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/mplock2.h> 91 #include <sys/spinlock2.h> 92 93 #define MAKEMPSAFE(have_mplock) \ 94 if (have_mplock == 0) { \ 95 get_mplock(); \ 96 have_mplock = 1; \ 97 } 98 99 extern void trap(struct trapframe *frame); 100 101 static int trap_pfault(struct trapframe *, int); 102 static void trap_fatal(struct trapframe *, vm_offset_t); 103 void dblfault_handler(struct trapframe *frame); 104 105 #define MAX_TRAP_MSG 30 106 static char *trap_msg[] = { 107 "", /* 0 unused */ 108 "privileged instruction fault", /* 1 T_PRIVINFLT */ 109 "", /* 2 unused */ 110 "breakpoint instruction fault", /* 3 T_BPTFLT */ 111 "", /* 4 unused */ 112 "", /* 5 unused */ 113 "arithmetic trap", /* 6 T_ARITHTRAP */ 114 "system forced exception", /* 7 T_ASTFLT */ 115 "", /* 8 unused */ 116 "general protection fault", /* 9 T_PROTFLT */ 117 "trace trap", /* 10 T_TRCTRAP */ 118 "", /* 11 unused */ 119 "page fault", /* 12 T_PAGEFLT */ 120 "", /* 13 unused */ 121 "alignment fault", /* 14 T_ALIGNFLT */ 122 "", /* 15 unused */ 123 "", /* 16 unused */ 124 "", /* 17 unused */ 125 "integer divide fault", /* 18 T_DIVIDE */ 126 "non-maskable interrupt trap", /* 19 T_NMI */ 127 "overflow trap", /* 20 T_OFLOW */ 128 "FPU bounds check fault", /* 21 T_BOUND */ 129 "FPU device not available", /* 22 T_DNA */ 130 "double fault", /* 23 T_DOUBLEFLT */ 131 "FPU operand fetch fault", /* 24 T_FPOPFLT */ 132 "invalid TSS fault", /* 25 T_TSSFLT */ 133 "segment not present fault", /* 26 T_SEGNPFLT */ 134 "stack fault", /* 27 T_STKFLT */ 135 "machine check trap", /* 28 T_MCHK */ 136 "SIMD floating-point exception", /* 29 T_XMMFLT */ 137 "reserved (unknown) fault", /* 30 T_RESERVED */ 138 }; 139 140 #ifdef DDB 141 static int ddb_on_nmi = 1; 142 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW, 143 &ddb_on_nmi, 0, "Go to DDB on NMI"); 144 static int ddb_on_seg_fault = 0; 145 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_seg_fault, CTLFLAG_RW, 146 &ddb_on_seg_fault, 0, "Go to DDB on user seg-fault"); 147 static int freeze_on_seg_fault = 0; 148 SYSCTL_INT(_machdep, OID_AUTO, freeze_on_seg_fault, CTLFLAG_RW, 149 &freeze_on_seg_fault, 0, "Go to DDB on user seg-fault"); 150 #endif 151 static int panic_on_nmi = 1; 152 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW, 153 &panic_on_nmi, 0, "Panic on NMI"); 154 static int fast_release; 155 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW, 156 &fast_release, 0, "Passive Release was optimal"); 157 static int slow_release; 158 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW, 159 &slow_release, 0, "Passive Release was nonoptimal"); 160 161 /* 162 * System call debugging records the worst-case system call 163 * overhead (inclusive of blocking), but may be inaccurate. 164 */ 165 /*#define SYSCALL_DEBUG*/ 166 #ifdef SYSCALL_DEBUG 167 uint64_t SysCallsWorstCase[SYS_MAXSYSCALL]; 168 #endif 169 170 /* 171 * Passively intercepts the thread switch function to increase 172 * the thread priority from a user priority to a kernel priority, reducing 173 * syscall and trap overhead for the case where no switch occurs. 174 * 175 * Synchronizes td_ucred with p_ucred. This is used by system calls, 176 * signal handling, faults, AST traps, and anything else that enters the 177 * kernel from userland and provides the kernel with a stable read-only 178 * copy of the process ucred. 179 * 180 * To avoid races with another thread updating p_ucred we obtain p_spin. 181 * The other thread doing the update will obtain both p_token and p_spin. 182 * In the case where the cached cred pointer matches, we will already have 183 * the ref and we don't have to do one blessed thing. 184 */ 185 static __inline void 186 userenter(struct thread *curtd, struct proc *curp) 187 { 188 struct ucred *ocred; 189 struct ucred *ncred; 190 191 curtd->td_release = lwkt_passive_release; 192 193 if (curtd->td_ucred != curp->p_ucred) { 194 spin_lock(&curp->p_spin); 195 ncred = crhold(curp->p_ucred); 196 spin_unlock(&curp->p_spin); 197 ocred = curtd->td_ucred; 198 curtd->td_ucred = ncred; 199 if (ocred) 200 crfree(ocred); 201 } 202 203 #ifdef DDB 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 #endif 213 } 214 215 /* 216 * Handle signals, upcalls, profiling, and other AST's and/or tasks that 217 * must be completed before we can return to or try to return to userland. 218 * 219 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64 220 * arithmatic on the delta calculation so the absolute tick values are 221 * truncated to an integer. 222 */ 223 static void 224 userret(struct lwp *lp, struct trapframe *frame, int sticks) 225 { 226 struct proc *p = lp->lwp_proc; 227 int sig; 228 229 /* 230 * Charge system time if profiling. Note: times are in microseconds. 231 * This may do a copyout and block, so do it first even though it 232 * means some system time will be charged as user time. 233 */ 234 if (p->p_flags & P_PROFIL) { 235 addupc_task(p, frame->tf_rip, 236 (u_int)((int)lp->lwp_thread->td_sticks - sticks)); 237 } 238 239 recheck: 240 /* 241 * Specific on-return-to-usermode checks (LWP_MP_WEXIT, 242 * LWP_MP_VNLRU, etc). 243 */ 244 if (lp->lwp_mpflags & LWP_MP_URETMASK) 245 lwpuserret(lp); 246 247 /* 248 * Block here if we are in a stopped state. 249 */ 250 if (p->p_stat == SSTOP || p->p_stat == SCORE || dump_stop_usertds) { 251 lwkt_gettoken(&p->p_token); 252 tstop(); 253 lwkt_reltoken(&p->p_token); 254 goto recheck; 255 } 256 257 /* 258 * Post any pending upcalls. If running a virtual kernel be sure 259 * to restore the virtual kernel's vmspace before posting the upcall. 260 */ 261 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF)) { 262 lwkt_gettoken(&p->p_token); 263 if (p->p_flags & P_SIGVTALRM) { 264 p->p_flags &= ~P_SIGVTALRM; 265 ksignal(p, SIGVTALRM); 266 } 267 if (p->p_flags & P_SIGPROF) { 268 p->p_flags &= ~P_SIGPROF; 269 ksignal(p, SIGPROF); 270 } 271 lwkt_reltoken(&p->p_token); 272 goto recheck; 273 } 274 275 /* 276 * Post any pending signals. If running a virtual kernel be sure 277 * to restore the virtual kernel's vmspace before posting the signal. 278 * 279 * WARNING! postsig() can exit and not return. 280 */ 281 if ((sig = CURSIG_TRACE(lp)) != 0) { 282 lwkt_gettoken(&p->p_token); 283 postsig(sig); 284 lwkt_reltoken(&p->p_token); 285 goto recheck; 286 } 287 288 /* 289 * block here if we are swapped out, but still process signals 290 * (such as SIGKILL). proc0 (the swapin scheduler) is already 291 * aware of our situation, we do not have to wake it up. 292 */ 293 if (p->p_flags & P_SWAPPEDOUT) { 294 lwkt_gettoken(&p->p_token); 295 get_mplock(); 296 p->p_flags |= P_SWAPWAIT; 297 swapin_request(); 298 if (p->p_flags & P_SWAPWAIT) 299 tsleep(p, PCATCH, "SWOUT", 0); 300 p->p_flags &= ~P_SWAPWAIT; 301 rel_mplock(); 302 lwkt_reltoken(&p->p_token); 303 goto recheck; 304 } 305 306 /* 307 * In a multi-threaded program it is possible for a thread to change 308 * signal state during a system call which temporarily changes the 309 * signal mask. In this case postsig() might not be run and we 310 * have to restore the mask ourselves. 311 */ 312 if (lp->lwp_flags & LWP_OLDMASK) { 313 lp->lwp_flags &= ~LWP_OLDMASK; 314 lp->lwp_sigmask = lp->lwp_oldsigmask; 315 goto recheck; 316 } 317 } 318 319 /* 320 * Cleanup from userenter and any passive release that might have occured. 321 * We must reclaim the current-process designation before we can return 322 * to usermode. We also handle both LWKT and USER reschedule requests. 323 */ 324 static __inline void 325 userexit(struct lwp *lp) 326 { 327 struct thread *td = lp->lwp_thread; 328 /* globaldata_t gd = td->td_gd; */ 329 330 /* 331 * Handle stop requests at kernel priority. Any requests queued 332 * after this loop will generate another AST. 333 */ 334 while (lp->lwp_proc->p_stat == SSTOP || 335 lp->lwp_proc->p_stat == SCORE) { 336 lwkt_gettoken(&lp->lwp_proc->p_token); 337 tstop(); 338 lwkt_reltoken(&lp->lwp_proc->p_token); 339 } 340 341 /* 342 * Reduce our priority in preparation for a return to userland. If 343 * our passive release function was still in place, our priority was 344 * never raised and does not need to be reduced. 345 */ 346 lwkt_passive_recover(td); 347 348 /* WARNING: we may have migrated cpu's */ 349 /* gd = td->td_gd; */ 350 351 /* 352 * Become the current user scheduled process if we aren't already, 353 * and deal with reschedule requests and other factors. 354 */ 355 lp->lwp_proc->p_usched->acquire_curproc(lp); 356 } 357 358 #if !defined(KTR_KERNENTRY) 359 #define KTR_KERNENTRY KTR_ALL 360 #endif 361 KTR_INFO_MASTER(kernentry); 362 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, 363 "TRAP(pid %d, tid %d, trapno %ld, eva %lu)", 364 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva); 365 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %d, tid %d)", 366 pid_t pid, lwpid_t tid); 367 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %d, tid %d, nr %ld)", 368 pid_t pid, lwpid_t tid, register_t trapno); 369 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %d, tid %d, err %d)", 370 pid_t pid, lwpid_t tid, int err); 371 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %d, tid %d)", 372 pid_t pid, lwpid_t tid); 373 374 /* 375 * Exception, fault, and trap interface to the kernel. 376 * This common code is called from assembly language IDT gate entry 377 * routines that prepare a suitable stack frame, and restore this 378 * frame after the exception has been processed. 379 * 380 * This function is also called from doreti in an interlock to handle ASTs. 381 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap 382 * 383 * NOTE! We have to retrieve the fault address prior to obtaining the 384 * MP lock because get_mplock() may switch out. YYY cr2 really ought 385 * to be retrieved by the assembly code, not here. 386 * 387 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing 388 * if an attempt is made to switch from a fast interrupt or IPI. This is 389 * necessary to properly take fatal kernel traps on SMP machines if 390 * get_mplock() has to block. 391 */ 392 393 void 394 trap(struct trapframe *frame) 395 { 396 struct globaldata *gd = mycpu; 397 struct thread *td = gd->gd_curthread; 398 struct lwp *lp = td->td_lwp; 399 struct proc *p; 400 int sticks = 0; 401 int i = 0, ucode = 0, type, code; 402 int have_mplock = 0; 403 #ifdef INVARIANTS 404 int crit_count = td->td_critcount; 405 lwkt_tokref_t curstop = td->td_toks_stop; 406 #endif 407 vm_offset_t eva; 408 409 p = td->td_proc; 410 clear_quickret(); 411 412 #ifdef DDB 413 /* 414 * We need to allow T_DNA faults when the debugger is active since 415 * some dumping paths do large bcopy() which use the floating 416 * point registers for faster copying. 417 */ 418 if (db_active && frame->tf_trapno != T_DNA) { 419 eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0); 420 ++gd->gd_trap_nesting_level; 421 MAKEMPSAFE(have_mplock); 422 trap_fatal(frame, eva); 423 --gd->gd_trap_nesting_level; 424 goto out2; 425 } 426 #endif 427 428 eva = 0; 429 430 if ((frame->tf_rflags & PSL_I) == 0) { 431 /* 432 * Buggy application or kernel code has disabled interrupts 433 * and then trapped. Enabling interrupts now is wrong, but 434 * it is better than running with interrupts disabled until 435 * they are accidentally enabled later. 436 */ 437 type = frame->tf_trapno; 438 if (ISPL(frame->tf_cs) == SEL_UPL) { 439 MAKEMPSAFE(have_mplock); 440 /* JG curproc can be NULL */ 441 kprintf( 442 "pid %ld (%s): trap %d with interrupts disabled\n", 443 (long)curproc->p_pid, curproc->p_comm, type); 444 } else if (type != T_NMI && type != T_BPTFLT && 445 type != T_TRCTRAP) { 446 /* 447 * XXX not quite right, since this may be for a 448 * multiple fault in user mode. 449 */ 450 MAKEMPSAFE(have_mplock); 451 kprintf("kernel trap %d (%s @ 0x%016jx) with " 452 "interrupts disabled\n", 453 type, 454 td->td_comm, 455 frame->tf_rip); 456 } 457 cpu_enable_intr(); 458 } 459 460 type = frame->tf_trapno; 461 code = frame->tf_err; 462 463 if (ISPL(frame->tf_cs) == SEL_UPL) { 464 /* user trap */ 465 466 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid, 467 frame->tf_trapno, eva); 468 469 userenter(td, p); 470 471 sticks = (int)td->td_sticks; 472 KASSERT(lp->lwp_md.md_regs == frame, 473 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame)); 474 475 switch (type) { 476 case T_PRIVINFLT: /* privileged instruction fault */ 477 i = SIGILL; 478 ucode = ILL_PRVOPC; 479 break; 480 481 case T_BPTFLT: /* bpt instruction fault */ 482 case T_TRCTRAP: /* trace trap */ 483 frame->tf_rflags &= ~PSL_T; 484 i = SIGTRAP; 485 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT); 486 break; 487 488 case T_ARITHTRAP: /* arithmetic trap */ 489 ucode = code; 490 i = SIGFPE; 491 break; 492 493 case T_ASTFLT: /* Allow process switch */ 494 mycpu->gd_cnt.v_soft++; 495 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) { 496 atomic_clear_int(&mycpu->gd_reqflags, 497 RQF_AST_OWEUPC); 498 addupc_task(p, p->p_prof.pr_addr, 499 p->p_prof.pr_ticks); 500 } 501 goto out; 502 503 case T_PROTFLT: /* general protection fault */ 504 i = SIGBUS; 505 ucode = BUS_OBJERR; 506 break; 507 case T_STKFLT: /* stack fault */ 508 case T_SEGNPFLT: /* segment not present fault */ 509 i = SIGBUS; 510 ucode = BUS_ADRERR; 511 break; 512 case T_TSSFLT: /* invalid TSS fault */ 513 case T_DOUBLEFLT: /* double fault */ 514 default: 515 i = SIGBUS; 516 ucode = BUS_OBJERR; 517 break; 518 519 case T_PAGEFLT: /* page fault */ 520 i = trap_pfault(frame, TRUE); 521 if (frame->tf_rip == 0) { 522 #ifdef DDB 523 /* used for kernel debugging only */ 524 while (freeze_on_seg_fault) 525 tsleep(p, 0, "freeze", hz * 20); 526 #endif 527 } 528 if (i == -1 || i == 0) 529 goto out; 530 531 532 if (i == SIGSEGV) 533 ucode = SEGV_MAPERR; 534 else { 535 i = SIGSEGV; 536 ucode = SEGV_ACCERR; 537 } 538 break; 539 540 case T_DIVIDE: /* integer divide fault */ 541 ucode = FPE_INTDIV; 542 i = SIGFPE; 543 break; 544 545 #if NISA > 0 546 case T_NMI: 547 MAKEMPSAFE(have_mplock); 548 /* machine/parity/power fail/"kitchen sink" faults */ 549 if (isa_nmi(code) == 0) { 550 #ifdef DDB 551 /* 552 * NMI can be hooked up to a pushbutton 553 * for debugging. 554 */ 555 if (ddb_on_nmi) { 556 kprintf ("NMI ... going to debugger\n"); 557 kdb_trap(type, 0, frame); 558 } 559 #endif /* DDB */ 560 goto out2; 561 } else if (panic_on_nmi) 562 panic("NMI indicates hardware failure"); 563 break; 564 #endif /* NISA > 0 */ 565 566 case T_OFLOW: /* integer overflow fault */ 567 ucode = FPE_INTOVF; 568 i = SIGFPE; 569 break; 570 571 case T_BOUND: /* bounds check fault */ 572 ucode = FPE_FLTSUB; 573 i = SIGFPE; 574 break; 575 576 case T_DNA: 577 /* 578 * Virtual kernel intercept - pass the DNA exception 579 * to the virtual kernel if it asked to handle it. 580 * This occurs when the virtual kernel is holding 581 * onto the FP context for a different emulated 582 * process then the one currently running. 583 * 584 * We must still call npxdna() since we may have 585 * saved FP state that the virtual kernel needs 586 * to hand over to a different emulated process. 587 */ 588 if (lp->lwp_vkernel && lp->lwp_vkernel->ve && 589 (td->td_pcb->pcb_flags & FP_VIRTFP) 590 ) { 591 npxdna(); 592 break; 593 } 594 595 /* 596 * The kernel may have switched out the FP unit's 597 * state, causing the user process to take a fault 598 * when it tries to use the FP unit. Restore the 599 * state here 600 */ 601 if (npxdna()) 602 goto out; 603 i = SIGFPE; 604 ucode = FPE_FPU_NP_TRAP; 605 break; 606 607 case T_FPOPFLT: /* FPU operand fetch fault */ 608 ucode = ILL_COPROC; 609 i = SIGILL; 610 break; 611 612 case T_XMMFLT: /* SIMD floating-point exception */ 613 ucode = 0; /* XXX */ 614 i = SIGFPE; 615 break; 616 } 617 } else { 618 /* kernel trap */ 619 620 switch (type) { 621 case T_PAGEFLT: /* page fault */ 622 trap_pfault(frame, FALSE); 623 goto out2; 624 625 case T_DNA: 626 /* 627 * The kernel is apparently using fpu for copying. 628 * XXX this should be fatal unless the kernel has 629 * registered such use. 630 */ 631 if (npxdna()) 632 goto out2; 633 break; 634 635 case T_STKFLT: /* stack fault */ 636 case T_PROTFLT: /* general protection fault */ 637 case T_SEGNPFLT: /* segment not present fault */ 638 /* 639 * Invalid segment selectors and out of bounds 640 * %rip's and %rsp's can be set up in user mode. 641 * This causes a fault in kernel mode when the 642 * kernel tries to return to user mode. We want 643 * to get this fault so that we can fix the 644 * problem here and not have to check all the 645 * selectors and pointers when the user changes 646 * them. 647 */ 648 if (mycpu->gd_intr_nesting_level == 0) { 649 /* 650 * NOTE: in 64-bit mode traps push rsp/ss 651 * even if no ring change occurs. 652 */ 653 if (td->td_pcb->pcb_onfault && 654 td->td_pcb->pcb_onfault_sp == 655 frame->tf_rsp) { 656 frame->tf_rip = (register_t) 657 td->td_pcb->pcb_onfault; 658 goto out2; 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 goto out2; 680 } 681 break; 682 683 case T_TRCTRAP: /* trace trap */ 684 #if 0 685 if (frame->tf_rip == (int)IDTVEC(syscall)) { 686 /* 687 * We've just entered system mode via the 688 * syscall lcall. Continue single stepping 689 * silently until the syscall handler has 690 * saved the flags. 691 */ 692 goto out2; 693 } 694 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) { 695 /* 696 * The syscall handler has now saved the 697 * flags. Stop single stepping it. 698 */ 699 frame->tf_rflags &= ~PSL_T; 700 goto out2; 701 } 702 #endif 703 704 /* 705 * Ignore debug register trace traps due to 706 * accesses in the user's address space, which 707 * can happen under several conditions such as 708 * if a user sets a watchpoint on a buffer and 709 * then passes that buffer to a system call. 710 * We still want to get TRCTRAPS for addresses 711 * in kernel space because that is useful when 712 * debugging the kernel. 713 */ 714 #if 0 /* JG */ 715 if (user_dbreg_trap()) { 716 /* 717 * Reset breakpoint bits because the 718 * processor doesn't 719 */ 720 /* XXX check upper bits here */ 721 load_dr6(rdr6() & 0xfffffff0); 722 goto out2; 723 } 724 #endif 725 /* 726 * FALLTHROUGH (TRCTRAP kernel mode, kernel address) 727 */ 728 case T_BPTFLT: 729 /* 730 * If DDB is enabled, let it handle the debugger trap. 731 * Otherwise, debugger traps "can't happen". 732 */ 733 ucode = TRAP_BRKPT; 734 #ifdef DDB 735 MAKEMPSAFE(have_mplock); 736 if (kdb_trap(type, 0, frame)) 737 goto out2; 738 #endif 739 break; 740 741 #if NISA > 0 742 case T_NMI: 743 MAKEMPSAFE(have_mplock); 744 /* machine/parity/power fail/"kitchen sink" faults */ 745 if (isa_nmi(code) == 0) { 746 #ifdef DDB 747 /* 748 * NMI can be hooked up to a pushbutton 749 * for debugging. 750 */ 751 if (ddb_on_nmi) { 752 kprintf ("NMI ... going to debugger\n"); 753 kdb_trap(type, 0, frame); 754 } 755 #endif /* DDB */ 756 goto out2; 757 } else if (panic_on_nmi == 0) 758 goto out2; 759 /* FALL THROUGH */ 760 #endif /* NISA > 0 */ 761 } 762 MAKEMPSAFE(have_mplock); 763 trap_fatal(frame, 0); 764 goto out2; 765 } 766 767 /* 768 * Virtual kernel intercept - if the fault is directly related to a 769 * VM context managed by a virtual kernel then let the virtual kernel 770 * handle it. 771 */ 772 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 773 vkernel_trap(lp, frame); 774 goto out; 775 } 776 777 /* Translate fault for emulators (e.g. Linux) */ 778 if (*p->p_sysent->sv_transtrap) 779 i = (*p->p_sysent->sv_transtrap)(i, type); 780 781 MAKEMPSAFE(have_mplock); 782 trapsignal(lp, i, ucode); 783 784 #ifdef DEBUG 785 if (type <= MAX_TRAP_MSG) { 786 uprintf("fatal process exception: %s", 787 trap_msg[type]); 788 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 789 uprintf(", fault VA = 0x%lx", frame->tf_addr); 790 uprintf("\n"); 791 } 792 #endif 793 794 out: 795 userret(lp, frame, sticks); 796 userexit(lp); 797 out2: ; 798 if (have_mplock) 799 rel_mplock(); 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 #if 0 /* JG */ 896 else if ((frame->tf_err & PGEX_I) && pg_nx != 0) 897 ftype = VM_PROT_EXECUTE; 898 #endif 899 else 900 ftype = VM_PROT_READ; 901 902 if (map != &kernel_map) { 903 /* 904 * Keep swapout from messing with us during this 905 * critical time. 906 */ 907 PHOLD(lp->lwp_proc); 908 909 /* 910 * Issue fault 911 */ 912 fault_flags = 0; 913 if (usermode) 914 fault_flags |= VM_FAULT_BURST; 915 if (ftype & VM_PROT_WRITE) 916 fault_flags |= VM_FAULT_DIRTY; 917 else 918 fault_flags |= VM_FAULT_NORMAL; 919 rv = vm_fault(map, va, ftype, fault_flags); 920 921 PRELE(lp->lwp_proc); 922 } else { 923 /* 924 * Don't have to worry about process locking or stacks in the 925 * kernel. 926 */ 927 fault_flags = VM_FAULT_NORMAL; 928 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 929 } 930 if (rv == KERN_SUCCESS) 931 return (0); 932 nogo: 933 if (!usermode) { 934 /* 935 * NOTE: in 64-bit mode traps push rsp/ss 936 * even if no ring change occurs. 937 */ 938 if (td->td_pcb->pcb_onfault && 939 td->td_pcb->pcb_onfault_sp == frame->tf_rsp && 940 td->td_gd->gd_intr_nesting_level == 0) { 941 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault; 942 return (0); 943 } 944 trap_fatal(frame, frame->tf_addr); 945 return (-1); 946 } 947 948 /* 949 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no 950 * kludge is needed to pass the fault address to signal handlers. 951 */ 952 p = td->td_proc; 953 #ifdef DDB 954 if (td->td_lwp->lwp_vkernel == NULL) { 955 while (freeze_on_seg_fault) { 956 tsleep(p, 0, "freeze", hz * 20); 957 } 958 if (ddb_on_seg_fault) 959 Debugger("ddb_on_seg_fault"); 960 } 961 #endif 962 963 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 964 } 965 966 static void 967 trap_fatal(struct trapframe *frame, vm_offset_t eva) 968 { 969 int code, ss; 970 u_int type; 971 long rsp; 972 struct soft_segment_descriptor softseg; 973 char *msg; 974 975 code = frame->tf_err; 976 type = frame->tf_trapno; 977 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg); 978 979 if (type <= MAX_TRAP_MSG) 980 msg = trap_msg[type]; 981 else 982 msg = "UNKNOWN"; 983 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg, 984 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 985 /* three separate prints in case of a trap on an unmapped page */ 986 kprintf("cpuid = %d; ", mycpu->gd_cpuid); 987 kprintf("lapic->id = %08x\n", lapic->id); 988 if (type == T_PAGEFLT) { 989 kprintf("fault virtual address = 0x%lx\n", eva); 990 kprintf("fault code = %s %s %s, %s\n", 991 code & PGEX_U ? "user" : "supervisor", 992 code & PGEX_W ? "write" : "read", 993 code & PGEX_I ? "instruction" : "data", 994 code & PGEX_P ? "protection violation" : "page not present"); 995 } 996 kprintf("instruction pointer = 0x%lx:0x%lx\n", 997 frame->tf_cs & 0xffff, frame->tf_rip); 998 if (ISPL(frame->tf_cs) == SEL_UPL) { 999 ss = frame->tf_ss & 0xffff; 1000 rsp = frame->tf_rsp; 1001 } else { 1002 /* 1003 * NOTE: in 64-bit mode traps push rsp/ss even if no ring 1004 * change occurs. 1005 */ 1006 ss = GSEL(GDATA_SEL, SEL_KPL); 1007 rsp = frame->tf_rsp; 1008 } 1009 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp); 1010 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp); 1011 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n", 1012 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); 1013 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n", 1014 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32, 1015 softseg.ssd_gran); 1016 kprintf("processor eflags = "); 1017 if (frame->tf_rflags & PSL_T) 1018 kprintf("trace trap, "); 1019 if (frame->tf_rflags & PSL_I) 1020 kprintf("interrupt enabled, "); 1021 if (frame->tf_rflags & PSL_NT) 1022 kprintf("nested task, "); 1023 if (frame->tf_rflags & PSL_RF) 1024 kprintf("resume, "); 1025 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12); 1026 kprintf("current process = "); 1027 if (curproc) { 1028 kprintf("%lu\n", 1029 (u_long)curproc->p_pid); 1030 } else { 1031 kprintf("Idle\n"); 1032 } 1033 kprintf("current thread = pri %d ", curthread->td_pri); 1034 if (curthread->td_critcount) 1035 kprintf("(CRIT)"); 1036 kprintf("\n"); 1037 1038 #ifdef DDB 1039 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame)) 1040 return; 1041 #endif 1042 kprintf("trap number = %d\n", type); 1043 if (type <= MAX_TRAP_MSG) 1044 panic("%s", trap_msg[type]); 1045 else 1046 panic("unknown/reserved trap"); 1047 } 1048 1049 /* 1050 * Double fault handler. Called when a fault occurs while writing 1051 * a frame for a trap/exception onto the stack. This usually occurs 1052 * when the stack overflows (such is the case with infinite recursion, 1053 * for example). 1054 */ 1055 static __inline 1056 int 1057 in_kstack_guard(register_t rptr) 1058 { 1059 thread_t td = curthread; 1060 1061 if ((char *)rptr >= td->td_kstack && 1062 (char *)rptr < td->td_kstack + PAGE_SIZE) { 1063 return 1; 1064 } 1065 return 0; 1066 } 1067 1068 void 1069 dblfault_handler(struct trapframe *frame) 1070 { 1071 thread_t td = curthread; 1072 1073 if (in_kstack_guard(frame->tf_rsp) || in_kstack_guard(frame->tf_rbp)) { 1074 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n"); 1075 if (in_kstack_guard(frame->tf_rsp)) 1076 frame->tf_rsp = (register_t)(td->td_kstack + PAGE_SIZE); 1077 if (in_kstack_guard(frame->tf_rbp)) 1078 frame->tf_rbp = (register_t)(td->td_kstack + PAGE_SIZE); 1079 } else { 1080 kprintf("DOUBLE FAULT\n"); 1081 } 1082 kprintf("\nFatal double fault\n"); 1083 kprintf("rip = 0x%lx\n", frame->tf_rip); 1084 kprintf("rsp = 0x%lx\n", frame->tf_rsp); 1085 kprintf("rbp = 0x%lx\n", frame->tf_rbp); 1086 /* three separate prints in case of a trap on an unmapped page */ 1087 kprintf("cpuid = %d; ", mycpu->gd_cpuid); 1088 kprintf("lapic->id = %08x\n", lapic->id); 1089 panic("double fault"); 1090 } 1091 1092 /* 1093 * syscall2 - MP aware system call request C handler 1094 * 1095 * A system call is essentially treated as a trap except that the 1096 * MP lock is not held on entry or return. We are responsible for 1097 * obtaining the MP lock if necessary and for handling ASTs 1098 * (e.g. a task switch) prior to return. 1099 * 1100 * MPSAFE 1101 */ 1102 void 1103 syscall2(struct trapframe *frame) 1104 { 1105 struct thread *td = curthread; 1106 struct proc *p = td->td_proc; 1107 struct lwp *lp = td->td_lwp; 1108 caddr_t params; 1109 struct sysent *callp; 1110 register_t orig_tf_rflags; 1111 int sticks; 1112 int error; 1113 int narg; 1114 #ifdef INVARIANTS 1115 int crit_count = td->td_critcount; 1116 #endif 1117 int have_mplock = 0; 1118 register_t *argp; 1119 u_int code; 1120 int reg, regcnt; 1121 union sysunion args; 1122 register_t *argsdst; 1123 1124 mycpu->gd_cnt.v_syscall++; 1125 1126 #ifdef DIAGNOSTIC 1127 if (ISPL(frame->tf_cs) != SEL_UPL) { 1128 get_mplock(); 1129 panic("syscall"); 1130 /* NOT REACHED */ 1131 } 1132 #endif 1133 1134 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid, 1135 frame->tf_rax); 1136 1137 userenter(td, p); /* lazy raise our priority */ 1138 1139 reg = 0; 1140 regcnt = 6; 1141 /* 1142 * Misc 1143 */ 1144 sticks = (int)td->td_sticks; 1145 orig_tf_rflags = frame->tf_rflags; 1146 1147 /* 1148 * Virtual kernel intercept - if a VM context managed by a virtual 1149 * kernel issues a system call the virtual kernel handles it, not us. 1150 * Restore the virtual kernel context and return from its system 1151 * call. The current frame is copied out to the virtual kernel. 1152 */ 1153 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 1154 vkernel_trap(lp, frame); 1155 error = EJUSTRETURN; 1156 goto out; 1157 } 1158 1159 /* 1160 * Get the system call parameters and account for time 1161 */ 1162 KASSERT(lp->lwp_md.md_regs == frame, 1163 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame)); 1164 params = (caddr_t)frame->tf_rsp + sizeof(register_t); 1165 code = frame->tf_rax; 1166 1167 if (p->p_sysent->sv_prepsyscall) { 1168 (*p->p_sysent->sv_prepsyscall)( 1169 frame, (int *)(&args.nosys.sysmsg + 1), 1170 &code, ¶ms); 1171 } else { 1172 if (code == SYS_syscall || code == SYS___syscall) { 1173 code = frame->tf_rdi; 1174 reg++; 1175 regcnt--; 1176 } 1177 } 1178 1179 if (p->p_sysent->sv_mask) 1180 code &= p->p_sysent->sv_mask; 1181 1182 if (code >= p->p_sysent->sv_size) 1183 callp = &p->p_sysent->sv_table[0]; 1184 else 1185 callp = &p->p_sysent->sv_table[code]; 1186 1187 narg = callp->sy_narg & SYF_ARGMASK; 1188 1189 /* 1190 * On x86_64 we get up to six arguments in registers. The rest are 1191 * on the stack. The first six members of 'struct trapframe' happen 1192 * to be the registers used to pass arguments, in exactly the right 1193 * order. 1194 */ 1195 argp = &frame->tf_rdi; 1196 argp += reg; 1197 argsdst = (register_t *)(&args.nosys.sysmsg + 1); 1198 /* 1199 * JG can we overflow the space pointed to by 'argsdst' 1200 * either with 'bcopy' or with 'copyin'? 1201 */ 1202 bcopy(argp, argsdst, sizeof(register_t) * regcnt); 1203 /* 1204 * copyin is MP aware, but the tracing code is not 1205 */ 1206 if (narg > regcnt) { 1207 KASSERT(params != NULL, ("copyin args with no params!")); 1208 error = copyin(params, &argsdst[regcnt], 1209 (narg - regcnt) * sizeof(register_t)); 1210 if (error) { 1211 #ifdef KTRACE 1212 if (KTRPOINT(td, KTR_SYSCALL)) { 1213 MAKEMPSAFE(have_mplock); 1214 1215 ktrsyscall(lp, code, narg, 1216 (void *)(&args.nosys.sysmsg + 1)); 1217 } 1218 #endif 1219 goto bad; 1220 } 1221 } 1222 1223 #ifdef KTRACE 1224 if (KTRPOINT(td, KTR_SYSCALL)) { 1225 MAKEMPSAFE(have_mplock); 1226 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1)); 1227 } 1228 #endif 1229 1230 /* 1231 * Default return value is 0 (will be copied to %rax). Double-value 1232 * returns use %rax and %rdx. %rdx is left unchanged for system 1233 * calls which return only one result. 1234 */ 1235 args.sysmsg_fds[0] = 0; 1236 args.sysmsg_fds[1] = frame->tf_rdx; 1237 1238 /* 1239 * The syscall might manipulate the trap frame. If it does it 1240 * will probably return EJUSTRETURN. 1241 */ 1242 args.sysmsg_frame = frame; 1243 1244 STOPEVENT(p, S_SCE, narg); /* MP aware */ 1245 1246 /* 1247 * NOTE: All system calls run MPSAFE now. The system call itself 1248 * is responsible for getting the MP lock. 1249 */ 1250 #ifdef SYSCALL_DEBUG 1251 uint64_t tscval = rdtsc(); 1252 #endif 1253 error = (*callp->sy_call)(&args); 1254 #ifdef SYSCALL_DEBUG 1255 tscval = rdtsc() - tscval; 1256 tscval = tscval * 1000000 / tsc_frequency; 1257 if (SysCallsWorstCase[code] < tscval) 1258 SysCallsWorstCase[code] = tscval; 1259 #endif 1260 1261 out: 1262 /* 1263 * MP SAFE (we may or may not have the MP lock at this point) 1264 */ 1265 //kprintf("SYSMSG %d ", error); 1266 switch (error) { 1267 case 0: 1268 /* 1269 * Reinitialize proc pointer `p' as it may be different 1270 * if this is a child returning from fork syscall. 1271 */ 1272 p = curproc; 1273 lp = curthread->td_lwp; 1274 frame->tf_rax = args.sysmsg_fds[0]; 1275 frame->tf_rdx = args.sysmsg_fds[1]; 1276 frame->tf_rflags &= ~PSL_C; 1277 break; 1278 case ERESTART: 1279 /* 1280 * Reconstruct pc, we know that 'syscall' is 2 bytes. 1281 * We have to do a full context restore so that %r10 1282 * (which was holding the value of %rcx) is restored for 1283 * the next iteration. 1284 */ 1285 if (frame->tf_err != 0 && frame->tf_err != 2) 1286 kprintf("lp %s:%d frame->tf_err is weird %ld\n", 1287 td->td_comm, lp->lwp_proc->p_pid, frame->tf_err); 1288 frame->tf_rip -= frame->tf_err; 1289 frame->tf_r10 = frame->tf_rcx; 1290 break; 1291 case EJUSTRETURN: 1292 break; 1293 case EASYNC: 1294 panic("Unexpected EASYNC return value (for now)"); 1295 default: 1296 bad: 1297 if (p->p_sysent->sv_errsize) { 1298 if (error >= p->p_sysent->sv_errsize) 1299 error = -1; /* XXX */ 1300 else 1301 error = p->p_sysent->sv_errtbl[error]; 1302 } 1303 frame->tf_rax = error; 1304 frame->tf_rflags |= PSL_C; 1305 break; 1306 } 1307 1308 /* 1309 * Traced syscall. trapsignal() is not MP aware. 1310 */ 1311 if (orig_tf_rflags & PSL_T) { 1312 MAKEMPSAFE(have_mplock); 1313 frame->tf_rflags &= ~PSL_T; 1314 trapsignal(lp, SIGTRAP, TRAP_TRACE); 1315 } 1316 1317 /* 1318 * Handle reschedule and other end-of-syscall issues 1319 */ 1320 userret(lp, frame, sticks); 1321 1322 #ifdef KTRACE 1323 if (KTRPOINT(td, KTR_SYSRET)) { 1324 MAKEMPSAFE(have_mplock); 1325 ktrsysret(lp, code, error, args.sysmsg_result); 1326 } 1327 #endif 1328 1329 /* 1330 * This works because errno is findable through the 1331 * register set. If we ever support an emulation where this 1332 * is not the case, this code will need to be revisited. 1333 */ 1334 STOPEVENT(p, S_SCX, code); 1335 1336 userexit(lp); 1337 /* 1338 * Release the MP lock if we had to get it 1339 */ 1340 if (have_mplock) 1341 rel_mplock(); 1342 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error); 1343 #ifdef INVARIANTS 1344 KASSERT(crit_count == td->td_critcount, 1345 ("syscall: critical section count mismatch! %d/%d", 1346 crit_count, td->td_pri)); 1347 KASSERT(&td->td_toks_base == td->td_toks_stop, 1348 ("syscall: extra tokens held after trap! %ld", 1349 td->td_toks_stop - &td->td_toks_base)); 1350 #endif 1351 } 1352 1353 /* 1354 * NOTE: mplock not held at any point 1355 */ 1356 void 1357 fork_return(struct lwp *lp, struct trapframe *frame) 1358 { 1359 frame->tf_rax = 0; /* Child returns zero */ 1360 frame->tf_rflags &= ~PSL_C; /* success */ 1361 frame->tf_rdx = 1; 1362 1363 generic_lwp_return(lp, frame); 1364 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid); 1365 } 1366 1367 /* 1368 * Simplified back end of syscall(), used when returning from fork() 1369 * directly into user mode. 1370 * 1371 * This code will return back into the fork trampoline code which then 1372 * runs doreti. 1373 * 1374 * NOTE: The mplock is not held at any point. 1375 */ 1376 void 1377 generic_lwp_return(struct lwp *lp, struct trapframe *frame) 1378 { 1379 struct proc *p = lp->lwp_proc; 1380 1381 /* 1382 * Newly forked processes are given a kernel priority. We have to 1383 * adjust the priority to a normal user priority and fake entry 1384 * into the kernel (call userenter()) to install a passive release 1385 * function just in case userret() decides to stop the process. This 1386 * can occur when ^Z races a fork. If we do not install the passive 1387 * release function the current process designation will not be 1388 * released when the thread goes to sleep. 1389 */ 1390 lwkt_setpri_self(TDPRI_USER_NORM); 1391 userenter(lp->lwp_thread, p); 1392 userret(lp, frame, 0); 1393 #ifdef KTRACE 1394 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET)) 1395 ktrsysret(lp, SYS_fork, 0, 0); 1396 #endif 1397 lp->lwp_flags |= LWP_PASSIVE_ACQ; 1398 userexit(lp); 1399 lp->lwp_flags &= ~LWP_PASSIVE_ACQ; 1400 } 1401 1402 /* 1403 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA 1404 * fault (which is then passed back to the virtual kernel) if an attempt is 1405 * made to use the FP unit. 1406 * 1407 * XXX this is a fairly big hack. 1408 */ 1409 void 1410 set_vkernel_fp(struct trapframe *frame) 1411 { 1412 struct thread *td = curthread; 1413 1414 if (frame->tf_xflags & PGEX_FPFAULT) { 1415 td->td_pcb->pcb_flags |= FP_VIRTFP; 1416 if (mdcpu->gd_npxthread == td) 1417 npxexit(); 1418 } else { 1419 td->td_pcb->pcb_flags &= ~FP_VIRTFP; 1420 } 1421 } 1422 1423 /* 1424 * Called from vkernel_trap() to fixup the vkernel's syscall 1425 * frame for vmspace_ctl() return. 1426 */ 1427 void 1428 cpu_vkernel_trap(struct trapframe *frame, int error) 1429 { 1430 frame->tf_rax = error; 1431 if (error) 1432 frame->tf_rflags |= PSL_C; 1433 else 1434 frame->tf_rflags &= ~PSL_C; 1435 } 1436