that is useful when 709 * debugging the kernel. 710 */ 711 #if JG 712 if (user_dbreg_trap()) { 713 /* 714 * Reset breakpoint bits because the 715 * processor doesn't 716 */ 717 /* XXX check upper bits here */ 718 load_dr6(rdr6() & 0xfffffff0); 719 goto out2; 720 } 721 #endif 722 /* 723 * FALLTHROUGH (TRCTRAP kernel mode, kernel address) 724 */ 725 case T_BPTFLT: 726 /* 727 * If DDB is enabled, let it handle the debugger trap. 728 * Otherwise, debugger traps "can't happen". 729 */ 730 ucode = TRAP_BRKPT; 731 #ifdef DDB 732 MAKEMPSAFE(have_mplock); 733 if (kdb_trap(type, 0, frame)) 734 goto out2; 735 #endif 736 break; 737 738 #if NISA > 0 739 case T_NMI: 740 MAKEMPSAFE(have_mplock); 741 /* machine/parity/power fail/"kitchen sink" faults */ 742 if (isa_nmi(code) == 0) { 743 #ifdef DDB 744 /* 745 * NMI can be hooked up to a pushbutton 746 * for debugging. 747 */ 748 if (ddb_on_nmi) { 749 kprintf ("NMI ... going to debugger\n"); 750 kdb_trap(type, 0, frame); 751 } 752 #endif /* DDB */ 753 goto out2; 754 } else if (panic_on_nmi == 0) 755 goto out2; 756 /* FALL THROUGH */ 757 #endif /* NISA > 0 */ 758 } 759 MAKEMPSAFE(have_mplock); 760 trap_fatal(frame, 0); 761 goto out2; 762 } 763 764 /* 765 * Virtual kernel intercept - if the fault is directly related to a 766 * VM context managed by a virtual kernel then let the virtual kernel 767 * handle it. 768 */ 769 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 770 vkernel_trap(lp, frame); 771 goto out; 772 } 773 774 /* Translate fault for emulators (e.g. Linux) */ 775 if (*p->p_sysent->sv_transtrap) 776 i = (*p->p_sysent->sv_transtrap)(i, type); 777 778 MAKEMPSAFE(have_mplock); 779 trapsignal(lp, i, ucode); 780 781 #ifdef DEBUG 782 if (type <= MAX_TRAP_MSG) { 783 uprintf("fatal process exception: %s", 784 trap_msg[type]); 785 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 786 uprintf(", fault VA = 0x%lx", frame->tf_addr); 787 uprintf("\n"); 788 } 789 #endif 790 791 out: 792 userret(lp, frame, sticks); 793 userexit(lp); 794 out2: ; 795 #ifdef SMP 796 if (have_mplock) 797 rel_mplock(); 798 #endif 799 if (p != NULL && lp != NULL) 800 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid); 801 #ifdef INVARIANTS 802 KASSERT(crit_count == td->td_critcount, 803 ("trap: critical section count mismatch! %d/%d", 804 crit_count, td->td_pri)); 805 KASSERT(curstop == td->td_toks_stop, 806 ("trap: extra tokens held after trap! %ld/%ld", 807 curstop - &td->td_toks_base, 808 td->td_toks_stop - &td->td_toks_base)); 809 #endif 810 } 811 812 static int 813 trap_pfault(struct trapframe *frame, int usermode) 814 { 815 vm_offset_t va; 816 struct vmspace *vm = NULL; 817 vm_map_t map; 818 int rv = 0; 819 int fault_flags; 820 vm_prot_t ftype; 821 thread_t td = curthread; 822 struct lwp *lp = td->td_lwp; 823 struct proc *p; 824 825 va = trunc_page(frame->tf_addr); 826 if (va >= VM_MIN_KERNEL_ADDRESS) { 827 /* 828 * Don't allow user-mode faults in kernel address space. 829 */ 830 if (usermode) { 831 fault_flags = -1; 832 ftype = -1; 833 goto nogo; 834 } 835 836 map = &kernel_map; 837 } else { 838 /* 839 * This is a fault on non-kernel virtual memory. 840 * vm is initialized above to NULL. If curproc is NULL 841 * or curproc->p_vmspace is NULL the fault is fatal. 842 */ 843 if (lp != NULL) 844 vm = lp->lwp_vmspace; 845 846 if (vm == NULL) { 847 fault_flags = -1; 848 ftype = -1; 849 goto nogo; 850 } 851 852 /* 853 * Debugging, try to catch kernel faults on the user address space when not inside 854 * on onfault (e.g. copyin/copyout) routine. 855 */ 856 if (usermode == 0 && (td->td_pcb == NULL || td->td_pcb->pcb_onfault == NULL)) { 857 if (freeze_on_seg_fault) { 858 kprintf("trap_pfault: user address fault from kernel mode " 859 "%016lx\n", (long)frame->tf_addr); 860 while (freeze_on_seg_fault) { 861 tsleep(&freeze_on_seg_fault, 0, "frzseg", hz * 20); 862 } 863 } 864 } 865 map = &vm->vm_map; 866 } 867 868 /* 869 * PGEX_I is defined only if the execute disable bit capability is 870 * supported and enabled. 871 */ 872 if (frame->tf_err & PGEX_W) 873 ftype = VM_PROT_WRITE; 874 #if JG 875 else if ((frame->tf_err & PGEX_I) && pg_nx != 0) 876 ftype = VM_PROT_EXECUTE; 877 #endif 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; 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 if (td->td_gd->gd_intr_nesting_level == 0 && 914 td->td_pcb->pcb_onfault) { 915 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault; 916 return (0); 917 } 918 trap_fatal(frame, frame->tf_addr); 919 return (-1); 920 } 921 922 /* 923 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no 924 * kludge is needed to pass the fault address to signal handlers. 925 */ 926 p = td->td_proc; 927 if (td->td_lwp->lwp_vkernel == NULL) { 928 if (bootverbose || freeze_on_seg_fault || ddb_on_seg_fault) { 929 kprintf("seg-fault ft=%04x ff=%04x addr=%p rip=%p " 930 "pid=%d cpu=%d p_comm=%s\n", 931 ftype, fault_flags, 932 (void *)frame->tf_addr, 933 (void *)frame->tf_rip, 934 p->p_pid, mycpu->gd_cpuid, p->p_comm); 935 } 936 #ifdef DDB 937 while (freeze_on_seg_fault) { 938 tsleep(p, 0, "freeze", hz * 20); 939 } 940 if (ddb_on_seg_fault) 941 Debugger("ddb_on_seg_fault"); 942 #endif 943 } 944 945 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 946 } 947 948 static void 949 trap_fatal(struct trapframe *frame, vm_offset_t eva) 950 { 951 int code, ss; 952 u_int type; 953 long rsp; 954 struct soft_segment_descriptor softseg; 955 char *msg; 956 957 code = frame->tf_err; 958 type = frame->tf_trapno; 959 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg); 960 961 if (type <= MAX_TRAP_MSG) 962 msg = trap_msg[type]; 963 else 964 msg = "UNKNOWN"; 965 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg, 966 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 967 #ifdef SMP 968 /* three separate prints in case of a trap on an unmapped page */ 969 kprintf("cpuid = %d; ", mycpu->gd_cpuid); 970 kprintf("lapic->id = %08x\n", lapic->id); 971 #endif 972 if (type == T_PAGEFLT) { 973 kprintf("fault virtual address = 0x%lx\n", eva); 974 kprintf("fault code = %s %s %s, %s\n", 975 code & PGEX_U ? "user" : "supervisor", 976 code & PGEX_W ? "write" : "read", 977 code & PGEX_I ? "instruction" : "data", 978 code & PGEX_P ? "protection violation" : "page not present"); 979 } 980 kprintf("instruction pointer = 0x%lx:0x%lx\n", 981 frame->tf_cs & 0xffff, frame->tf_rip); 982 if (ISPL(frame->tf_cs) == SEL_UPL) { 983 ss = frame->tf_ss & 0xffff; 984 rsp = frame->tf_rsp; 985 } else { 986 ss = GSEL(GDATA_SEL, SEL_KPL); 987 rsp = (long)&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 #ifdef SMP 1067 /* three separate prints in case of a trap on an unmapped page */ 1068 kprintf("cpuid = %d; ", mycpu->gd_cpuid); 1069 kprintf("lapic->id = %08x\n", lapic->id); 1070 #endif 1071 panic("double fault"); 1072 } 1073 1074 /* 1075 * syscall2 - MP aware system call request C handler 1076 * 1077 * A system call is essentially treated as a trap except that the 1078 * MP lock is not held on entry or return. We are responsible for 1079 * obtaining the MP lock if necessary and for handling ASTs 1080 * (e.g. a task switch) prior to return. 1081 * 1082 * MPSAFE 1083 */ 1084 void 1085 syscall2(struct trapframe *frame) 1086 { 1087 struct thread *td = curthread; 1088 struct proc *p = td->td_proc; 1089 struct lwp *lp = td->td_lwp; 1090 caddr_t params; 1091 struct sysent *callp; 1092 register_t orig_tf_rflags; 1093 int sticks; 1094 int error; 1095 int narg; 1096 #ifdef INVARIANTS 1097 int crit_count = td->td_critcount; 1098 #endif 1099 #ifdef SMP 1100 int have_mplock = 0; 1101 #endif 1102 register_t *argp; 1103 u_int code; 1104 int reg, regcnt; 1105 union sysunion args; 1106 register_t *argsdst; 1107 1108 mycpu->gd_cnt.v_syscall++; 1109 1110 #ifdef DIAGNOSTIC 1111 if (ISPL(frame->tf_cs) != SEL_UPL) { 1112 get_mplock(); 1113 panic("syscall"); 1114 /* NOT REACHED */ 1115 } 1116 #endif 1117 1118 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid, 1119 frame->tf_rax); 1120 1121 userenter(td, p); /* lazy raise our priority */ 1122 1123 reg = 0; 1124 regcnt = 6; 1125 /* 1126 * Misc 1127 */ 1128 sticks = (int)td->td_sticks; 1129 orig_tf_rflags = frame->tf_rflags; 1130 1131 /* 1132 * Virtual kernel intercept - if a VM context managed by a virtual 1133 * kernel issues a system call the virtual kernel handles it, not us. 1134 * Restore the virtual kernel context and return from its system 1135 * call. The current frame is copied out to the virtual kernel. 1136 */ 1137 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 1138 vkernel_trap(lp, frame); 1139 error = EJUSTRETURN; 1140 goto out; 1141 } 1142 1143 /* 1144 * Get the system call parameters and account for time 1145 */ 1146 KASSERT(lp->lwp_md.md_regs == frame, 1147 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame)); 1148 params = (caddr_t)frame->tf_rsp + sizeof(register_t); 1149 code = frame->tf_rax; 1150 1151 if (p->p_sysent->sv_prepsyscall) { 1152 (*p->p_sysent->sv_prepsyscall)( 1153 frame, (int *)(&args.nosys.sysmsg + 1), 1154 &code, ¶ms); 1155 } else { 1156 if (code == SYS_syscall || code == SYS___syscall) { 1157 code = frame->tf_rdi; 1158 reg++; 1159 regcnt--; 1160 } 1161 } 1162 1163 if (p->p_sysent->sv_mask) 1164 code &= p->p_sysent->sv_mask; 1165 1166 if (code >= p->p_sysent->sv_size) 1167 callp = &p->p_sysent->sv_table[0]; 1168 else 1169 callp = &p->p_sysent->sv_table[code]; 1170 1171 narg = callp->sy_narg & SYF_ARGMASK; 1172 1173 /* 1174 * On x86_64 we get up to six arguments in registers. The rest are 1175 * on the stack. The first six members of 'struct trapframe' happen 1176 * to be the registers used to pass arguments, in exactly the right 1177 * order. 1178 */ 1179 argp = &frame->tf_rdi; 1180 argp += reg; 1181 argsdst = (register_t *)(&args.nosys.sysmsg + 1); 1182 /* 1183 * JG can we overflow the space pointed to by 'argsdst' 1184 * either with 'bcopy' or with 'copyin'? 1185 */ 1186 bcopy(argp, argsdst, sizeof(register_t) * regcnt); 1187 /* 1188 * copyin is MP aware, but the tracing code is not 1189 */ 1190 if (narg > regcnt) { 1191 KASSERT(params != NULL, ("copyin args with no params!")); 1192 error = copyin(params, &argsdst[regcnt], 1193 (narg - regcnt) * sizeof(register_t)); 1194 if (error) { 1195 #ifdef KTRACE 1196 if (KTRPOINT(td, KTR_SYSCALL)) { 1197 MAKEMPSAFE(have_mplock); 1198 1199 ktrsyscall(lp, code, narg, 1200 (void *)(&args.nosys.sysmsg + 1)); 1201 } 1202 #endif 1203 goto bad; 1204 } 1205 } 1206 1207 #ifdef KTRACE 1208 if (KTRPOINT(td, KTR_SYSCALL)) { 1209 MAKEMPSAFE(have_mplock); 1210 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1)); 1211 } 1212 #endif 1213 1214 /* 1215 * Default return value is 0 (will be copied to %rax). Double-value 1216 * returns use %rax and %rdx. %rdx is left unchanged for system 1217 * calls which return only one result. 1218 */ 1219 args.sysmsg_fds[0] = 0; 1220 args.sysmsg_fds[1] = frame->tf_rdx; 1221 1222 /* 1223 * The syscall might manipulate the trap frame. If it does it 1224 * will probably return EJUSTRETURN. 1225 */ 1226 args.sysmsg_frame = frame; 1227 1228 STOPEVENT(p, S_SCE, narg); /* MP aware */ 1229 1230 /* 1231 * NOTE: All system calls run MPSAFE now. The system call itself 1232 * is responsible for getting the MP lock. 1233 */ 1234 #ifdef SYSCALL_DEBUG 1235 uint64_t tscval = rdtsc(); 1236 #endif 1237 error = (*callp->sy_call)(&args); 1238 #ifdef SYSCALL_DEBUG 1239 tscval = rdtsc() - tscval; 1240 tscval = tscval * 1000000 / tsc_frequency; 1241 if (SysCallsWorstCase[code] < tscval) 1242 SysCallsWorstCase[code] = tscval; 1243 #endif 1244 1245 out: 1246 /* 1247 * MP SAFE (we may or may not have the MP lock at this point) 1248 */ 1249 //kprintf("SYSMSG %d ", error); 1250 switch (error) { 1251 case 0: 1252 /* 1253 * Reinitialize proc pointer `p' as it may be different 1254 * if this is a child returning from fork syscall. 1255 */ 1256 p = curproc; 1257 lp = curthread->td_lwp; 1258 frame->tf_rax = args.sysmsg_fds[0]; 1259 frame->tf_rdx = args.sysmsg_fds[1]; 1260 frame->tf_rflags &= ~PSL_C; 1261 break; 1262 case ERESTART: 1263 /* 1264 * Reconstruct pc, we know that 'syscall' is 2 bytes. 1265 * We have to do a full context restore so that %r10 1266 * (which was holding the value of %rcx) is restored for 1267 * the next iteration. 1268 */ 1269 if (frame->tf_err != 0 && frame->tf_err != 2) 1270 kprintf("lp %s:%d frame->tf_err is weird %ld\n", 1271 td->td_comm, lp->lwp_proc->p_pid, frame->tf_err); 1272 frame->tf_rip -= frame->tf_err; 1273 frame->tf_r10 = frame->tf_rcx; 1274 break; 1275 case EJUSTRETURN: 1276 break; 1277 case EASYNC: 1278 panic("Unexpected EASYNC return value (for now)"); 1279 default: 1280 bad: 1281 if (p->p_sysent->sv_errsize) { 1282 if (error >= p->p_sysent->sv_errsize) 1283 error = -1; /* XXX */ 1284 else 1285 error = p->p_sysent->sv_errtbl[error]; 1286 } 1287 frame->tf_rax = error; 1288 frame->tf_rflags |= PSL_C; 1289 break; 1290 } 1291 1292 /* 1293 * Traced syscall. trapsignal() is not MP aware. 1294 */ 1295 if (orig_tf_rflags & PSL_T) { 1296 MAKEMPSAFE(have_mplock); 1297 frame->tf_rflags &= ~PSL_T; 1298 trapsignal(lp, SIGTRAP, TRAP_TRACE); 1299 } 1300 1301 /* 1302 * Handle reschedule and other end-of-syscall issues 1303 */ 1304 userret(lp, frame, sticks); 1305 1306 #ifdef KTRACE 1307 if (KTRPOINT(td, KTR_SYSRET)) { 1308 MAKEMPSAFE(have_mplock); 1309 ktrsysret(lp, code, error, args.sysmsg_result); 1310 } 1311 #endif 1312 1313 /* 1314 * This works because errno is findable through the 1315 * register set. If we ever support an emulation where this 1316 * is not the case, this code will need to be revisited. 1317 */ 1318 STOPEVENT(p, S_SCX, code); 1319 1320 userexit(lp); 1321 #ifdef SMP 1322 /* 1323 * Release the MP lock if we had to get it 1324 */ 1325 if (have_mplock) 1326 rel_mplock(); 1327 #endif 1328 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error); 1329 #ifdef INVARIANTS 1330 KASSERT(crit_count == td->td_critcount, 1331 ("syscall: critical section count mismatch! %d/%d", 1332 crit_count, td->td_pri)); 1333 KASSERT(&td->td_toks_base == td->td_toks_stop, 1334 ("syscall: extra tokens held after trap! %ld", 1335 td->td_toks_stop - &td->td_toks_base)); 1336 #endif 1337 } 1338 1339 /* 1340 * NOTE: mplock not held at any point 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 * NOTE: The mplock is not held at any point. 1361 */ 1362 void 1363 generic_lwp_return(struct lwp *lp, struct trapframe *frame) 1364 { 1365 struct proc *p = lp->lwp_proc; 1366 1367 /* 1368 * Newly forked processes are given a kernel priority. We have to 1369 * adjust the priority to a normal user priority and fake entry 1370 * into the kernel (call userenter()) to install a passive release 1371 * function just in case userret() decides to stop the process. This 1372 * can occur when ^Z races a fork. If we do not install the passive 1373 * release function the current process designation will not be 1374 * released when the thread goes to sleep. 1375 */ 1376 lwkt_setpri_self(TDPRI_USER_NORM); 1377 userenter(lp->lwp_thread, p); 1378 userret(lp, frame, 0); 1379 #ifdef KTRACE 1380 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET)) 1381 ktrsysret(lp, SYS_fork, 0, 0); 1382 #endif 1383 lp->lwp_flags |= LWP_PASSIVE_ACQ; 1384 userexit(lp); 1385 lp->lwp_flags &= ~LWP_PASSIVE_ACQ; 1386 } 1387 1388 /* 1389 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA 1390 * fault (which is then passed back to the virtual kernel) if an attempt is 1391 * made to use the FP unit. 1392 * 1393 * XXX this is a fairly big hack. 1394 */ 1395 void 1396 set_vkernel_fp(struct trapframe *frame) 1397 { 1398 struct thread *td = curthread; 1399 1400 if (frame->tf_xflags & PGEX_FPFAULT) { 1401 td->td_pcb->pcb_flags |= FP_VIRTFP; 1402 if (mdcpu->gd_npxthread == td) 1403 npxexit(); 1404 } else { 1405 td->td_pcb->pcb_flags &= ~FP_VIRTFP; 1406 } 1407 } 1408 1409 /* 1410 * Called from vkernel_trap() to fixup the vkernel's syscall 1411 * frame for vmspace_ctl() return. 1412 */ 1413 void 1414 cpu_vkernel_trap(struct trapframe *frame, int error) 1415 { 1416 frame->tf_rax = error; 1417 if (error) 1418 frame->tf_rflags |= PSL_C; 1419 else 1420 frame->tf_rflags &= ~PSL_C; 1421 } 1422