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 92 #define MAKEMPSAFE(have_mplock) \ 93 if (have_mplock == 0) { \ 94 get_mplock(); \ 95 have_mplock = 1; \ 96 } 97 98 extern void trap(struct trapframe *frame); 99 100 static int trap_pfault(struct trapframe *, int); 101 static void trap_fatal(struct trapframe *, vm_offset_t); 102 void dblfault_handler(struct trapframe *frame); 103 104 #define MAX_TRAP_MSG 30 105 static char *trap_msg[] = { 106 "", /* 0 unused */ 107 "privileged instruction fault", /* 1 T_PRIVINFLT */ 108 "", /* 2 unused */ 109 "breakpoint instruction fault", /* 3 T_BPTFLT */ 110 "", /* 4 unused */ 111 "", /* 5 unused */ 112 "arithmetic trap", /* 6 T_ARITHTRAP */ 113 "system forced exception", /* 7 T_ASTFLT */ 114 "", /* 8 unused */ 115 "general protection fault", /* 9 T_PROTFLT */ 116 "trace trap", /* 10 T_TRCTRAP */ 117 "", /* 11 unused */ 118 "page fault", /* 12 T_PAGEFLT */ 119 "", /* 13 unused */ 120 "alignment fault", /* 14 T_ALIGNFLT */ 121 "", /* 15 unused */ 122 "", /* 16 unused */ 123 "", /* 17 unused */ 124 "integer divide fault", /* 18 T_DIVIDE */ 125 "non-maskable interrupt trap", /* 19 T_NMI */ 126 "overflow trap", /* 20 T_OFLOW */ 127 "FPU bounds check fault", /* 21 T_BOUND */ 128 "FPU device not available", /* 22 T_DNA */ 129 "double fault", /* 23 T_DOUBLEFLT */ 130 "FPU operand fetch fault", /* 24 T_FPOPFLT */ 131 "invalid TSS fault", /* 25 T_TSSFLT */ 132 "segment not present fault", /* 26 T_SEGNPFLT */ 133 "stack fault", /* 27 T_STKFLT */ 134 "machine check trap", /* 28 T_MCHK */ 135 "SIMD floating-point exception", /* 29 T_XMMFLT */ 136 "reserved (unknown) fault", /* 30 T_RESERVED */ 137 }; 138 139 #ifdef DDB 140 static int ddb_on_nmi = 1; 141 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW, 142 &ddb_on_nmi, 0, "Go to DDB on NMI"); 143 static int ddb_on_seg_fault = 0; 144 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_seg_fault, CTLFLAG_RW, 145 &ddb_on_seg_fault, 0, "Go to DDB on user seg-fault"); 146 static int freeze_on_seg_fault = 0; 147 SYSCTL_INT(_machdep, OID_AUTO, freeze_on_seg_fault, CTLFLAG_RW, 148 &freeze_on_seg_fault, 0, "Go to DDB on user seg-fault"); 149 #endif 150 static int panic_on_nmi = 1; 151 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW, 152 &panic_on_nmi, 0, "Panic on NMI"); 153 static int fast_release; 154 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW, 155 &fast_release, 0, "Passive Release was optimal"); 156 static int slow_release; 157 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW, 158 &slow_release, 0, "Passive Release was nonoptimal"); 159 160 /* 161 * System call debugging records the worst-case system call 162 * overhead (inclusive of blocking), but may be inaccurate. 163 */ 164 /*#define SYSCALL_DEBUG*/ 165 #ifdef SYSCALL_DEBUG 166 uint64_t SysCallsWorstCase[SYS_MAXSYSCALL]; 167 #endif 168 169 /* 170 * Passively intercepts the thread switch function to increase 171 * the thread priority from a user priority to a kernel priority, reducing 172 * syscall and trap overhead for the case where no switch occurs. 173 * 174 * Synchronizes td_ucred with p_ucred. This is used by system calls, 175 * signal handling, faults, AST traps, and anything else that enters the 176 * kernel from userland and provides the kernel with a stable read-only 177 * copy of the process ucred. 178 */ 179 static __inline void 180 userenter(struct thread *curtd, struct proc *curp) 181 { 182 struct ucred *ocred; 183 struct ucred *ncred; 184 185 curtd->td_release = lwkt_passive_release; 186 187 if (curtd->td_ucred != curp->p_ucred) { 188 ncred = crhold(curp->p_ucred); 189 ocred = curtd->td_ucred; 190 curtd->td_ucred = ncred; 191 if (ocred) 192 crfree(ocred); 193 } 194 195 #ifdef DDB 196 /* 197 * Debugging, remove top two user stack pages to catch kernel faults 198 */ 199 if (freeze_on_seg_fault > 1 && curtd->td_lwp) { 200 pmap_remove(vmspace_pmap(curtd->td_lwp->lwp_vmspace), 201 0x00007FFFFFFFD000LU, 202 0x0000800000000000LU); 203 } 204 #endif 205 } 206 207 /* 208 * Handle signals, upcalls, profiling, and other AST's and/or tasks that 209 * must be completed before we can return to or try to return to userland. 210 * 211 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64 212 * arithmatic on the delta calculation so the absolute tick values are 213 * truncated to an integer. 214 */ 215 static void 216 userret(struct lwp *lp, struct trapframe *frame, int sticks) 217 { 218 struct proc *p = lp->lwp_proc; 219 int sig; 220 221 /* 222 * Charge system time if profiling. Note: times are in microseconds. 223 * This may do a copyout and block, so do it first even though it 224 * means some system time will be charged as user time. 225 */ 226 if (p->p_flags & P_PROFIL) { 227 addupc_task(p, frame->tf_rip, 228 (u_int)((int)lp->lwp_thread->td_sticks - sticks)); 229 } 230 231 recheck: 232 /* 233 * Specific on-return-to-usermode checks (LWP_MP_WEXIT, 234 * LWP_MP_VNLRU, etc). 235 */ 236 if (lp->lwp_mpflags & LWP_MP_URETMASK) 237 lwpuserret(lp); 238 239 /* 240 * Block here if we are in a stopped state. 241 */ 242 if (p->p_stat == SSTOP || dump_stop_usertds) { 243 lwkt_gettoken(&p->p_token); 244 tstop(); 245 lwkt_reltoken(&p->p_token); 246 goto recheck; 247 } 248 249 /* 250 * Post any pending upcalls. If running a virtual kernel be sure 251 * to restore the virtual kernel's vmspace before posting the upcall. 252 */ 253 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF)) { 254 lwkt_gettoken(&p->p_token); 255 if (p->p_flags & P_SIGVTALRM) { 256 p->p_flags &= ~P_SIGVTALRM; 257 ksignal(p, SIGVTALRM); 258 } 259 if (p->p_flags & P_SIGPROF) { 260 p->p_flags &= ~P_SIGPROF; 261 ksignal(p, SIGPROF); 262 } 263 lwkt_reltoken(&p->p_token); 264 goto recheck; 265 } 266 267 /* 268 * Post any pending signals. If running a virtual kernel be sure 269 * to restore the virtual kernel's vmspace before posting the signal. 270 * 271 * WARNING! postsig() can exit and not return. 272 */ 273 if ((sig = CURSIG_TRACE(lp)) != 0) { 274 lwkt_gettoken(&p->p_token); 275 postsig(sig); 276 lwkt_reltoken(&p->p_token); 277 goto recheck; 278 } 279 280 /* 281 * block here if we are swapped out, but still process signals 282 * (such as SIGKILL). proc0 (the swapin scheduler) is already 283 * aware of our situation, we do not have to wake it up. 284 */ 285 if (p->p_flags & P_SWAPPEDOUT) { 286 lwkt_gettoken(&p->p_token); 287 get_mplock(); 288 p->p_flags |= P_SWAPWAIT; 289 swapin_request(); 290 if (p->p_flags & P_SWAPWAIT) 291 tsleep(p, PCATCH, "SWOUT", 0); 292 p->p_flags &= ~P_SWAPWAIT; 293 rel_mplock(); 294 lwkt_reltoken(&p->p_token); 295 goto recheck; 296 } 297 298 /* 299 * In a multi-threaded program it is possible for a thread to change 300 * signal state during a system call which temporarily changes the 301 * signal mask. In this case postsig() might not be run and we 302 * have to restore the mask ourselves. 303 */ 304 if (lp->lwp_flags & LWP_OLDMASK) { 305 lp->lwp_flags &= ~LWP_OLDMASK; 306 lp->lwp_sigmask = lp->lwp_oldsigmask; 307 goto recheck; 308 } 309 } 310 311 /* 312 * Cleanup from userenter and any passive release that might have occured. 313 * We must reclaim the current-process designation before we can return 314 * to usermode. We also handle both LWKT and USER reschedule requests. 315 */ 316 static __inline void 317 userexit(struct lwp *lp) 318 { 319 struct thread *td = lp->lwp_thread; 320 /* globaldata_t gd = td->td_gd; */ 321 322 /* 323 * Handle stop requests at kernel priority. Any requests queued 324 * after this loop will generate another AST. 325 */ 326 while (lp->lwp_proc->p_stat == SSTOP) { 327 lwkt_gettoken(&lp->lwp_proc->p_token); 328 tstop(); 329 lwkt_reltoken(&lp->lwp_proc->p_token); 330 } 331 332 /* 333 * Reduce our priority in preparation for a return to userland. If 334 * our passive release function was still in place, our priority was 335 * never raised and does not need to be reduced. 336 */ 337 lwkt_passive_recover(td); 338 339 /* WARNING: we may have migrated cpu's */ 340 /* gd = td->td_gd; */ 341 342 /* 343 * Become the current user scheduled process if we aren't already, 344 * and deal with reschedule requests and other factors. 345 */ 346 lp->lwp_proc->p_usched->acquire_curproc(lp); 347 } 348 349 #if !defined(KTR_KERNENTRY) 350 #define KTR_KERNENTRY KTR_ALL 351 #endif 352 KTR_INFO_MASTER(kernentry); 353 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, 354 "TRAP(pid %d, tid %d, trapno %ld, eva %lu)", 355 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva); 356 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %d, tid %d)", 357 pid_t pid, lwpid_t tid); 358 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %d, tid %d, nr %ld)", 359 pid_t pid, lwpid_t tid, register_t trapno); 360 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %d, tid %d, err %d)", 361 pid_t pid, lwpid_t tid, int err); 362 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %d, tid %d)", 363 pid_t pid, lwpid_t tid); 364 365 /* 366 * Exception, fault, and trap interface to the kernel. 367 * This common code is called from assembly language IDT gate entry 368 * routines that prepare a suitable stack frame, and restore this 369 * frame after the exception has been processed. 370 * 371 * This function is also called from doreti in an interlock to handle ASTs. 372 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap 373 * 374 * NOTE! We have to retrieve the fault address prior to obtaining the 375 * MP lock because get_mplock() may switch out. YYY cr2 really ought 376 * to be retrieved by the assembly code, not here. 377 * 378 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing 379 * if an attempt is made to switch from a fast interrupt or IPI. This is 380 * necessary to properly take fatal kernel traps on SMP machines if 381 * get_mplock() has to block. 382 */ 383 384 void 385 trap(struct trapframe *frame) 386 { 387 struct globaldata *gd = mycpu; 388 struct thread *td = gd->gd_curthread; 389 struct lwp *lp = td->td_lwp; 390 struct proc *p; 391 int sticks = 0; 392 int i = 0, ucode = 0, type, code; 393 int have_mplock = 0; 394 #ifdef INVARIANTS 395 int crit_count = td->td_critcount; 396 lwkt_tokref_t curstop = td->td_toks_stop; 397 #endif 398 vm_offset_t eva; 399 400 p = td->td_proc; 401 clear_quickret(); 402 403 #ifdef DDB 404 /* 405 * We need to allow T_DNA faults when the debugger is active since 406 * some dumping paths do large bcopy() which use the floating 407 * point registers for faster copying. 408 */ 409 if (db_active && frame->tf_trapno != T_DNA) { 410 eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0); 411 ++gd->gd_trap_nesting_level; 412 MAKEMPSAFE(have_mplock); 413 trap_fatal(frame, eva); 414 --gd->gd_trap_nesting_level; 415 goto out2; 416 } 417 #endif 418 419 eva = 0; 420 421 if ((frame->tf_rflags & PSL_I) == 0) { 422 /* 423 * Buggy application or kernel code has disabled interrupts 424 * and then trapped. Enabling interrupts now is wrong, but 425 * it is better than running with interrupts disabled until 426 * they are accidentally enabled later. 427 */ 428 type = frame->tf_trapno; 429 if (ISPL(frame->tf_cs) == SEL_UPL) { 430 MAKEMPSAFE(have_mplock); 431 /* JG curproc can be NULL */ 432 kprintf( 433 "pid %ld (%s): trap %d with interrupts disabled\n", 434 (long)curproc->p_pid, curproc->p_comm, type); 435 } else if (type != T_NMI && type != T_BPTFLT && 436 type != T_TRCTRAP) { 437 /* 438 * XXX not quite right, since this may be for a 439 * multiple fault in user mode. 440 */ 441 MAKEMPSAFE(have_mplock); 442 kprintf("kernel trap %d with interrupts disabled\n", 443 type); 444 } 445 cpu_enable_intr(); 446 } 447 448 type = frame->tf_trapno; 449 code = frame->tf_err; 450 451 if (ISPL(frame->tf_cs) == SEL_UPL) { 452 /* user trap */ 453 454 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid, 455 frame->tf_trapno, eva); 456 457 userenter(td, p); 458 459 sticks = (int)td->td_sticks; 460 KASSERT(lp->lwp_md.md_regs == frame, 461 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame)); 462 463 switch (type) { 464 case T_PRIVINFLT: /* privileged instruction fault */ 465 i = SIGILL; 466 ucode = ILL_PRVOPC; 467 break; 468 469 case T_BPTFLT: /* bpt instruction fault */ 470 case T_TRCTRAP: /* trace trap */ 471 frame->tf_rflags &= ~PSL_T; 472 i = SIGTRAP; 473 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT); 474 break; 475 476 case T_ARITHTRAP: /* arithmetic trap */ 477 ucode = code; 478 i = SIGFPE; 479 break; 480 481 case T_ASTFLT: /* Allow process switch */ 482 mycpu->gd_cnt.v_soft++; 483 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) { 484 atomic_clear_int(&mycpu->gd_reqflags, 485 RQF_AST_OWEUPC); 486 addupc_task(p, p->p_prof.pr_addr, 487 p->p_prof.pr_ticks); 488 } 489 goto out; 490 491 case T_PROTFLT: /* general protection fault */ 492 i = SIGBUS; 493 ucode = BUS_OBJERR; 494 break; 495 case T_STKFLT: /* stack fault */ 496 case T_SEGNPFLT: /* segment not present fault */ 497 i = SIGBUS; 498 ucode = BUS_ADRERR; 499 break; 500 case T_TSSFLT: /* invalid TSS fault */ 501 case T_DOUBLEFLT: /* double fault */ 502 default: 503 i = SIGBUS; 504 ucode = BUS_OBJERR; 505 break; 506 507 case T_PAGEFLT: /* page fault */ 508 i = trap_pfault(frame, TRUE); 509 if (frame->tf_rip == 0) { 510 kprintf("T_PAGEFLT: Warning %%rip == 0!\n"); 511 #ifdef DDB 512 while (freeze_on_seg_fault) 513 tsleep(p, 0, "freeze", hz * 20); 514 #endif 515 } 516 if (i == -1 || i == 0) 517 goto out; 518 519 520 if (i == SIGSEGV) 521 ucode = SEGV_MAPERR; 522 else { 523 i = SIGSEGV; 524 ucode = SEGV_ACCERR; 525 } 526 break; 527 528 case T_DIVIDE: /* integer divide fault */ 529 ucode = FPE_INTDIV; 530 i = SIGFPE; 531 break; 532 533 #if NISA > 0 534 case T_NMI: 535 MAKEMPSAFE(have_mplock); 536 /* machine/parity/power fail/"kitchen sink" faults */ 537 if (isa_nmi(code) == 0) { 538 #ifdef DDB 539 /* 540 * NMI can be hooked up to a pushbutton 541 * for debugging. 542 */ 543 if (ddb_on_nmi) { 544 kprintf ("NMI ... going to debugger\n"); 545 kdb_trap(type, 0, frame); 546 } 547 #endif /* DDB */ 548 goto out2; 549 } else if (panic_on_nmi) 550 panic("NMI indicates hardware failure"); 551 break; 552 #endif /* NISA > 0 */ 553 554 case T_OFLOW: /* integer overflow fault */ 555 ucode = FPE_INTOVF; 556 i = SIGFPE; 557 break; 558 559 case T_BOUND: /* bounds check fault */ 560 ucode = FPE_FLTSUB; 561 i = SIGFPE; 562 break; 563 564 case T_DNA: 565 /* 566 * Virtual kernel intercept - pass the DNA exception 567 * to the virtual kernel if it asked to handle it. 568 * This occurs when the virtual kernel is holding 569 * onto the FP context for a different emulated 570 * process then the one currently running. 571 * 572 * We must still call npxdna() since we may have 573 * saved FP state that the virtual kernel needs 574 * to hand over to a different emulated process. 575 */ 576 if (lp->lwp_vkernel && lp->lwp_vkernel->ve && 577 (td->td_pcb->pcb_flags & FP_VIRTFP) 578 ) { 579 npxdna(); 580 break; 581 } 582 583 /* 584 * The kernel may have switched out the FP unit's 585 * state, causing the user process to take a fault 586 * when it tries to use the FP unit. Restore the 587 * state here 588 */ 589 if (npxdna()) 590 goto out; 591 i = SIGFPE; 592 ucode = FPE_FPU_NP_TRAP; 593 break; 594 595 case T_FPOPFLT: /* FPU operand fetch fault */ 596 ucode = ILL_COPROC; 597 i = SIGILL; 598 break; 599 600 case T_XMMFLT: /* SIMD floating-point exception */ 601 ucode = 0; /* XXX */ 602 i = SIGFPE; 603 break; 604 } 605 } else { 606 /* kernel trap */ 607 608 switch (type) { 609 case T_PAGEFLT: /* page fault */ 610 trap_pfault(frame, FALSE); 611 goto out2; 612 613 case T_DNA: 614 /* 615 * The kernel is apparently using fpu for copying. 616 * XXX this should be fatal unless the kernel has 617 * registered such use. 618 */ 619 if (npxdna()) 620 goto out2; 621 break; 622 623 case T_STKFLT: /* stack fault */ 624 break; 625 626 case T_PROTFLT: /* general protection fault */ 627 case T_SEGNPFLT: /* segment not present fault */ 628 /* 629 * Invalid segment selectors and out of bounds 630 * %rip's and %rsp's can be set up in user mode. 631 * This causes a fault in kernel mode when the 632 * kernel tries to return to user mode. We want 633 * to get this fault so that we can fix the 634 * problem here and not have to check all the 635 * selectors and pointers when the user changes 636 * them. 637 */ 638 if (mycpu->gd_intr_nesting_level == 0) { 639 /* 640 * NOTE: in 64-bit mode traps push rsp/ss 641 * even if no ring change occurs. 642 */ 643 if (td->td_pcb->pcb_onfault && 644 td->td_pcb->pcb_onfault_sp == 645 frame->tf_rsp) { 646 frame->tf_rip = (register_t) 647 td->td_pcb->pcb_onfault; 648 goto out2; 649 } 650 if (frame->tf_rip == (long)doreti_iret) { 651 frame->tf_rip = (long)doreti_iret_fault; 652 goto out2; 653 } 654 } 655 break; 656 657 case T_TSSFLT: 658 /* 659 * PSL_NT can be set in user mode and isn't cleared 660 * automatically when the kernel is entered. This 661 * causes a TSS fault when the kernel attempts to 662 * `iret' because the TSS link is uninitialized. We 663 * want to get this fault so that we can fix the 664 * problem here and not every time the kernel is 665 * entered. 666 */ 667 if (frame->tf_rflags & PSL_NT) { 668 frame->tf_rflags &= ~PSL_NT; 669 goto out2; 670 } 671 break; 672 673 case T_TRCTRAP: /* trace trap */ 674 #if 0 675 if (frame->tf_rip == (int)IDTVEC(syscall)) { 676 /* 677 * We've just entered system mode via the 678 * syscall lcall. Continue single stepping 679 * silently until the syscall handler has 680 * saved the flags. 681 */ 682 goto out2; 683 } 684 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) { 685 /* 686 * The syscall handler has now saved the 687 * flags. Stop single stepping it. 688 */ 689 frame->tf_rflags &= ~PSL_T; 690 goto out2; 691 } 692 #endif 693 694 /* 695 * Ignore debug register trace traps due to 696 * accesses in the user's address space, which 697 * can happen under several conditions such as 698 * if a user sets a watchpoint on a buffer and 699 * then passes that buffer to a system call. 700 * We still want to get TRCTRAPS for addresses 701 * in kernel space because that is useful when 702 * debugging the kernel. 703 */ 704 #if JG 705 if (user_dbreg_trap()) { 706 /* 707 * Reset breakpoint bits because the 708 * processor doesn't 709 */ 710 /* XXX check upper bits here */ 711 load_dr6(rdr6() & 0xfffffff0); 712 goto out2; 713 } 714 #endif 715 /* 716 * FALLTHROUGH (TRCTRAP kernel mode, kernel address) 717 */ 718 case T_BPTFLT: 719 /* 720 * If DDB is enabled, let it handle the debugger trap. 721 * Otherwise, debugger traps "can't happen". 722 */ 723 ucode = TRAP_BRKPT; 724 #ifdef DDB 725 MAKEMPSAFE(have_mplock); 726 if (kdb_trap(type, 0, frame)) 727 goto out2; 728 #endif 729 break; 730 731 #if NISA > 0 732 case T_NMI: 733 MAKEMPSAFE(have_mplock); 734 /* machine/parity/power fail/"kitchen sink" faults */ 735 if (isa_nmi(code) == 0) { 736 #ifdef DDB 737 /* 738 * NMI can be hooked up to a pushbutton 739 * for debugging. 740 */ 741 if (ddb_on_nmi) { 742 kprintf ("NMI ... going to debugger\n"); 743 kdb_trap(type, 0, frame); 744 } 745 #endif /* DDB */ 746 goto out2; 747 } else if (panic_on_nmi == 0) 748 goto out2; 749 /* FALL THROUGH */ 750 #endif /* NISA > 0 */ 751 } 752 MAKEMPSAFE(have_mplock); 753 trap_fatal(frame, 0); 754 goto out2; 755 } 756 757 /* 758 * Virtual kernel intercept - if the fault is directly related to a 759 * VM context managed by a virtual kernel then let the virtual kernel 760 * handle it. 761 */ 762 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 763 vkernel_trap(lp, frame); 764 goto out; 765 } 766 767 /* Translate fault for emulators (e.g. Linux) */ 768 if (*p->p_sysent->sv_transtrap) 769 i = (*p->p_sysent->sv_transtrap)(i, type); 770 771 MAKEMPSAFE(have_mplock); 772 trapsignal(lp, i, ucode); 773 774 #ifdef DEBUG 775 if (type <= MAX_TRAP_MSG) { 776 uprintf("fatal process exception: %s", 777 trap_msg[type]); 778 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 779 uprintf(", fault VA = 0x%lx", frame->tf_addr); 780 uprintf("\n"); 781 } 782 #endif 783 784 out: 785 userret(lp, frame, sticks); 786 userexit(lp); 787 out2: ; 788 if (have_mplock) 789 rel_mplock(); 790 if (p != NULL && lp != NULL) 791 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid); 792 #ifdef INVARIANTS 793 KASSERT(crit_count == td->td_critcount, 794 ("trap: critical section count mismatch! %d/%d", 795 crit_count, td->td_pri)); 796 KASSERT(curstop == td->td_toks_stop, 797 ("trap: extra tokens held after trap! %ld/%ld", 798 curstop - &td->td_toks_base, 799 td->td_toks_stop - &td->td_toks_base)); 800 #endif 801 } 802 803 void 804 trap_handle_userenter(struct thread *td) 805 { 806 userenter(td, td->td_proc); 807 } 808 809 void 810 trap_handle_userexit(struct trapframe *frame, int sticks) 811 { 812 struct lwp *lp = curthread->td_lwp; 813 814 if (lp) { 815 userret(lp, frame, sticks); 816 userexit(lp); 817 } 818 } 819 820 static int 821 trap_pfault(struct trapframe *frame, int usermode) 822 { 823 vm_offset_t va; 824 struct vmspace *vm = NULL; 825 vm_map_t map; 826 int rv = 0; 827 int fault_flags; 828 vm_prot_t ftype; 829 thread_t td = curthread; 830 struct lwp *lp = td->td_lwp; 831 struct proc *p; 832 833 va = trunc_page(frame->tf_addr); 834 if (va >= VM_MIN_KERNEL_ADDRESS) { 835 /* 836 * Don't allow user-mode faults in kernel address space. 837 */ 838 if (usermode) { 839 fault_flags = -1; 840 ftype = -1; 841 goto nogo; 842 } 843 844 map = &kernel_map; 845 } else { 846 /* 847 * This is a fault on non-kernel virtual memory. 848 * vm is initialized above to NULL. If curproc is NULL 849 * or curproc->p_vmspace is NULL the fault is fatal. 850 */ 851 if (lp != NULL) 852 vm = lp->lwp_vmspace; 853 854 if (vm == NULL) { 855 fault_flags = -1; 856 ftype = -1; 857 goto nogo; 858 } 859 860 /* 861 * Debugging, try to catch kernel faults on the user address space when not inside 862 * on onfault (e.g. copyin/copyout) routine. 863 */ 864 if (usermode == 0 && (td->td_pcb == NULL || 865 td->td_pcb->pcb_onfault == NULL)) { 866 #ifdef DDB 867 if (freeze_on_seg_fault) { 868 kprintf("trap_pfault: user address fault from kernel mode " 869 "%016lx\n", (long)frame->tf_addr); 870 while (freeze_on_seg_fault) 871 tsleep(&freeze_on_seg_fault, 0, "frzseg", hz * 20); 872 } 873 #endif 874 } 875 map = &vm->vm_map; 876 } 877 878 /* 879 * PGEX_I is defined only if the execute disable bit capability is 880 * supported and enabled. 881 */ 882 if (frame->tf_err & PGEX_W) 883 ftype = VM_PROT_WRITE; 884 #if JG 885 else if ((frame->tf_err & PGEX_I) && pg_nx != 0) 886 ftype = VM_PROT_EXECUTE; 887 #endif 888 else 889 ftype = VM_PROT_READ; 890 891 if (map != &kernel_map) { 892 /* 893 * Keep swapout from messing with us during this 894 * critical time. 895 */ 896 PHOLD(lp->lwp_proc); 897 898 /* 899 * Issue fault 900 */ 901 fault_flags = 0; 902 if (usermode) 903 fault_flags |= VM_FAULT_BURST; 904 if (ftype & VM_PROT_WRITE) 905 fault_flags |= VM_FAULT_DIRTY; 906 else 907 fault_flags |= VM_FAULT_NORMAL; 908 rv = vm_fault(map, va, ftype, fault_flags); 909 910 PRELE(lp->lwp_proc); 911 } else { 912 /* 913 * Don't have to worry about process locking or stacks in the 914 * kernel. 915 */ 916 fault_flags = VM_FAULT_NORMAL; 917 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 918 } 919 if (rv == KERN_SUCCESS) 920 return (0); 921 nogo: 922 if (!usermode) { 923 /* 924 * NOTE: in 64-bit mode traps push rsp/ss 925 * even if no ring change occurs. 926 */ 927 if (td->td_pcb->pcb_onfault && 928 td->td_pcb->pcb_onfault_sp == frame->tf_rsp && 929 td->td_gd->gd_intr_nesting_level == 0) { 930 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault; 931 return (0); 932 } 933 trap_fatal(frame, frame->tf_addr); 934 return (-1); 935 } 936 937 /* 938 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no 939 * kludge is needed to pass the fault address to signal handlers. 940 */ 941 p = td->td_proc; 942 if (td->td_lwp->lwp_vkernel == NULL) { 943 #ifdef DDB 944 if (bootverbose || freeze_on_seg_fault || ddb_on_seg_fault) { 945 #else 946 if (bootverbose) { 947 #endif 948 kprintf("seg-fault ft=%04x ff=%04x addr=%p rip=%p " 949 "pid=%d cpu=%d p_comm=%s\n", 950 ftype, fault_flags, 951 (void *)frame->tf_addr, 952 (void *)frame->tf_rip, 953 p->p_pid, mycpu->gd_cpuid, p->p_comm); 954 } 955 #ifdef DDB 956 while (freeze_on_seg_fault) { 957 tsleep(p, 0, "freeze", hz * 20); 958 } 959 if (ddb_on_seg_fault) 960 Debugger("ddb_on_seg_fault"); 961 #endif 962 } 963 964 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 965 } 966 967 static void 968 trap_fatal(struct trapframe *frame, vm_offset_t eva) 969 { 970 int code, ss; 971 u_int type; 972 long rsp; 973 struct soft_segment_descriptor softseg; 974 char *msg; 975 976 code = frame->tf_err; 977 type = frame->tf_trapno; 978 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg); 979 980 if (type <= MAX_TRAP_MSG) 981 msg = trap_msg[type]; 982 else 983 msg = "UNKNOWN"; 984 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg, 985 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 986 /* three separate prints in case of a trap on an unmapped page */ 987 kprintf("cpuid = %d; ", mycpu->gd_cpuid); 988 kprintf("lapic->id = %08x\n", lapic->id); 989 if (type == T_PAGEFLT) { 990 kprintf("fault virtual address = 0x%lx\n", eva); 991 kprintf("fault code = %s %s %s, %s\n", 992 code & PGEX_U ? "user" : "supervisor", 993 code & PGEX_W ? "write" : "read", 994 code & PGEX_I ? "instruction" : "data", 995 code & PGEX_P ? "protection violation" : "page not present"); 996 } 997 kprintf("instruction pointer = 0x%lx:0x%lx\n", 998 frame->tf_cs & 0xffff, frame->tf_rip); 999 if (ISPL(frame->tf_cs) == SEL_UPL) { 1000 ss = frame->tf_ss & 0xffff; 1001 rsp = frame->tf_rsp; 1002 } else { 1003 /* 1004 * NOTE: in 64-bit mode traps push rsp/ss even if no ring 1005 * change occurs. 1006 */ 1007 ss = GSEL(GDATA_SEL, SEL_KPL); 1008 rsp = frame->tf_rsp; 1009 } 1010 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp); 1011 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp); 1012 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n", 1013 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); 1014 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n", 1015 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32, 1016 softseg.ssd_gran); 1017 kprintf("processor eflags = "); 1018 if (frame->tf_rflags & PSL_T) 1019 kprintf("trace trap, "); 1020 if (frame->tf_rflags & PSL_I) 1021 kprintf("interrupt enabled, "); 1022 if (frame->tf_rflags & PSL_NT) 1023 kprintf("nested task, "); 1024 if (frame->tf_rflags & PSL_RF) 1025 kprintf("resume, "); 1026 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12); 1027 kprintf("current process = "); 1028 if (curproc) { 1029 kprintf("%lu\n", 1030 (u_long)curproc->p_pid); 1031 } else { 1032 kprintf("Idle\n"); 1033 } 1034 kprintf("current thread = pri %d ", curthread->td_pri); 1035 if (curthread->td_critcount) 1036 kprintf("(CRIT)"); 1037 kprintf("\n"); 1038 1039 #ifdef DDB 1040 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame)) 1041 return; 1042 #endif 1043 kprintf("trap number = %d\n", type); 1044 if (type <= MAX_TRAP_MSG) 1045 panic("%s", trap_msg[type]); 1046 else 1047 panic("unknown/reserved trap"); 1048 } 1049 1050 /* 1051 * Double fault handler. Called when a fault occurs while writing 1052 * a frame for a trap/exception onto the stack. This usually occurs 1053 * when the stack overflows (such is the case with infinite recursion, 1054 * for example). 1055 */ 1056 static __inline 1057 int 1058 in_kstack_guard(register_t rptr) 1059 { 1060 thread_t td = curthread; 1061 1062 if ((char *)rptr >= td->td_kstack && 1063 (char *)rptr < td->td_kstack + PAGE_SIZE) { 1064 return 1; 1065 } 1066 return 0; 1067 } 1068 1069 void 1070 dblfault_handler(struct trapframe *frame) 1071 { 1072 thread_t td = curthread; 1073 1074 if (in_kstack_guard(frame->tf_rsp) || in_kstack_guard(frame->tf_rbp)) { 1075 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n"); 1076 if (in_kstack_guard(frame->tf_rsp)) 1077 frame->tf_rsp = (register_t)(td->td_kstack + PAGE_SIZE); 1078 if (in_kstack_guard(frame->tf_rbp)) 1079 frame->tf_rbp = (register_t)(td->td_kstack + PAGE_SIZE); 1080 } else { 1081 kprintf("DOUBLE FAULT\n"); 1082 } 1083 kprintf("\nFatal double fault\n"); 1084 kprintf("rip = 0x%lx\n", frame->tf_rip); 1085 kprintf("rsp = 0x%lx\n", frame->tf_rsp); 1086 kprintf("rbp = 0x%lx\n", frame->tf_rbp); 1087 /* three separate prints in case of a trap on an unmapped page */ 1088 kprintf("cpuid = %d; ", mycpu->gd_cpuid); 1089 kprintf("lapic->id = %08x\n", lapic->id); 1090 panic("double fault"); 1091 } 1092 1093 /* 1094 * syscall2 - MP aware system call request C handler 1095 * 1096 * A system call is essentially treated as a trap except that the 1097 * MP lock is not held on entry or return. We are responsible for 1098 * obtaining the MP lock if necessary and for handling ASTs 1099 * (e.g. a task switch) prior to return. 1100 * 1101 * MPSAFE 1102 */ 1103 void 1104 syscall2(struct trapframe *frame) 1105 { 1106 struct thread *td = curthread; 1107 struct proc *p = td->td_proc; 1108 struct lwp *lp = td->td_lwp; 1109 caddr_t params; 1110 struct sysent *callp; 1111 register_t orig_tf_rflags; 1112 int sticks; 1113 int error; 1114 int narg; 1115 #ifdef INVARIANTS 1116 int crit_count = td->td_critcount; 1117 #endif 1118 int have_mplock = 0; 1119 register_t *argp; 1120 u_int code; 1121 int reg, regcnt; 1122 union sysunion args; 1123 register_t *argsdst; 1124 1125 mycpu->gd_cnt.v_syscall++; 1126 1127 #ifdef DIAGNOSTIC 1128 if (ISPL(frame->tf_cs) != SEL_UPL) { 1129 get_mplock(); 1130 panic("syscall"); 1131 /* NOT REACHED */ 1132 } 1133 #endif 1134 1135 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid, 1136 frame->tf_rax); 1137 1138 userenter(td, p); /* lazy raise our priority */ 1139 1140 reg = 0; 1141 regcnt = 6; 1142 /* 1143 * Misc 1144 */ 1145 sticks = (int)td->td_sticks; 1146 orig_tf_rflags = frame->tf_rflags; 1147 1148 /* 1149 * Virtual kernel intercept - if a VM context managed by a virtual 1150 * kernel issues a system call the virtual kernel handles it, not us. 1151 * Restore the virtual kernel context and return from its system 1152 * call. The current frame is copied out to the virtual kernel. 1153 */ 1154 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 1155 vkernel_trap(lp, frame); 1156 error = EJUSTRETURN; 1157 goto out; 1158 } 1159 1160 /* 1161 * Get the system call parameters and account for time 1162 */ 1163 KASSERT(lp->lwp_md.md_regs == frame, 1164 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame)); 1165 params = (caddr_t)frame->tf_rsp + sizeof(register_t); 1166 code = frame->tf_rax; 1167 1168 if (p->p_sysent->sv_prepsyscall) { 1169 (*p->p_sysent->sv_prepsyscall)( 1170 frame, (int *)(&args.nosys.sysmsg + 1), 1171 &code, ¶ms); 1172 } else { 1173 if (code == SYS_syscall || code == SYS___syscall) { 1174 code = frame->tf_rdi; 1175 reg++; 1176 regcnt--; 1177 } 1178 } 1179 1180 if (p->p_sysent->sv_mask) 1181 code &= p->p_sysent->sv_mask; 1182 1183 if (code >= p->p_sysent->sv_size) 1184 callp = &p->p_sysent->sv_table[0]; 1185 else 1186 callp = &p->p_sysent->sv_table[code]; 1187 1188 narg = callp->sy_narg & SYF_ARGMASK; 1189 1190 /* 1191 * On x86_64 we get up to six arguments in registers. The rest are 1192 * on the stack. The first six members of 'struct trapframe' happen 1193 * to be the registers used to pass arguments, in exactly the right 1194 * order. 1195 */ 1196 argp = &frame->tf_rdi; 1197 argp += reg; 1198 argsdst = (register_t *)(&args.nosys.sysmsg + 1); 1199 /* 1200 * JG can we overflow the space pointed to by 'argsdst' 1201 * either with 'bcopy' or with 'copyin'? 1202 */ 1203 bcopy(argp, argsdst, sizeof(register_t) * regcnt); 1204 /* 1205 * copyin is MP aware, but the tracing code is not 1206 */ 1207 if (narg > regcnt) { 1208 KASSERT(params != NULL, ("copyin args with no params!")); 1209 error = copyin(params, &argsdst[regcnt], 1210 (narg - regcnt) * sizeof(register_t)); 1211 if (error) { 1212 #ifdef KTRACE 1213 if (KTRPOINT(td, KTR_SYSCALL)) { 1214 MAKEMPSAFE(have_mplock); 1215 1216 ktrsyscall(lp, code, narg, 1217 (void *)(&args.nosys.sysmsg + 1)); 1218 } 1219 #endif 1220 goto bad; 1221 } 1222 } 1223 1224 #ifdef KTRACE 1225 if (KTRPOINT(td, KTR_SYSCALL)) { 1226 MAKEMPSAFE(have_mplock); 1227 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1)); 1228 } 1229 #endif 1230 1231 /* 1232 * Default return value is 0 (will be copied to %rax). Double-value 1233 * returns use %rax and %rdx. %rdx is left unchanged for system 1234 * calls which return only one result. 1235 */ 1236 args.sysmsg_fds[0] = 0; 1237 args.sysmsg_fds[1] = frame->tf_rdx; 1238 1239 /* 1240 * The syscall might manipulate the trap frame. If it does it 1241 * will probably return EJUSTRETURN. 1242 */ 1243 args.sysmsg_frame = frame; 1244 1245 STOPEVENT(p, S_SCE, narg); /* MP aware */ 1246 1247 /* 1248 * NOTE: All system calls run MPSAFE now. The system call itself 1249 * is responsible for getting the MP lock. 1250 */ 1251 #ifdef SYSCALL_DEBUG 1252 uint64_t tscval = rdtsc(); 1253 #endif 1254 error = (*callp->sy_call)(&args); 1255 #ifdef SYSCALL_DEBUG 1256 tscval = rdtsc() - tscval; 1257 tscval = tscval * 1000000 / tsc_frequency; 1258 if (SysCallsWorstCase[code] < tscval) 1259 SysCallsWorstCase[code] = tscval; 1260 #endif 1261 1262 out: 1263 /* 1264 * MP SAFE (we may or may not have the MP lock at this point) 1265 */ 1266 //kprintf("SYSMSG %d ", error); 1267 switch (error) { 1268 case 0: 1269 /* 1270 * Reinitialize proc pointer `p' as it may be different 1271 * if this is a child returning from fork syscall. 1272 */ 1273 p = curproc; 1274 lp = curthread->td_lwp; 1275 frame->tf_rax = args.sysmsg_fds[0]; 1276 frame->tf_rdx = args.sysmsg_fds[1]; 1277 frame->tf_rflags &= ~PSL_C; 1278 break; 1279 case ERESTART: 1280 /* 1281 * Reconstruct pc, we know that 'syscall' is 2 bytes. 1282 * We have to do a full context restore so that %r10 1283 * (which was holding the value of %rcx) is restored for 1284 * the next iteration. 1285 */ 1286 if (frame->tf_err != 0 && frame->tf_err != 2) 1287 kprintf("lp %s:%d frame->tf_err is weird %ld\n", 1288 td->td_comm, lp->lwp_proc->p_pid, frame->tf_err); 1289 frame->tf_rip -= frame->tf_err; 1290 frame->tf_r10 = frame->tf_rcx; 1291 break; 1292 case EJUSTRETURN: 1293 break; 1294 case EASYNC: 1295 panic("Unexpected EASYNC return value (for now)"); 1296 default: 1297 bad: 1298 if (p->p_sysent->sv_errsize) { 1299 if (error >= p->p_sysent->sv_errsize) 1300 error = -1; /* XXX */ 1301 else 1302 error = p->p_sysent->sv_errtbl[error]; 1303 } 1304 frame->tf_rax = error; 1305 frame->tf_rflags |= PSL_C; 1306 break; 1307 } 1308 1309 /* 1310 * Traced syscall. trapsignal() is not MP aware. 1311 */ 1312 if (orig_tf_rflags & PSL_T) { 1313 MAKEMPSAFE(have_mplock); 1314 frame->tf_rflags &= ~PSL_T; 1315 trapsignal(lp, SIGTRAP, TRAP_TRACE); 1316 } 1317 1318 /* 1319 * Handle reschedule and other end-of-syscall issues 1320 */ 1321 userret(lp, frame, sticks); 1322 1323 #ifdef KTRACE 1324 if (KTRPOINT(td, KTR_SYSRET)) { 1325 MAKEMPSAFE(have_mplock); 1326 ktrsysret(lp, code, error, args.sysmsg_result); 1327 } 1328 #endif 1329 1330 /* 1331 * This works because errno is findable through the 1332 * register set. If we ever support an emulation where this 1333 * is not the case, this code will need to be revisited. 1334 */ 1335 STOPEVENT(p, S_SCX, code); 1336 1337 userexit(lp); 1338 /* 1339 * Release the MP lock if we had to get it 1340 */ 1341 if (have_mplock) 1342 rel_mplock(); 1343 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error); 1344 #ifdef INVARIANTS 1345 KASSERT(crit_count == td->td_critcount, 1346 ("syscall: critical section count mismatch! %d/%d", 1347 crit_count, td->td_pri)); 1348 KASSERT(&td->td_toks_base == td->td_toks_stop, 1349 ("syscall: extra tokens held after trap! %ld", 1350 td->td_toks_stop - &td->td_toks_base)); 1351 #endif 1352 } 1353 1354 /* 1355 * NOTE: mplock not held at any point 1356 */ 1357 void 1358 fork_return(struct lwp *lp, struct trapframe *frame) 1359 { 1360 frame->tf_rax = 0; /* Child returns zero */ 1361 frame->tf_rflags &= ~PSL_C; /* success */ 1362 frame->tf_rdx = 1; 1363 1364 generic_lwp_return(lp, frame); 1365 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid); 1366 } 1367 1368 /* 1369 * Simplified back end of syscall(), used when returning from fork() 1370 * directly into user mode. 1371 * 1372 * This code will return back into the fork trampoline code which then 1373 * runs doreti. 1374 * 1375 * NOTE: The mplock is not held at any point. 1376 */ 1377 void 1378 generic_lwp_return(struct lwp *lp, struct trapframe *frame) 1379 { 1380 struct proc *p = lp->lwp_proc; 1381 1382 /* 1383 * Newly forked processes are given a kernel priority. We have to 1384 * adjust the priority to a normal user priority and fake entry 1385 * into the kernel (call userenter()) to install a passive release 1386 * function just in case userret() decides to stop the process. This 1387 * can occur when ^Z races a fork. If we do not install the passive 1388 * release function the current process designation will not be 1389 * released when the thread goes to sleep. 1390 */ 1391 lwkt_setpri_self(TDPRI_USER_NORM); 1392 userenter(lp->lwp_thread, p); 1393 userret(lp, frame, 0); 1394 #ifdef KTRACE 1395 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET)) 1396 ktrsysret(lp, SYS_fork, 0, 0); 1397 #endif 1398 lp->lwp_flags |= LWP_PASSIVE_ACQ; 1399 userexit(lp); 1400 lp->lwp_flags &= ~LWP_PASSIVE_ACQ; 1401 } 1402 1403 /* 1404 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA 1405 * fault (which is then passed back to the virtual kernel) if an attempt is 1406 * made to use the FP unit. 1407 * 1408 * XXX this is a fairly big hack. 1409 */ 1410 void 1411 set_vkernel_fp(struct trapframe *frame) 1412 { 1413 struct thread *td = curthread; 1414 1415 if (frame->tf_xflags & PGEX_FPFAULT) { 1416 td->td_pcb->pcb_flags |= FP_VIRTFP; 1417 if (mdcpu->gd_npxthread == td) 1418 npxexit(); 1419 } else { 1420 td->td_pcb->pcb_flags &= ~FP_VIRTFP; 1421 } 1422 } 1423 1424 /* 1425 * Called from vkernel_trap() to fixup the vkernel's syscall 1426 * frame for vmspace_ctl() return. 1427 */ 1428 void 1429 cpu_vkernel_trap(struct trapframe *frame, int error) 1430 { 1431 frame->tf_rax = error; 1432 if (error) 1433 frame->tf_rflags |= PSL_C; 1434 else 1435 frame->tf_rflags &= ~PSL_C; 1436 } 1437