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 || 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 lwkt_gettoken(&lp->lwp_proc->p_token); 336 tstop(); 337 lwkt_reltoken(&lp->lwp_proc->p_token); 338 } 339 340 /* 341 * Reduce our priority in preparation for a return to userland. If 342 * our passive release function was still in place, our priority was 343 * never raised and does not need to be reduced. 344 */ 345 lwkt_passive_recover(td); 346 347 /* WARNING: we may have migrated cpu's */ 348 /* gd = td->td_gd; */ 349 350 /* 351 * Become the current user scheduled process if we aren't already, 352 * and deal with reschedule requests and other factors. 353 */ 354 lp->lwp_proc->p_usched->acquire_curproc(lp); 355 } 356 357 #if !defined(KTR_KERNENTRY) 358 #define KTR_KERNENTRY KTR_ALL 359 #endif 360 KTR_INFO_MASTER(kernentry); 361 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, 362 "TRAP(pid %d, tid %d, trapno %ld, eva %lu)", 363 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva); 364 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %d, tid %d)", 365 pid_t pid, lwpid_t tid); 366 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %d, tid %d, nr %ld)", 367 pid_t pid, lwpid_t tid, register_t trapno); 368 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %d, tid %d, err %d)", 369 pid_t pid, lwpid_t tid, int err); 370 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %d, tid %d)", 371 pid_t pid, lwpid_t tid); 372 373 /* 374 * Exception, fault, and trap interface to the kernel. 375 * This common code is called from assembly language IDT gate entry 376 * routines that prepare a suitable stack frame, and restore this 377 * frame after the exception has been processed. 378 * 379 * This function is also called from doreti in an interlock to handle ASTs. 380 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap 381 * 382 * NOTE! We have to retrieve the fault address prior to obtaining the 383 * MP lock because get_mplock() may switch out. YYY cr2 really ought 384 * to be retrieved by the assembly code, not here. 385 * 386 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing 387 * if an attempt is made to switch from a fast interrupt or IPI. This is 388 * necessary to properly take fatal kernel traps on SMP machines if 389 * get_mplock() has to block. 390 */ 391 392 void 393 trap(struct trapframe *frame) 394 { 395 struct globaldata *gd = mycpu; 396 struct thread *td = gd->gd_curthread; 397 struct lwp *lp = td->td_lwp; 398 struct proc *p; 399 int sticks = 0; 400 int i = 0, ucode = 0, type, code; 401 int have_mplock = 0; 402 #ifdef INVARIANTS 403 int crit_count = td->td_critcount; 404 lwkt_tokref_t curstop = td->td_toks_stop; 405 #endif 406 vm_offset_t eva; 407 408 p = td->td_proc; 409 clear_quickret(); 410 411 #ifdef DDB 412 /* 413 * We need to allow T_DNA faults when the debugger is active since 414 * some dumping paths do large bcopy() which use the floating 415 * point registers for faster copying. 416 */ 417 if (db_active && frame->tf_trapno != T_DNA) { 418 eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0); 419 ++gd->gd_trap_nesting_level; 420 MAKEMPSAFE(have_mplock); 421 trap_fatal(frame, eva); 422 --gd->gd_trap_nesting_level; 423 goto out2; 424 } 425 #endif 426 427 eva = 0; 428 429 if ((frame->tf_rflags & PSL_I) == 0) { 430 /* 431 * Buggy application or kernel code has disabled interrupts 432 * and then trapped. Enabling interrupts now is wrong, but 433 * it is better than running with interrupts disabled until 434 * they are accidentally enabled later. 435 */ 436 type = frame->tf_trapno; 437 if (ISPL(frame->tf_cs) == SEL_UPL) { 438 MAKEMPSAFE(have_mplock); 439 /* JG curproc can be NULL */ 440 kprintf( 441 "pid %ld (%s): trap %d with interrupts disabled\n", 442 (long)curproc->p_pid, curproc->p_comm, type); 443 } else if (type != T_NMI && type != T_BPTFLT && 444 type != T_TRCTRAP) { 445 /* 446 * XXX not quite right, since this may be for a 447 * multiple fault in user mode. 448 */ 449 MAKEMPSAFE(have_mplock); 450 kprintf("kernel trap %d with interrupts disabled\n", 451 type); 452 } 453 cpu_enable_intr(); 454 } 455 456 type = frame->tf_trapno; 457 code = frame->tf_err; 458 459 if (ISPL(frame->tf_cs) == SEL_UPL) { 460 /* user trap */ 461 462 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid, 463 frame->tf_trapno, eva); 464 465 userenter(td, p); 466 467 sticks = (int)td->td_sticks; 468 KASSERT(lp->lwp_md.md_regs == frame, 469 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame)); 470 471 switch (type) { 472 case T_PRIVINFLT: /* privileged instruction fault */ 473 i = SIGILL; 474 ucode = ILL_PRVOPC; 475 break; 476 477 case T_BPTFLT: /* bpt instruction fault */ 478 case T_TRCTRAP: /* trace trap */ 479 frame->tf_rflags &= ~PSL_T; 480 i = SIGTRAP; 481 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT); 482 break; 483 484 case T_ARITHTRAP: /* arithmetic trap */ 485 ucode = code; 486 i = SIGFPE; 487 break; 488 489 case T_ASTFLT: /* Allow process switch */ 490 mycpu->gd_cnt.v_soft++; 491 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) { 492 atomic_clear_int(&mycpu->gd_reqflags, 493 RQF_AST_OWEUPC); 494 addupc_task(p, p->p_prof.pr_addr, 495 p->p_prof.pr_ticks); 496 } 497 goto out; 498 499 case T_PROTFLT: /* general protection fault */ 500 i = SIGBUS; 501 ucode = BUS_OBJERR; 502 break; 503 case T_STKFLT: /* stack fault */ 504 case T_SEGNPFLT: /* segment not present fault */ 505 i = SIGBUS; 506 ucode = BUS_ADRERR; 507 break; 508 case T_TSSFLT: /* invalid TSS fault */ 509 case T_DOUBLEFLT: /* double fault */ 510 default: 511 i = SIGBUS; 512 ucode = BUS_OBJERR; 513 break; 514 515 case T_PAGEFLT: /* page fault */ 516 i = trap_pfault(frame, TRUE); 517 if (frame->tf_rip == 0) { 518 kprintf("T_PAGEFLT: Warning %%rip == 0!\n"); 519 #ifdef DDB 520 while (freeze_on_seg_fault) 521 tsleep(p, 0, "freeze", hz * 20); 522 #endif 523 } 524 if (i == -1 || i == 0) 525 goto out; 526 527 528 if (i == SIGSEGV) 529 ucode = SEGV_MAPERR; 530 else { 531 i = SIGSEGV; 532 ucode = SEGV_ACCERR; 533 } 534 break; 535 536 case T_DIVIDE: /* integer divide fault */ 537 ucode = FPE_INTDIV; 538 i = SIGFPE; 539 break; 540 541 #if NISA > 0 542 case T_NMI: 543 MAKEMPSAFE(have_mplock); 544 /* machine/parity/power fail/"kitchen sink" faults */ 545 if (isa_nmi(code) == 0) { 546 #ifdef DDB 547 /* 548 * NMI can be hooked up to a pushbutton 549 * for debugging. 550 */ 551 if (ddb_on_nmi) { 552 kprintf ("NMI ... going to debugger\n"); 553 kdb_trap(type, 0, frame); 554 } 555 #endif /* DDB */ 556 goto out2; 557 } else if (panic_on_nmi) 558 panic("NMI indicates hardware failure"); 559 break; 560 #endif /* NISA > 0 */ 561 562 case T_OFLOW: /* integer overflow fault */ 563 ucode = FPE_INTOVF; 564 i = SIGFPE; 565 break; 566 567 case T_BOUND: /* bounds check fault */ 568 ucode = FPE_FLTSUB; 569 i = SIGFPE; 570 break; 571 572 case T_DNA: 573 /* 574 * Virtual kernel intercept - pass the DNA exception 575 * to the virtual kernel if it asked to handle it. 576 * This occurs when the virtual kernel is holding 577 * onto the FP context for a different emulated 578 * process then the one currently running. 579 * 580 * We must still call npxdna() since we may have 581 * saved FP state that the virtual kernel needs 582 * to hand over to a different emulated process. 583 */ 584 if (lp->lwp_vkernel && lp->lwp_vkernel->ve && 585 (td->td_pcb->pcb_flags & FP_VIRTFP) 586 ) { 587 npxdna(); 588 break; 589 } 590 591 /* 592 * The kernel may have switched out the FP unit's 593 * state, causing the user process to take a fault 594 * when it tries to use the FP unit. Restore the 595 * state here 596 */ 597 if (npxdna()) 598 goto out; 599 i = SIGFPE; 600 ucode = FPE_FPU_NP_TRAP; 601 break; 602 603 case T_FPOPFLT: /* FPU operand fetch fault */ 604 ucode = ILL_COPROC; 605 i = SIGILL; 606 break; 607 608 case T_XMMFLT: /* SIMD floating-point exception */ 609 ucode = 0; /* XXX */ 610 i = SIGFPE; 611 break; 612 } 613 } else { 614 /* kernel trap */ 615 616 switch (type) { 617 case T_PAGEFLT: /* page fault */ 618 trap_pfault(frame, FALSE); 619 goto out2; 620 621 case T_DNA: 622 /* 623 * The kernel is apparently using fpu for copying. 624 * XXX this should be fatal unless the kernel has 625 * registered such use. 626 */ 627 if (npxdna()) 628 goto out2; 629 break; 630 631 case T_STKFLT: /* stack fault */ 632 case T_PROTFLT: /* general protection fault */ 633 case T_SEGNPFLT: /* segment not present fault */ 634 /* 635 * Invalid segment selectors and out of bounds 636 * %rip's and %rsp's can be set up in user mode. 637 * This causes a fault in kernel mode when the 638 * kernel tries to return to user mode. We want 639 * to get this fault so that we can fix the 640 * problem here and not have to check all the 641 * selectors and pointers when the user changes 642 * them. 643 */ 644 if (mycpu->gd_intr_nesting_level == 0) { 645 /* 646 * NOTE: in 64-bit mode traps push rsp/ss 647 * even if no ring change occurs. 648 */ 649 if (td->td_pcb->pcb_onfault && 650 td->td_pcb->pcb_onfault_sp == 651 frame->tf_rsp) { 652 frame->tf_rip = (register_t) 653 td->td_pcb->pcb_onfault; 654 goto out2; 655 } 656 if (frame->tf_rip == (long)doreti_iret) { 657 frame->tf_rip = (long)doreti_iret_fault; 658 goto out2; 659 } 660 } 661 break; 662 663 case T_TSSFLT: 664 /* 665 * PSL_NT can be set in user mode and isn't cleared 666 * automatically when the kernel is entered. This 667 * causes a TSS fault when the kernel attempts to 668 * `iret' because the TSS link is uninitialized. We 669 * want to get this fault so that we can fix the 670 * problem here and not every time the kernel is 671 * entered. 672 */ 673 if (frame->tf_rflags & PSL_NT) { 674 frame->tf_rflags &= ~PSL_NT; 675 goto out2; 676 } 677 break; 678 679 case T_TRCTRAP: /* trace trap */ 680 #if 0 681 if (frame->tf_rip == (int)IDTVEC(syscall)) { 682 /* 683 * We've just entered system mode via the 684 * syscall lcall. Continue single stepping 685 * silently until the syscall handler has 686 * saved the flags. 687 */ 688 goto out2; 689 } 690 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) { 691 /* 692 * The syscall handler has now saved the 693 * flags. Stop single stepping it. 694 */ 695 frame->tf_rflags &= ~PSL_T; 696 goto out2; 697 } 698 #endif 699 700 /* 701 * Ignore debug register trace traps due to 702 * accesses in the user's address space, which 703 * can happen under several conditions such as 704 * if a user sets a watchpoint on a buffer and 705 * then passes that buffer to a system call. 706 * We still want to get TRCTRAPS for addresses 707 * in kernel space because that is useful when 708 * debugging the kernel. 709 */ 710 #if 0 /* JG */ 711 if (user_dbreg_trap()) { 712 /* 713 * Reset breakpoint bits because the 714 * processor doesn't 715 */ 716 /* XXX check upper bits here */ 717 load_dr6(rdr6() & 0xfffffff0); 718 goto out2; 719 } 720 #endif 721 /* 722 * FALLTHROUGH (TRCTRAP kernel mode, kernel address) 723 */ 724 case T_BPTFLT: 725 /* 726 * If DDB is enabled, let it handle the debugger trap. 727 * Otherwise, debugger traps "can't happen". 728 */ 729 ucode = TRAP_BRKPT; 730 #ifdef DDB 731 MAKEMPSAFE(have_mplock); 732 if (kdb_trap(type, 0, frame)) 733 goto out2; 734 #endif 735 break; 736 737 #if NISA > 0 738 case T_NMI: 739 MAKEMPSAFE(have_mplock); 740 /* machine/parity/power fail/"kitchen sink" faults */ 741 if (isa_nmi(code) == 0) { 742 #ifdef DDB 743 /* 744 * NMI can be hooked up to a pushbutton 745 * for debugging. 746 */ 747 if (ddb_on_nmi) { 748 kprintf ("NMI ... going to debugger\n"); 749 kdb_trap(type, 0, frame); 750 } 751 #endif /* DDB */ 752 goto out2; 753 } else if (panic_on_nmi == 0) 754 goto out2; 755 /* FALL THROUGH */ 756 #endif /* NISA > 0 */ 757 } 758 MAKEMPSAFE(have_mplock); 759 trap_fatal(frame, 0); 760 goto out2; 761 } 762 763 /* 764 * Virtual kernel intercept - if the fault is directly related to a 765 * VM context managed by a virtual kernel then let the virtual kernel 766 * handle it. 767 */ 768 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 769 vkernel_trap(lp, frame); 770 goto out; 771 } 772 773 /* Translate fault for emulators (e.g. Linux) */ 774 if (*p->p_sysent->sv_transtrap) 775 i = (*p->p_sysent->sv_transtrap)(i, type); 776 777 MAKEMPSAFE(have_mplock); 778 trapsignal(lp, i, ucode); 779 780 #ifdef DEBUG 781 if (type <= MAX_TRAP_MSG) { 782 uprintf("fatal process exception: %s", 783 trap_msg[type]); 784 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 785 uprintf(", fault VA = 0x%lx", frame->tf_addr); 786 uprintf("\n"); 787 } 788 #endif 789 790 out: 791 userret(lp, frame, sticks); 792 userexit(lp); 793 out2: ; 794 if (have_mplock) 795 rel_mplock(); 796 if (p != NULL && lp != NULL) 797 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid); 798 #ifdef INVARIANTS 799 KASSERT(crit_count == td->td_critcount, 800 ("trap: critical section count mismatch! %d/%d", 801 crit_count, td->td_pri)); 802 KASSERT(curstop == td->td_toks_stop, 803 ("trap: extra tokens held after trap! %ld/%ld", 804 curstop - &td->td_toks_base, 805 td->td_toks_stop - &td->td_toks_base)); 806 #endif 807 } 808 809 void 810 trap_handle_userenter(struct thread *td) 811 { 812 userenter(td, td->td_proc); 813 } 814 815 void 816 trap_handle_userexit(struct trapframe *frame, int sticks) 817 { 818 struct lwp *lp = curthread->td_lwp; 819 820 if (lp) { 821 userret(lp, frame, sticks); 822 userexit(lp); 823 } 824 } 825 826 static int 827 trap_pfault(struct trapframe *frame, int usermode) 828 { 829 vm_offset_t va; 830 struct vmspace *vm = NULL; 831 vm_map_t map; 832 int rv = 0; 833 int fault_flags; 834 vm_prot_t ftype; 835 thread_t td = curthread; 836 struct lwp *lp = td->td_lwp; 837 struct proc *p; 838 839 va = trunc_page(frame->tf_addr); 840 if (va >= VM_MIN_KERNEL_ADDRESS) { 841 /* 842 * Don't allow user-mode faults in kernel address space. 843 */ 844 if (usermode) { 845 fault_flags = -1; 846 ftype = -1; 847 goto nogo; 848 } 849 850 map = &kernel_map; 851 } else { 852 /* 853 * This is a fault on non-kernel virtual memory. 854 * vm is initialized above to NULL. If curproc is NULL 855 * or curproc->p_vmspace is NULL the fault is fatal. 856 */ 857 if (lp != NULL) 858 vm = lp->lwp_vmspace; 859 860 if (vm == NULL) { 861 fault_flags = -1; 862 ftype = -1; 863 goto nogo; 864 } 865 866 /* 867 * Debugging, try to catch kernel faults on the user address 868 * space when not inside on onfault (e.g. copyin/copyout) 869 * routine. 870 */ 871 if (usermode == 0 && (td->td_pcb == NULL || 872 td->td_pcb->pcb_onfault == NULL)) { 873 #ifdef DDB 874 if (freeze_on_seg_fault) { 875 kprintf("trap_pfault: user address fault from kernel mode " 876 "%016lx\n", (long)frame->tf_addr); 877 while (freeze_on_seg_fault) 878 tsleep(&freeze_on_seg_fault, 0, "frzseg", hz * 20); 879 } 880 #endif 881 } 882 map = &vm->vm_map; 883 } 884 885 /* 886 * PGEX_I is defined only if the execute disable bit capability is 887 * supported and enabled. 888 */ 889 if (frame->tf_err & PGEX_W) 890 ftype = VM_PROT_WRITE; 891 #if 0 /* JG */ 892 else if ((frame->tf_err & PGEX_I) && pg_nx != 0) 893 ftype = VM_PROT_EXECUTE; 894 #endif 895 else 896 ftype = VM_PROT_READ; 897 898 if (map != &kernel_map) { 899 /* 900 * Keep swapout from messing with us during this 901 * critical time. 902 */ 903 PHOLD(lp->lwp_proc); 904 905 /* 906 * Issue fault 907 */ 908 fault_flags = 0; 909 if (usermode) 910 fault_flags |= VM_FAULT_BURST; 911 if (ftype & VM_PROT_WRITE) 912 fault_flags |= VM_FAULT_DIRTY; 913 else 914 fault_flags |= VM_FAULT_NORMAL; 915 rv = vm_fault(map, va, ftype, fault_flags); 916 917 PRELE(lp->lwp_proc); 918 } else { 919 /* 920 * Don't have to worry about process locking or stacks in the 921 * kernel. 922 */ 923 fault_flags = VM_FAULT_NORMAL; 924 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 925 } 926 if (rv == KERN_SUCCESS) 927 return (0); 928 nogo: 929 if (!usermode) { 930 /* 931 * NOTE: in 64-bit mode traps push rsp/ss 932 * even if no ring change occurs. 933 */ 934 if (td->td_pcb->pcb_onfault && 935 td->td_pcb->pcb_onfault_sp == frame->tf_rsp && 936 td->td_gd->gd_intr_nesting_level == 0) { 937 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault; 938 return (0); 939 } 940 trap_fatal(frame, frame->tf_addr); 941 return (-1); 942 } 943 944 /* 945 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no 946 * kludge is needed to pass the fault address to signal handlers. 947 */ 948 p = td->td_proc; 949 #ifdef DDB 950 if (td->td_lwp->lwp_vkernel == NULL) { 951 while (freeze_on_seg_fault) { 952 tsleep(p, 0, "freeze", hz * 20); 953 } 954 if (ddb_on_seg_fault) 955 Debugger("ddb_on_seg_fault"); 956 } 957 #endif 958 959 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 960 } 961 962 static void 963 trap_fatal(struct trapframe *frame, vm_offset_t eva) 964 { 965 int code, ss; 966 u_int type; 967 long rsp; 968 struct soft_segment_descriptor softseg; 969 char *msg; 970 971 code = frame->tf_err; 972 type = frame->tf_trapno; 973 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg); 974 975 if (type <= MAX_TRAP_MSG) 976 msg = trap_msg[type]; 977 else 978 msg = "UNKNOWN"; 979 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg, 980 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 981 /* three separate prints in case of a trap on an unmapped page */ 982 kprintf("cpuid = %d; ", mycpu->gd_cpuid); 983 kprintf("lapic->id = %08x\n", lapic->id); 984 if (type == T_PAGEFLT) { 985 kprintf("fault virtual address = 0x%lx\n", eva); 986 kprintf("fault code = %s %s %s, %s\n", 987 code & PGEX_U ? "user" : "supervisor", 988 code & PGEX_W ? "write" : "read", 989 code & PGEX_I ? "instruction" : "data", 990 code & PGEX_P ? "protection violation" : "page not present"); 991 } 992 kprintf("instruction pointer = 0x%lx:0x%lx\n", 993 frame->tf_cs & 0xffff, frame->tf_rip); 994 if (ISPL(frame->tf_cs) == SEL_UPL) { 995 ss = frame->tf_ss & 0xffff; 996 rsp = frame->tf_rsp; 997 } else { 998 /* 999 * NOTE: in 64-bit mode traps push rsp/ss even if no ring 1000 * change occurs. 1001 */ 1002 ss = GSEL(GDATA_SEL, SEL_KPL); 1003 rsp = frame->tf_rsp; 1004 } 1005 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp); 1006 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp); 1007 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n", 1008 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); 1009 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n", 1010 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32, 1011 softseg.ssd_gran); 1012 kprintf("processor eflags = "); 1013 if (frame->tf_rflags & PSL_T) 1014 kprintf("trace trap, "); 1015 if (frame->tf_rflags & PSL_I) 1016 kprintf("interrupt enabled, "); 1017 if (frame->tf_rflags & PSL_NT) 1018 kprintf("nested task, "); 1019 if (frame->tf_rflags & PSL_RF) 1020 kprintf("resume, "); 1021 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12); 1022 kprintf("current process = "); 1023 if (curproc) { 1024 kprintf("%lu\n", 1025 (u_long)curproc->p_pid); 1026 } else { 1027 kprintf("Idle\n"); 1028 } 1029 kprintf("current thread = pri %d ", curthread->td_pri); 1030 if (curthread->td_critcount) 1031 kprintf("(CRIT)"); 1032 kprintf("\n"); 1033 1034 #ifdef DDB 1035 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame)) 1036 return; 1037 #endif 1038 kprintf("trap number = %d\n", type); 1039 if (type <= MAX_TRAP_MSG) 1040 panic("%s", trap_msg[type]); 1041 else 1042 panic("unknown/reserved trap"); 1043 } 1044 1045 /* 1046 * Double fault handler. Called when a fault occurs while writing 1047 * a frame for a trap/exception onto the stack. This usually occurs 1048 * when the stack overflows (such is the case with infinite recursion, 1049 * for example). 1050 */ 1051 static __inline 1052 int 1053 in_kstack_guard(register_t rptr) 1054 { 1055 thread_t td = curthread; 1056 1057 if ((char *)rptr >= td->td_kstack && 1058 (char *)rptr < td->td_kstack + PAGE_SIZE) { 1059 return 1; 1060 } 1061 return 0; 1062 } 1063 1064 void 1065 dblfault_handler(struct trapframe *frame) 1066 { 1067 thread_t td = curthread; 1068 1069 if (in_kstack_guard(frame->tf_rsp) || in_kstack_guard(frame->tf_rbp)) { 1070 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n"); 1071 if (in_kstack_guard(frame->tf_rsp)) 1072 frame->tf_rsp = (register_t)(td->td_kstack + PAGE_SIZE); 1073 if (in_kstack_guard(frame->tf_rbp)) 1074 frame->tf_rbp = (register_t)(td->td_kstack + PAGE_SIZE); 1075 } else { 1076 kprintf("DOUBLE FAULT\n"); 1077 } 1078 kprintf("\nFatal double fault\n"); 1079 kprintf("rip = 0x%lx\n", frame->tf_rip); 1080 kprintf("rsp = 0x%lx\n", frame->tf_rsp); 1081 kprintf("rbp = 0x%lx\n", frame->tf_rbp); 1082 /* three separate prints in case of a trap on an unmapped page */ 1083 kprintf("cpuid = %d; ", mycpu->gd_cpuid); 1084 kprintf("lapic->id = %08x\n", lapic->id); 1085 panic("double fault"); 1086 } 1087 1088 /* 1089 * syscall2 - MP aware system call request C handler 1090 * 1091 * A system call is essentially treated as a trap except that the 1092 * MP lock is not held on entry or return. We are responsible for 1093 * obtaining the MP lock if necessary and for handling ASTs 1094 * (e.g. a task switch) prior to return. 1095 * 1096 * MPSAFE 1097 */ 1098 void 1099 syscall2(struct trapframe *frame) 1100 { 1101 struct thread *td = curthread; 1102 struct proc *p = td->td_proc; 1103 struct lwp *lp = td->td_lwp; 1104 caddr_t params; 1105 struct sysent *callp; 1106 register_t orig_tf_rflags; 1107 int sticks; 1108 int error; 1109 int narg; 1110 #ifdef INVARIANTS 1111 int crit_count = td->td_critcount; 1112 #endif 1113 int have_mplock = 0; 1114 register_t *argp; 1115 u_int code; 1116 int reg, regcnt; 1117 union sysunion args; 1118 register_t *argsdst; 1119 1120 mycpu->gd_cnt.v_syscall++; 1121 1122 #ifdef DIAGNOSTIC 1123 if (ISPL(frame->tf_cs) != SEL_UPL) { 1124 get_mplock(); 1125 panic("syscall"); 1126 /* NOT REACHED */ 1127 } 1128 #endif 1129 1130 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid, 1131 frame->tf_rax); 1132 1133 userenter(td, p); /* lazy raise our priority */ 1134 1135 reg = 0; 1136 regcnt = 6; 1137 /* 1138 * Misc 1139 */ 1140 sticks = (int)td->td_sticks; 1141 orig_tf_rflags = frame->tf_rflags; 1142 1143 /* 1144 * Virtual kernel intercept - if a VM context managed by a virtual 1145 * kernel issues a system call the virtual kernel handles it, not us. 1146 * Restore the virtual kernel context and return from its system 1147 * call. The current frame is copied out to the virtual kernel. 1148 */ 1149 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 1150 vkernel_trap(lp, frame); 1151 error = EJUSTRETURN; 1152 goto out; 1153 } 1154 1155 /* 1156 * Get the system call parameters and account for time 1157 */ 1158 KASSERT(lp->lwp_md.md_regs == frame, 1159 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame)); 1160 params = (caddr_t)frame->tf_rsp + sizeof(register_t); 1161 code = frame->tf_rax; 1162 1163 if (p->p_sysent->sv_prepsyscall) { 1164 (*p->p_sysent->sv_prepsyscall)( 1165 frame, (int *)(&args.nosys.sysmsg + 1), 1166 &code, ¶ms); 1167 } else { 1168 if (code == SYS_syscall || code == SYS___syscall) { 1169 code = frame->tf_rdi; 1170 reg++; 1171 regcnt--; 1172 } 1173 } 1174 1175 if (p->p_sysent->sv_mask) 1176 code &= p->p_sysent->sv_mask; 1177 1178 if (code >= p->p_sysent->sv_size) 1179 callp = &p->p_sysent->sv_table[0]; 1180 else 1181 callp = &p->p_sysent->sv_table[code]; 1182 1183 narg = callp->sy_narg & SYF_ARGMASK; 1184 1185 /* 1186 * On x86_64 we get up to six arguments in registers. The rest are 1187 * on the stack. The first six members of 'struct trapframe' happen 1188 * to be the registers used to pass arguments, in exactly the right 1189 * order. 1190 */ 1191 argp = &frame->tf_rdi; 1192 argp += reg; 1193 argsdst = (register_t *)(&args.nosys.sysmsg + 1); 1194 /* 1195 * JG can we overflow the space pointed to by 'argsdst' 1196 * either with 'bcopy' or with 'copyin'? 1197 */ 1198 bcopy(argp, argsdst, sizeof(register_t) * regcnt); 1199 /* 1200 * copyin is MP aware, but the tracing code is not 1201 */ 1202 if (narg > regcnt) { 1203 KASSERT(params != NULL, ("copyin args with no params!")); 1204 error = copyin(params, &argsdst[regcnt], 1205 (narg - regcnt) * sizeof(register_t)); 1206 if (error) { 1207 #ifdef KTRACE 1208 if (KTRPOINT(td, KTR_SYSCALL)) { 1209 MAKEMPSAFE(have_mplock); 1210 1211 ktrsyscall(lp, code, narg, 1212 (void *)(&args.nosys.sysmsg + 1)); 1213 } 1214 #endif 1215 goto bad; 1216 } 1217 } 1218 1219 #ifdef KTRACE 1220 if (KTRPOINT(td, KTR_SYSCALL)) { 1221 MAKEMPSAFE(have_mplock); 1222 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1)); 1223 } 1224 #endif 1225 1226 /* 1227 * Default return value is 0 (will be copied to %rax). Double-value 1228 * returns use %rax and %rdx. %rdx is left unchanged for system 1229 * calls which return only one result. 1230 */ 1231 args.sysmsg_fds[0] = 0; 1232 args.sysmsg_fds[1] = frame->tf_rdx; 1233 1234 /* 1235 * The syscall might manipulate the trap frame. If it does it 1236 * will probably return EJUSTRETURN. 1237 */ 1238 args.sysmsg_frame = frame; 1239 1240 STOPEVENT(p, S_SCE, narg); /* MP aware */ 1241 1242 /* 1243 * NOTE: All system calls run MPSAFE now. The system call itself 1244 * is responsible for getting the MP lock. 1245 */ 1246 #ifdef SYSCALL_DEBUG 1247 uint64_t tscval = rdtsc(); 1248 #endif 1249 error = (*callp->sy_call)(&args); 1250 #ifdef SYSCALL_DEBUG 1251 tscval = rdtsc() - tscval; 1252 tscval = tscval * 1000000 / tsc_frequency; 1253 if (SysCallsWorstCase[code] < tscval) 1254 SysCallsWorstCase[code] = tscval; 1255 #endif 1256 1257 out: 1258 /* 1259 * MP SAFE (we may or may not have the MP lock at this point) 1260 */ 1261 //kprintf("SYSMSG %d ", error); 1262 switch (error) { 1263 case 0: 1264 /* 1265 * Reinitialize proc pointer `p' as it may be different 1266 * if this is a child returning from fork syscall. 1267 */ 1268 p = curproc; 1269 lp = curthread->td_lwp; 1270 frame->tf_rax = args.sysmsg_fds[0]; 1271 frame->tf_rdx = args.sysmsg_fds[1]; 1272 frame->tf_rflags &= ~PSL_C; 1273 break; 1274 case ERESTART: 1275 /* 1276 * Reconstruct pc, we know that 'syscall' is 2 bytes. 1277 * We have to do a full context restore so that %r10 1278 * (which was holding the value of %rcx) is restored for 1279 * the next iteration. 1280 */ 1281 if (frame->tf_err != 0 && frame->tf_err != 2) 1282 kprintf("lp %s:%d frame->tf_err is weird %ld\n", 1283 td->td_comm, lp->lwp_proc->p_pid, frame->tf_err); 1284 frame->tf_rip -= frame->tf_err; 1285 frame->tf_r10 = frame->tf_rcx; 1286 break; 1287 case EJUSTRETURN: 1288 break; 1289 case EASYNC: 1290 panic("Unexpected EASYNC return value (for now)"); 1291 default: 1292 bad: 1293 if (p->p_sysent->sv_errsize) { 1294 if (error >= p->p_sysent->sv_errsize) 1295 error = -1; /* XXX */ 1296 else 1297 error = p->p_sysent->sv_errtbl[error]; 1298 } 1299 frame->tf_rax = error; 1300 frame->tf_rflags |= PSL_C; 1301 break; 1302 } 1303 1304 /* 1305 * Traced syscall. trapsignal() is not MP aware. 1306 */ 1307 if (orig_tf_rflags & PSL_T) { 1308 MAKEMPSAFE(have_mplock); 1309 frame->tf_rflags &= ~PSL_T; 1310 trapsignal(lp, SIGTRAP, TRAP_TRACE); 1311 } 1312 1313 /* 1314 * Handle reschedule and other end-of-syscall issues 1315 */ 1316 userret(lp, frame, sticks); 1317 1318 #ifdef KTRACE 1319 if (KTRPOINT(td, KTR_SYSRET)) { 1320 MAKEMPSAFE(have_mplock); 1321 ktrsysret(lp, code, error, args.sysmsg_result); 1322 } 1323 #endif 1324 1325 /* 1326 * This works because errno is findable through the 1327 * register set. If we ever support an emulation where this 1328 * is not the case, this code will need to be revisited. 1329 */ 1330 STOPEVENT(p, S_SCX, code); 1331 1332 userexit(lp); 1333 /* 1334 * Release the MP lock if we had to get it 1335 */ 1336 if (have_mplock) 1337 rel_mplock(); 1338 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error); 1339 #ifdef INVARIANTS 1340 KASSERT(crit_count == td->td_critcount, 1341 ("syscall: critical section count mismatch! %d/%d", 1342 crit_count, td->td_pri)); 1343 KASSERT(&td->td_toks_base == td->td_toks_stop, 1344 ("syscall: extra tokens held after trap! %ld", 1345 td->td_toks_stop - &td->td_toks_base)); 1346 #endif 1347 } 1348 1349 /* 1350 * NOTE: mplock not held at any point 1351 */ 1352 void 1353 fork_return(struct lwp *lp, struct trapframe *frame) 1354 { 1355 frame->tf_rax = 0; /* Child returns zero */ 1356 frame->tf_rflags &= ~PSL_C; /* success */ 1357 frame->tf_rdx = 1; 1358 1359 generic_lwp_return(lp, frame); 1360 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid); 1361 } 1362 1363 /* 1364 * Simplified back end of syscall(), used when returning from fork() 1365 * directly into user mode. 1366 * 1367 * This code will return back into the fork trampoline code which then 1368 * runs doreti. 1369 * 1370 * NOTE: The mplock is not held at any point. 1371 */ 1372 void 1373 generic_lwp_return(struct lwp *lp, struct trapframe *frame) 1374 { 1375 struct proc *p = lp->lwp_proc; 1376 1377 /* 1378 * Newly forked processes are given a kernel priority. We have to 1379 * adjust the priority to a normal user priority and fake entry 1380 * into the kernel (call userenter()) to install a passive release 1381 * function just in case userret() decides to stop the process. This 1382 * can occur when ^Z races a fork. If we do not install the passive 1383 * release function the current process designation will not be 1384 * released when the thread goes to sleep. 1385 */ 1386 lwkt_setpri_self(TDPRI_USER_NORM); 1387 userenter(lp->lwp_thread, p); 1388 userret(lp, frame, 0); 1389 #ifdef KTRACE 1390 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET)) 1391 ktrsysret(lp, SYS_fork, 0, 0); 1392 #endif 1393 lp->lwp_flags |= LWP_PASSIVE_ACQ; 1394 userexit(lp); 1395 lp->lwp_flags &= ~LWP_PASSIVE_ACQ; 1396 } 1397 1398 /* 1399 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA 1400 * fault (which is then passed back to the virtual kernel) if an attempt is 1401 * made to use the FP unit. 1402 * 1403 * XXX this is a fairly big hack. 1404 */ 1405 void 1406 set_vkernel_fp(struct trapframe *frame) 1407 { 1408 struct thread *td = curthread; 1409 1410 if (frame->tf_xflags & PGEX_FPFAULT) { 1411 td->td_pcb->pcb_flags |= FP_VIRTFP; 1412 if (mdcpu->gd_npxthread == td) 1413 npxexit(); 1414 } else { 1415 td->td_pcb->pcb_flags &= ~FP_VIRTFP; 1416 } 1417 } 1418 1419 /* 1420 * Called from vkernel_trap() to fixup the vkernel's syscall 1421 * frame for vmspace_ctl() return. 1422 */ 1423 void 1424 cpu_vkernel_trap(struct trapframe *frame, int error) 1425 { 1426 frame->tf_rax = error; 1427 if (error) 1428 frame->tf_rflags |= PSL_C; 1429 else 1430 frame->tf_rflags &= ~PSL_C; 1431 } 1432