1 /*- 2 * Copyright (c) 1990, 1993 3 * The Regents of the University of California. All rights reserved. 4 * Copyright (C) 1994, David Greenman 5 * Copyright (c) 2008 The DragonFly Project. 6 * Copyright (c) 2008 Jordan Gordeev. 7 * 8 * This code is derived from software contributed to Berkeley by 9 * the University of Utah, and William Jolitz. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. All advertising materials mentioning features or use of this software 20 * must display the following acknowledgement: 21 * This product includes software developed by the University of 22 * California, Berkeley and its contributors. 23 * 4. Neither the name of the University nor the names of its contributors 24 * may be used to endorse or promote products derived from this software 25 * without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 37 * SUCH DAMAGE. 38 * 39 * from: @(#)trap.c 7.4 (Berkeley) 5/13/91 40 * $FreeBSD: src/sys/i386/i386/trap.c,v 1.147.2.11 2003/02/27 19:09:59 luoqi Exp $ 41 */ 42 43 /* 44 * x86_64 Trap and System call handling 45 */ 46 47 #include "use_isa.h" 48 49 #include "opt_ddb.h" 50 #include "opt_ktrace.h" 51 52 #include <machine/frame.h> 53 #include <sys/param.h> 54 #include <sys/systm.h> 55 #include <sys/kernel.h> 56 #include <sys/kerneldump.h> 57 #include <sys/proc.h> 58 #include <sys/pioctl.h> 59 #include <sys/types.h> 60 #include <sys/signal2.h> 61 #include <sys/syscall.h> 62 #include <sys/sysctl.h> 63 #include <sys/sysent.h> 64 #include <sys/systm.h> 65 #ifdef KTRACE 66 #include <sys/ktrace.h> 67 #endif 68 #include <sys/ktr.h> 69 #include <sys/sysmsg.h> 70 #include <sys/sysproto.h> 71 #include <sys/sysunion.h> 72 73 #include <vm/pmap.h> 74 #include <vm/vm.h> 75 #include <vm/vm_extern.h> 76 #include <vm/vm_kern.h> 77 #include <vm/vm_param.h> 78 #include <machine/cpu.h> 79 #include <machine/pcb.h> 80 #include <machine/smp.h> 81 #include <machine/thread.h> 82 #include <machine/clock.h> 83 #include <machine/vmparam.h> 84 #include <machine/md_var.h> 85 #include <machine_base/isa/isa_intr.h> 86 #include <machine_base/apic/lapic.h> 87 88 #include <ddb/ddb.h> 89 90 #include <sys/thread2.h> 91 #include <sys/mplock2.h> 92 93 #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 static __inline void 181 userenter(struct thread *curtd, struct proc *curp) 182 { 183 struct ucred *ocred; 184 struct ucred *ncred; 185 186 curtd->td_release = lwkt_passive_release; 187 188 if (curtd->td_ucred != curp->p_ucred) { 189 ncred = crhold(curp->p_ucred); 190 ocred = curtd->td_ucred; 191 curtd->td_ucred = ncred; 192 if (ocred) 193 crfree(ocred); 194 } 195 196 #ifdef DDB 197 /* 198 * Debugging, remove top two user stack pages to catch kernel faults 199 */ 200 if (freeze_on_seg_fault > 1 && curtd->td_lwp) { 201 pmap_remove(vmspace_pmap(curtd->td_lwp->lwp_vmspace), 202 0x00007FFFFFFFD000LU, 203 0x0000800000000000LU); 204 } 205 #endif 206 } 207 208 /* 209 * Handle signals, upcalls, profiling, and other AST's and/or tasks that 210 * must be completed before we can return to or try to return to userland. 211 * 212 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64 213 * arithmatic on the delta calculation so the absolute tick values are 214 * truncated to an integer. 215 */ 216 static void 217 userret(struct lwp *lp, struct trapframe *frame, int sticks) 218 { 219 struct proc *p = lp->lwp_proc; 220 int sig; 221 222 /* 223 * Charge system time if profiling. Note: times are in microseconds. 224 * This may do a copyout and block, so do it first even though it 225 * means some system time will be charged as user time. 226 */ 227 if (p->p_flags & P_PROFIL) { 228 addupc_task(p, frame->tf_rip, 229 (u_int)((int)lp->lwp_thread->td_sticks - sticks)); 230 } 231 232 recheck: 233 /* 234 * Specific on-return-to-usermode checks (LWP_MP_WEXIT, 235 * LWP_MP_VNLRU, etc). 236 */ 237 if (lp->lwp_mpflags & LWP_MP_URETMASK) 238 lwpuserret(lp); 239 240 /* 241 * Block here if we are in a stopped state. 242 */ 243 if (p->p_stat == SSTOP || dump_stop_usertds) { 244 lwkt_gettoken(&p->p_token); 245 tstop(); 246 lwkt_reltoken(&p->p_token); 247 goto recheck; 248 } 249 250 /* 251 * Post any pending upcalls. If running a virtual kernel be sure 252 * to restore the virtual kernel's vmspace before posting the upcall. 253 */ 254 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF)) { 255 lwkt_gettoken(&p->p_token); 256 if (p->p_flags & P_SIGVTALRM) { 257 p->p_flags &= ~P_SIGVTALRM; 258 ksignal(p, SIGVTALRM); 259 } 260 if (p->p_flags & P_SIGPROF) { 261 p->p_flags &= ~P_SIGPROF; 262 ksignal(p, SIGPROF); 263 } 264 lwkt_reltoken(&p->p_token); 265 goto recheck; 266 } 267 268 /* 269 * Post any pending signals. If running a virtual kernel be sure 270 * to restore the virtual kernel's vmspace before posting the signal. 271 * 272 * WARNING! postsig() can exit and not return. 273 */ 274 if ((sig = CURSIG_TRACE(lp)) != 0) { 275 lwkt_gettoken(&p->p_token); 276 postsig(sig); 277 lwkt_reltoken(&p->p_token); 278 goto recheck; 279 } 280 281 /* 282 * block here if we are swapped out, but still process signals 283 * (such as SIGKILL). proc0 (the swapin scheduler) is already 284 * aware of our situation, we do not have to wake it up. 285 */ 286 if (p->p_flags & P_SWAPPEDOUT) { 287 lwkt_gettoken(&p->p_token); 288 get_mplock(); 289 p->p_flags |= P_SWAPWAIT; 290 swapin_request(); 291 if (p->p_flags & P_SWAPWAIT) 292 tsleep(p, PCATCH, "SWOUT", 0); 293 p->p_flags &= ~P_SWAPWAIT; 294 rel_mplock(); 295 lwkt_reltoken(&p->p_token); 296 goto recheck; 297 } 298 299 /* 300 * In a multi-threaded program it is possible for a thread to change 301 * signal state during a system call which temporarily changes the 302 * signal mask. In this case postsig() might not be run and we 303 * have to restore the mask ourselves. 304 */ 305 if (lp->lwp_flags & LWP_OLDMASK) { 306 lp->lwp_flags &= ~LWP_OLDMASK; 307 lp->lwp_sigmask = lp->lwp_oldsigmask; 308 goto recheck; 309 } 310 } 311 312 /* 313 * Cleanup from userenter and any passive release that might have occured. 314 * We must reclaim the current-process designation before we can return 315 * to usermode. We also handle both LWKT and USER reschedule requests. 316 */ 317 static __inline void 318 userexit(struct lwp *lp) 319 { 320 struct thread *td = lp->lwp_thread; 321 /* globaldata_t gd = td->td_gd; */ 322 323 /* 324 * Handle stop requests at kernel priority. Any requests queued 325 * after this loop will generate another AST. 326 */ 327 while (lp->lwp_proc->p_stat == SSTOP) { 328 lwkt_gettoken(&lp->lwp_proc->p_token); 329 tstop(); 330 lwkt_reltoken(&lp->lwp_proc->p_token); 331 } 332 333 /* 334 * Reduce our priority in preparation for a return to userland. If 335 * our passive release function was still in place, our priority was 336 * never raised and does not need to be reduced. 337 */ 338 lwkt_passive_recover(td); 339 340 /* WARNING: we may have migrated cpu's */ 341 /* gd = td->td_gd; */ 342 343 /* 344 * Become the current user scheduled process if we aren't already, 345 * and deal with reschedule requests and other factors. 346 */ 347 lp->lwp_proc->p_usched->acquire_curproc(lp); 348 } 349 350 #if !defined(KTR_KERNENTRY) 351 #define KTR_KERNENTRY KTR_ALL 352 #endif 353 KTR_INFO_MASTER(kernentry); 354 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, 355 "TRAP(pid %d, tid %d, trapno %ld, eva %lu)", 356 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva); 357 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %d, tid %d)", 358 pid_t pid, lwpid_t tid); 359 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %d, tid %d, nr %ld)", 360 pid_t pid, lwpid_t tid, register_t trapno); 361 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %d, tid %d, err %d)", 362 pid_t pid, lwpid_t tid, int err); 363 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %d, tid %d)", 364 pid_t pid, lwpid_t tid); 365 366 /* 367 * Exception, fault, and trap interface to the kernel. 368 * This common code is called from assembly language IDT gate entry 369 * routines that prepare a suitable stack frame, and restore this 370 * frame after the exception has been processed. 371 * 372 * This function is also called from doreti in an interlock to handle ASTs. 373 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap 374 * 375 * NOTE! We have to retrieve the fault address prior to obtaining the 376 * MP lock because get_mplock() may switch out. YYY cr2 really ought 377 * to be retrieved by the assembly code, not here. 378 * 379 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing 380 * if an attempt is made to switch from a fast interrupt or IPI. This is 381 * necessary to properly take fatal kernel traps on SMP machines if 382 * get_mplock() has to block. 383 */ 384 385 void 386 trap(struct trapframe *frame) 387 { 388 struct globaldata *gd = mycpu; 389 struct thread *td = gd->gd_curthread; 390 struct lwp *lp = td->td_lwp; 391 struct proc *p; 392 int sticks = 0; 393 int i = 0, ucode = 0, type, code; 394 int have_mplock = 0; 395 #ifdef INVARIANTS 396 int crit_count = td->td_critcount; 397 lwkt_tokref_t curstop = td->td_toks_stop; 398 #endif 399 vm_offset_t eva; 400 401 p = td->td_proc; 402 clear_quickret(); 403 404 #ifdef DDB 405 /* 406 * We need to allow T_DNA faults when the debugger is active since 407 * some dumping paths do large bcopy() which use the floating 408 * point registers for faster copying. 409 */ 410 if (db_active && frame->tf_trapno != T_DNA) { 411 eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0); 412 ++gd->gd_trap_nesting_level; 413 MAKEMPSAFE(have_mplock); 414 trap_fatal(frame, eva); 415 --gd->gd_trap_nesting_level; 416 goto out2; 417 } 418 #endif 419 420 eva = 0; 421 422 if ((frame->tf_rflags & PSL_I) == 0) { 423 /* 424 * Buggy application or kernel code has disabled interrupts 425 * and then trapped. Enabling interrupts now is wrong, but 426 * it is better than running with interrupts disabled until 427 * they are accidentally enabled later. 428 */ 429 type = frame->tf_trapno; 430 if (ISPL(frame->tf_cs) == SEL_UPL) { 431 MAKEMPSAFE(have_mplock); 432 /* JG curproc can be NULL */ 433 kprintf( 434 "pid %ld (%s): trap %d with interrupts disabled\n", 435 (long)curproc->p_pid, curproc->p_comm, type); 436 } else if (type != T_NMI && type != T_BPTFLT && 437 type != T_TRCTRAP) { 438 /* 439 * XXX not quite right, since this may be for a 440 * multiple fault in user mode. 441 */ 442 MAKEMPSAFE(have_mplock); 443 kprintf("kernel trap %d with interrupts disabled\n", 444 type); 445 } 446 cpu_enable_intr(); 447 } 448 449 type = frame->tf_trapno; 450 code = frame->tf_err; 451 452 if (ISPL(frame->tf_cs) == SEL_UPL) { 453 /* user trap */ 454 455 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid, 456 frame->tf_trapno, eva); 457 458 userenter(td, p); 459 460 sticks = (int)td->td_sticks; 461 KASSERT(lp->lwp_md.md_regs == frame, 462 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame)); 463 464 switch (type) { 465 case T_PRIVINFLT: /* privileged instruction fault */ 466 i = SIGILL; 467 ucode = ILL_PRVOPC; 468 break; 469 470 case T_BPTFLT: /* bpt instruction fault */ 471 case T_TRCTRAP: /* trace trap */ 472 frame->tf_rflags &= ~PSL_T; 473 i = SIGTRAP; 474 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT); 475 break; 476 477 case T_ARITHTRAP: /* arithmetic trap */ 478 ucode = code; 479 i = SIGFPE; 480 break; 481 482 case T_ASTFLT: /* Allow process switch */ 483 mycpu->gd_cnt.v_soft++; 484 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) { 485 atomic_clear_int(&mycpu->gd_reqflags, 486 RQF_AST_OWEUPC); 487 addupc_task(p, p->p_prof.pr_addr, 488 p->p_prof.pr_ticks); 489 } 490 goto out; 491 492 case T_PROTFLT: /* general protection fault */ 493 i = SIGBUS; 494 ucode = BUS_OBJERR; 495 break; 496 case T_STKFLT: /* stack fault */ 497 case T_SEGNPFLT: /* segment not present fault */ 498 i = SIGBUS; 499 ucode = BUS_ADRERR; 500 break; 501 case T_TSSFLT: /* invalid TSS fault */ 502 case T_DOUBLEFLT: /* double fault */ 503 default: 504 i = SIGBUS; 505 ucode = BUS_OBJERR; 506 break; 507 508 case T_PAGEFLT: /* page fault */ 509 i = trap_pfault(frame, TRUE); 510 if (frame->tf_rip == 0) { 511 kprintf("T_PAGEFLT: Warning %%rip == 0!\n"); 512 #ifdef DDB 513 while (freeze_on_seg_fault) 514 tsleep(p, 0, "freeze", hz * 20); 515 #endif 516 } 517 if (i == -1 || i == 0) 518 goto out; 519 520 521 if (i == SIGSEGV) 522 ucode = SEGV_MAPERR; 523 else { 524 i = SIGSEGV; 525 ucode = SEGV_ACCERR; 526 } 527 break; 528 529 case T_DIVIDE: /* integer divide fault */ 530 ucode = FPE_INTDIV; 531 i = SIGFPE; 532 break; 533 534 #if NISA > 0 535 case T_NMI: 536 MAKEMPSAFE(have_mplock); 537 /* machine/parity/power fail/"kitchen sink" faults */ 538 if (isa_nmi(code) == 0) { 539 #ifdef DDB 540 /* 541 * NMI can be hooked up to a pushbutton 542 * for debugging. 543 */ 544 if (ddb_on_nmi) { 545 kprintf ("NMI ... going to debugger\n"); 546 kdb_trap(type, 0, frame); 547 } 548 #endif /* DDB */ 549 goto out2; 550 } else if (panic_on_nmi) 551 panic("NMI indicates hardware failure"); 552 break; 553 #endif /* NISA > 0 */ 554 555 case T_OFLOW: /* integer overflow fault */ 556 ucode = FPE_INTOVF; 557 i = SIGFPE; 558 break; 559 560 case T_BOUND: /* bounds check fault */ 561 ucode = FPE_FLTSUB; 562 i = SIGFPE; 563 break; 564 565 case T_DNA: 566 /* 567 * Virtual kernel intercept - pass the DNA exception 568 * to the virtual kernel if it asked to handle it. 569 * This occurs when the virtual kernel is holding 570 * onto the FP context for a different emulated 571 * process then the one currently running. 572 * 573 * We must still call npxdna() since we may have 574 * saved FP state that the virtual kernel needs 575 * to hand over to a different emulated process. 576 */ 577 if (lp->lwp_vkernel && lp->lwp_vkernel->ve && 578 (td->td_pcb->pcb_flags & FP_VIRTFP) 579 ) { 580 npxdna(); 581 break; 582 } 583 584 /* 585 * The kernel may have switched out the FP unit's 586 * state, causing the user process to take a fault 587 * when it tries to use the FP unit. Restore the 588 * state here 589 */ 590 if (npxdna()) 591 goto out; 592 i = SIGFPE; 593 ucode = FPE_FPU_NP_TRAP; 594 break; 595 596 case T_FPOPFLT: /* FPU operand fetch fault */ 597 ucode = ILL_COPROC; 598 i = SIGILL; 599 break; 600 601 case T_XMMFLT: /* SIMD floating-point exception */ 602 ucode = 0; /* XXX */ 603 i = SIGFPE; 604 break; 605 } 606 } else { 607 /* kernel trap */ 608 609 switch (type) { 610 case T_PAGEFLT: /* page fault */ 611 trap_pfault(frame, FALSE); 612 goto out2; 613 614 case T_DNA: 615 /* 616 * The kernel is apparently using fpu for copying. 617 * XXX this should be fatal unless the kernel has 618 * registered such use. 619 */ 620 if (npxdna()) 621 goto out2; 622 break; 623 624 case T_STKFLT: /* stack fault */ 625 break; 626 627 case T_PROTFLT: /* general protection fault */ 628 case T_SEGNPFLT: /* segment not present fault */ 629 /* 630 * Invalid segment selectors and out of bounds 631 * %rip's and %rsp's can be set up in user mode. 632 * This causes a fault in kernel mode when the 633 * kernel tries to return to user mode. We want 634 * to get this fault so that we can fix the 635 * problem here and not have to check all the 636 * selectors and pointers when the user changes 637 * them. 638 */ 639 if (mycpu->gd_intr_nesting_level == 0) { 640 /* 641 * NOTE: in 64-bit mode traps push rsp/ss 642 * even if no ring change occurs. 643 */ 644 if (td->td_pcb->pcb_onfault && 645 td->td_pcb->pcb_onfault_sp == 646 frame->tf_rsp) { 647 frame->tf_rip = (register_t) 648 td->td_pcb->pcb_onfault; 649 goto out2; 650 } 651 if (frame->tf_rip == (long)doreti_iret) { 652 frame->tf_rip = (long)doreti_iret_fault; 653 goto out2; 654 } 655 } 656 break; 657 658 case T_TSSFLT: 659 /* 660 * PSL_NT can be set in user mode and isn't cleared 661 * automatically when the kernel is entered. This 662 * causes a TSS fault when the kernel attempts to 663 * `iret' because the TSS link is uninitialized. We 664 * want to get this fault so that we can fix the 665 * problem here and not every time the kernel is 666 * entered. 667 */ 668 if (frame->tf_rflags & PSL_NT) { 669 frame->tf_rflags &= ~PSL_NT; 670 goto out2; 671 } 672 break; 673 674 case T_TRCTRAP: /* trace trap */ 675 #if 0 676 if (frame->tf_rip == (int)IDTVEC(syscall)) { 677 /* 678 * We've just entered system mode via the 679 * syscall lcall. Continue single stepping 680 * silently until the syscall handler has 681 * saved the flags. 682 */ 683 goto out2; 684 } 685 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) { 686 /* 687 * The syscall handler has now saved the 688 * flags. Stop single stepping it. 689 */ 690 frame->tf_rflags &= ~PSL_T; 691 goto out2; 692 } 693 #endif 694 695 /* 696 * Ignore debug register trace traps due to 697 * accesses in the user's address space, which 698 * can happen under several conditions such as 699 * if a user sets a watchpoint on a buffer and 700 * then passes that buffer to a system call. 701 * We still want to get TRCTRAPS for addresses 702 * in kernel space because that is useful when 703 * debugging the kernel. 704 */ 705 #if JG 706 if (user_dbreg_trap()) { 707 /* 708 * Reset breakpoint bits because the 709 * processor doesn't 710 */ 711 /* XXX check upper bits here */ 712 load_dr6(rdr6() & 0xfffffff0); 713 goto out2; 714 } 715 #endif 716 /* 717 * FALLTHROUGH (TRCTRAP kernel mode, kernel address) 718 */ 719 case T_BPTFLT: 720 /* 721 * If DDB is enabled, let it handle the debugger trap. 722 * Otherwise, debugger traps "can't happen". 723 */ 724 ucode = TRAP_BRKPT; 725 #ifdef DDB 726 MAKEMPSAFE(have_mplock); 727 if (kdb_trap(type, 0, frame)) 728 goto out2; 729 #endif 730 break; 731 732 #if NISA > 0 733 case T_NMI: 734 MAKEMPSAFE(have_mplock); 735 /* machine/parity/power fail/"kitchen sink" faults */ 736 if (isa_nmi(code) == 0) { 737 #ifdef DDB 738 /* 739 * NMI can be hooked up to a pushbutton 740 * for debugging. 741 */ 742 if (ddb_on_nmi) { 743 kprintf ("NMI ... going to debugger\n"); 744 kdb_trap(type, 0, frame); 745 } 746 #endif /* DDB */ 747 goto out2; 748 } else if (panic_on_nmi == 0) 749 goto out2; 750 /* FALL THROUGH */ 751 #endif /* NISA > 0 */ 752 } 753 MAKEMPSAFE(have_mplock); 754 trap_fatal(frame, 0); 755 goto out2; 756 } 757 758 /* 759 * Virtual kernel intercept - if the fault is directly related to a 760 * VM context managed by a virtual kernel then let the virtual kernel 761 * handle it. 762 */ 763 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 764 vkernel_trap(lp, frame); 765 goto out; 766 } 767 768 /* Translate fault for emulators (e.g. Linux) */ 769 if (*p->p_sysent->sv_transtrap) 770 i = (*p->p_sysent->sv_transtrap)(i, type); 771 772 MAKEMPSAFE(have_mplock); 773 trapsignal(lp, i, ucode); 774 775 #ifdef DEBUG 776 if (type <= MAX_TRAP_MSG) { 777 uprintf("fatal process exception: %s", 778 trap_msg[type]); 779 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 780 uprintf(", fault VA = 0x%lx", frame->tf_addr); 781 uprintf("\n"); 782 } 783 #endif 784 785 out: 786 userret(lp, frame, sticks); 787 userexit(lp); 788 out2: ; 789 if (have_mplock) 790 rel_mplock(); 791 if (p != NULL && lp != NULL) 792 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid); 793 #ifdef INVARIANTS 794 KASSERT(crit_count == td->td_critcount, 795 ("trap: critical section count mismatch! %d/%d", 796 crit_count, td->td_pri)); 797 KASSERT(curstop == td->td_toks_stop, 798 ("trap: extra tokens held after trap! %ld/%ld", 799 curstop - &td->td_toks_base, 800 td->td_toks_stop - &td->td_toks_base)); 801 #endif 802 } 803 804 static int 805 trap_pfault(struct trapframe *frame, int usermode) 806 { 807 vm_offset_t va; 808 struct vmspace *vm = NULL; 809 vm_map_t map; 810 int rv = 0; 811 int fault_flags; 812 vm_prot_t ftype; 813 thread_t td = curthread; 814 struct lwp *lp = td->td_lwp; 815 struct proc *p; 816 817 va = trunc_page(frame->tf_addr); 818 if (va >= VM_MIN_KERNEL_ADDRESS) { 819 /* 820 * Don't allow user-mode faults in kernel address space. 821 */ 822 if (usermode) { 823 fault_flags = -1; 824 ftype = -1; 825 goto nogo; 826 } 827 828 map = &kernel_map; 829 } else { 830 /* 831 * This is a fault on non-kernel virtual memory. 832 * vm is initialized above to NULL. If curproc is NULL 833 * or curproc->p_vmspace is NULL the fault is fatal. 834 */ 835 if (lp != NULL) 836 vm = lp->lwp_vmspace; 837 838 if (vm == NULL) { 839 fault_flags = -1; 840 ftype = -1; 841 goto nogo; 842 } 843 844 /* 845 * Debugging, try to catch kernel faults on the user address space when not inside 846 * on onfault (e.g. copyin/copyout) routine. 847 */ 848 if (usermode == 0 && (td->td_pcb == NULL || 849 td->td_pcb->pcb_onfault == NULL)) { 850 #ifdef DDB 851 if (freeze_on_seg_fault) { 852 kprintf("trap_pfault: user address fault from kernel mode " 853 "%016lx\n", (long)frame->tf_addr); 854 while (freeze_on_seg_fault) 855 tsleep(&freeze_on_seg_fault, 0, "frzseg", hz * 20); 856 } 857 #endif 858 } 859 map = &vm->vm_map; 860 } 861 862 /* 863 * PGEX_I is defined only if the execute disable bit capability is 864 * supported and enabled. 865 */ 866 if (frame->tf_err & PGEX_W) 867 ftype = VM_PROT_WRITE; 868 #if JG 869 else if ((frame->tf_err & PGEX_I) && pg_nx != 0) 870 ftype = VM_PROT_EXECUTE; 871 #endif 872 else 873 ftype = VM_PROT_READ; 874 875 if (map != &kernel_map) { 876 /* 877 * Keep swapout from messing with us during this 878 * critical time. 879 */ 880 PHOLD(lp->lwp_proc); 881 882 /* 883 * Issue fault 884 */ 885 fault_flags = 0; 886 if (usermode) 887 fault_flags |= VM_FAULT_BURST; 888 if (ftype & VM_PROT_WRITE) 889 fault_flags |= VM_FAULT_DIRTY; 890 else 891 fault_flags |= VM_FAULT_NORMAL; 892 rv = vm_fault(map, va, ftype, fault_flags); 893 894 PRELE(lp->lwp_proc); 895 } else { 896 /* 897 * Don't have to worry about process locking or stacks in the 898 * kernel. 899 */ 900 fault_flags = VM_FAULT_NORMAL; 901 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 902 } 903 if (rv == KERN_SUCCESS) 904 return (0); 905 nogo: 906 if (!usermode) { 907 /* 908 * NOTE: in 64-bit mode traps push rsp/ss 909 * even if no ring change occurs. 910 */ 911 if (td->td_pcb->pcb_onfault && 912 td->td_pcb->pcb_onfault_sp == frame->tf_rsp && 913 td->td_gd->gd_intr_nesting_level == 0) { 914 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault; 915 return (0); 916 } 917 trap_fatal(frame, frame->tf_addr); 918 return (-1); 919 } 920 921 /* 922 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no 923 * kludge is needed to pass the fault address to signal handlers. 924 */ 925 p = td->td_proc; 926 if (td->td_lwp->lwp_vkernel == NULL) { 927 #ifdef DDB 928 if (bootverbose || freeze_on_seg_fault || ddb_on_seg_fault) { 929 #else 930 if (bootverbose) { 931 #endif 932 kprintf("seg-fault ft=%04x ff=%04x addr=%p rip=%p " 933 "pid=%d cpu=%d p_comm=%s\n", 934 ftype, fault_flags, 935 (void *)frame->tf_addr, 936 (void *)frame->tf_rip, 937 p->p_pid, mycpu->gd_cpuid, p->p_comm); 938 } 939 #ifdef DDB 940 while (freeze_on_seg_fault) { 941 tsleep(p, 0, "freeze", hz * 20); 942 } 943 if (ddb_on_seg_fault) 944 Debugger("ddb_on_seg_fault"); 945 #endif 946 } 947 948 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 949 } 950 951 static void 952 trap_fatal(struct trapframe *frame, vm_offset_t eva) 953 { 954 int code, ss; 955 u_int type; 956 long rsp; 957 struct soft_segment_descriptor softseg; 958 char *msg; 959 960 code = frame->tf_err; 961 type = frame->tf_trapno; 962 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg); 963 964 if (type <= MAX_TRAP_MSG) 965 msg = trap_msg[type]; 966 else 967 msg = "UNKNOWN"; 968 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg, 969 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel"); 970 /* three separate prints in case of a trap on an unmapped page */ 971 kprintf("cpuid = %d; ", mycpu->gd_cpuid); 972 kprintf("lapic->id = %08x\n", lapic->id); 973 if (type == T_PAGEFLT) { 974 kprintf("fault virtual address = 0x%lx\n", eva); 975 kprintf("fault code = %s %s %s, %s\n", 976 code & PGEX_U ? "user" : "supervisor", 977 code & PGEX_W ? "write" : "read", 978 code & PGEX_I ? "instruction" : "data", 979 code & PGEX_P ? "protection violation" : "page not present"); 980 } 981 kprintf("instruction pointer = 0x%lx:0x%lx\n", 982 frame->tf_cs & 0xffff, frame->tf_rip); 983 if (ISPL(frame->tf_cs) == SEL_UPL) { 984 ss = frame->tf_ss & 0xffff; 985 rsp = frame->tf_rsp; 986 } else { 987 /* 988 * NOTE: in 64-bit mode traps push rsp/ss even if no ring 989 * change occurs. 990 */ 991 ss = GSEL(GDATA_SEL, SEL_KPL); 992 rsp = frame->tf_rsp; 993 } 994 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp); 995 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp); 996 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n", 997 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type); 998 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n", 999 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32, 1000 softseg.ssd_gran); 1001 kprintf("processor eflags = "); 1002 if (frame->tf_rflags & PSL_T) 1003 kprintf("trace trap, "); 1004 if (frame->tf_rflags & PSL_I) 1005 kprintf("interrupt enabled, "); 1006 if (frame->tf_rflags & PSL_NT) 1007 kprintf("nested task, "); 1008 if (frame->tf_rflags & PSL_RF) 1009 kprintf("resume, "); 1010 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12); 1011 kprintf("current process = "); 1012 if (curproc) { 1013 kprintf("%lu\n", 1014 (u_long)curproc->p_pid); 1015 } else { 1016 kprintf("Idle\n"); 1017 } 1018 kprintf("current thread = pri %d ", curthread->td_pri); 1019 if (curthread->td_critcount) 1020 kprintf("(CRIT)"); 1021 kprintf("\n"); 1022 1023 #ifdef DDB 1024 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame)) 1025 return; 1026 #endif 1027 kprintf("trap number = %d\n", type); 1028 if (type <= MAX_TRAP_MSG) 1029 panic("%s", trap_msg[type]); 1030 else 1031 panic("unknown/reserved trap"); 1032 } 1033 1034 /* 1035 * Double fault handler. Called when a fault occurs while writing 1036 * a frame for a trap/exception onto the stack. This usually occurs 1037 * when the stack overflows (such is the case with infinite recursion, 1038 * for example). 1039 */ 1040 static __inline 1041 int 1042 in_kstack_guard(register_t rptr) 1043 { 1044 thread_t td = curthread; 1045 1046 if ((char *)rptr >= td->td_kstack && 1047 (char *)rptr < td->td_kstack + PAGE_SIZE) { 1048 return 1; 1049 } 1050 return 0; 1051 } 1052 1053 void 1054 dblfault_handler(struct trapframe *frame) 1055 { 1056 thread_t td = curthread; 1057 1058 if (in_kstack_guard(frame->tf_rsp) || in_kstack_guard(frame->tf_rbp)) { 1059 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n"); 1060 if (in_kstack_guard(frame->tf_rsp)) 1061 frame->tf_rsp = (register_t)(td->td_kstack + PAGE_SIZE); 1062 if (in_kstack_guard(frame->tf_rbp)) 1063 frame->tf_rbp = (register_t)(td->td_kstack + PAGE_SIZE); 1064 } else { 1065 kprintf("DOUBLE FAULT\n"); 1066 } 1067 kprintf("\nFatal double fault\n"); 1068 kprintf("rip = 0x%lx\n", frame->tf_rip); 1069 kprintf("rsp = 0x%lx\n", frame->tf_rsp); 1070 kprintf("rbp = 0x%lx\n", frame->tf_rbp); 1071 /* three separate prints in case of a trap on an unmapped page */ 1072 kprintf("cpuid = %d; ", mycpu->gd_cpuid); 1073 kprintf("lapic->id = %08x\n", lapic->id); 1074 panic("double fault"); 1075 } 1076 1077 /* 1078 * syscall2 - MP aware system call request C handler 1079 * 1080 * A system call is essentially treated as a trap except that the 1081 * MP lock is not held on entry or return. We are responsible for 1082 * obtaining the MP lock if necessary and for handling ASTs 1083 * (e.g. a task switch) prior to return. 1084 * 1085 * MPSAFE 1086 */ 1087 void 1088 syscall2(struct trapframe *frame) 1089 { 1090 struct thread *td = curthread; 1091 struct proc *p = td->td_proc; 1092 struct lwp *lp = td->td_lwp; 1093 caddr_t params; 1094 struct sysent *callp; 1095 register_t orig_tf_rflags; 1096 int sticks; 1097 int error; 1098 int narg; 1099 #ifdef INVARIANTS 1100 int crit_count = td->td_critcount; 1101 #endif 1102 int have_mplock = 0; 1103 register_t *argp; 1104 u_int code; 1105 int reg, regcnt; 1106 union sysunion args; 1107 register_t *argsdst; 1108 1109 mycpu->gd_cnt.v_syscall++; 1110 1111 #ifdef DIAGNOSTIC 1112 if (ISPL(frame->tf_cs) != SEL_UPL) { 1113 get_mplock(); 1114 panic("syscall"); 1115 /* NOT REACHED */ 1116 } 1117 #endif 1118 1119 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid, 1120 frame->tf_rax); 1121 1122 userenter(td, p); /* lazy raise our priority */ 1123 1124 reg = 0; 1125 regcnt = 6; 1126 /* 1127 * Misc 1128 */ 1129 sticks = (int)td->td_sticks; 1130 orig_tf_rflags = frame->tf_rflags; 1131 1132 /* 1133 * Virtual kernel intercept - if a VM context managed by a virtual 1134 * kernel issues a system call the virtual kernel handles it, not us. 1135 * Restore the virtual kernel context and return from its system 1136 * call. The current frame is copied out to the virtual kernel. 1137 */ 1138 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 1139 vkernel_trap(lp, frame); 1140 error = EJUSTRETURN; 1141 goto out; 1142 } 1143 1144 /* 1145 * Get the system call parameters and account for time 1146 */ 1147 KASSERT(lp->lwp_md.md_regs == frame, 1148 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame)); 1149 params = (caddr_t)frame->tf_rsp + sizeof(register_t); 1150 code = frame->tf_rax; 1151 1152 if (p->p_sysent->sv_prepsyscall) { 1153 (*p->p_sysent->sv_prepsyscall)( 1154 frame, (int *)(&args.nosys.sysmsg + 1), 1155 &code, ¶ms); 1156 } else { 1157 if (code == SYS_syscall || code == SYS___syscall) { 1158 code = frame->tf_rdi; 1159 reg++; 1160 regcnt--; 1161 } 1162 } 1163 1164 if (p->p_sysent->sv_mask) 1165 code &= p->p_sysent->sv_mask; 1166 1167 if (code >= p->p_sysent->sv_size) 1168 callp = &p->p_sysent->sv_table[0]; 1169 else 1170 callp = &p->p_sysent->sv_table[code]; 1171 1172 narg = callp->sy_narg & SYF_ARGMASK; 1173 1174 /* 1175 * On x86_64 we get up to six arguments in registers. The rest are 1176 * on the stack. The first six members of 'struct trapframe' happen 1177 * to be the registers used to pass arguments, in exactly the right 1178 * order. 1179 */ 1180 argp = &frame->tf_rdi; 1181 argp += reg; 1182 argsdst = (register_t *)(&args.nosys.sysmsg + 1); 1183 /* 1184 * JG can we overflow the space pointed to by 'argsdst' 1185 * either with 'bcopy' or with 'copyin'? 1186 */ 1187 bcopy(argp, argsdst, sizeof(register_t) * regcnt); 1188 /* 1189 * copyin is MP aware, but the tracing code is not 1190 */ 1191 if (narg > regcnt) { 1192 KASSERT(params != NULL, ("copyin args with no params!")); 1193 error = copyin(params, &argsdst[regcnt], 1194 (narg - regcnt) * sizeof(register_t)); 1195 if (error) { 1196 #ifdef KTRACE 1197 if (KTRPOINT(td, KTR_SYSCALL)) { 1198 MAKEMPSAFE(have_mplock); 1199 1200 ktrsyscall(lp, code, narg, 1201 (void *)(&args.nosys.sysmsg + 1)); 1202 } 1203 #endif 1204 goto bad; 1205 } 1206 } 1207 1208 #ifdef KTRACE 1209 if (KTRPOINT(td, KTR_SYSCALL)) { 1210 MAKEMPSAFE(have_mplock); 1211 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1)); 1212 } 1213 #endif 1214 1215 /* 1216 * Default return value is 0 (will be copied to %rax). Double-value 1217 * returns use %rax and %rdx. %rdx is left unchanged for system 1218 * calls which return only one result. 1219 */ 1220 args.sysmsg_fds[0] = 0; 1221 args.sysmsg_fds[1] = frame->tf_rdx; 1222 1223 /* 1224 * The syscall might manipulate the trap frame. If it does it 1225 * will probably return EJUSTRETURN. 1226 */ 1227 args.sysmsg_frame = frame; 1228 1229 STOPEVENT(p, S_SCE, narg); /* MP aware */ 1230 1231 /* 1232 * NOTE: All system calls run MPSAFE now. The system call itself 1233 * is responsible for getting the MP lock. 1234 */ 1235 #ifdef SYSCALL_DEBUG 1236 uint64_t tscval = rdtsc(); 1237 #endif 1238 error = (*callp->sy_call)(&args); 1239 #ifdef SYSCALL_DEBUG 1240 tscval = rdtsc() - tscval; 1241 tscval = tscval * 1000000 / tsc_frequency; 1242 if (SysCallsWorstCase[code] < tscval) 1243 SysCallsWorstCase[code] = tscval; 1244 #endif 1245 1246 out: 1247 /* 1248 * MP SAFE (we may or may not have the MP lock at this point) 1249 */ 1250 //kprintf("SYSMSG %d ", error); 1251 switch (error) { 1252 case 0: 1253 /* 1254 * Reinitialize proc pointer `p' as it may be different 1255 * if this is a child returning from fork syscall. 1256 */ 1257 p = curproc; 1258 lp = curthread->td_lwp; 1259 frame->tf_rax = args.sysmsg_fds[0]; 1260 frame->tf_rdx = args.sysmsg_fds[1]; 1261 frame->tf_rflags &= ~PSL_C; 1262 break; 1263 case ERESTART: 1264 /* 1265 * Reconstruct pc, we know that 'syscall' is 2 bytes. 1266 * We have to do a full context restore so that %r10 1267 * (which was holding the value of %rcx) is restored for 1268 * the next iteration. 1269 */ 1270 if (frame->tf_err != 0 && frame->tf_err != 2) 1271 kprintf("lp %s:%d frame->tf_err is weird %ld\n", 1272 td->td_comm, lp->lwp_proc->p_pid, frame->tf_err); 1273 frame->tf_rip -= frame->tf_err; 1274 frame->tf_r10 = frame->tf_rcx; 1275 break; 1276 case EJUSTRETURN: 1277 break; 1278 case EASYNC: 1279 panic("Unexpected EASYNC return value (for now)"); 1280 default: 1281 bad: 1282 if (p->p_sysent->sv_errsize) { 1283 if (error >= p->p_sysent->sv_errsize) 1284 error = -1; /* XXX */ 1285 else 1286 error = p->p_sysent->sv_errtbl[error]; 1287 } 1288 frame->tf_rax = error; 1289 frame->tf_rflags |= PSL_C; 1290 break; 1291 } 1292 1293 /* 1294 * Traced syscall. trapsignal() is not MP aware. 1295 */ 1296 if (orig_tf_rflags & PSL_T) { 1297 MAKEMPSAFE(have_mplock); 1298 frame->tf_rflags &= ~PSL_T; 1299 trapsignal(lp, SIGTRAP, TRAP_TRACE); 1300 } 1301 1302 /* 1303 * Handle reschedule and other end-of-syscall issues 1304 */ 1305 userret(lp, frame, sticks); 1306 1307 #ifdef KTRACE 1308 if (KTRPOINT(td, KTR_SYSRET)) { 1309 MAKEMPSAFE(have_mplock); 1310 ktrsysret(lp, code, error, args.sysmsg_result); 1311 } 1312 #endif 1313 1314 /* 1315 * This works because errno is findable through the 1316 * register set. If we ever support an emulation where this 1317 * is not the case, this code will need to be revisited. 1318 */ 1319 STOPEVENT(p, S_SCX, code); 1320 1321 userexit(lp); 1322 /* 1323 * Release the MP lock if we had to get it 1324 */ 1325 if (have_mplock) 1326 rel_mplock(); 1327 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error); 1328 #ifdef INVARIANTS 1329 KASSERT(crit_count == td->td_critcount, 1330 ("syscall: critical section count mismatch! %d/%d", 1331 crit_count, td->td_pri)); 1332 KASSERT(&td->td_toks_base == td->td_toks_stop, 1333 ("syscall: extra tokens held after trap! %ld", 1334 td->td_toks_stop - &td->td_toks_base)); 1335 #endif 1336 } 1337 1338 /* 1339 * NOTE: mplock not held at any point 1340 */ 1341 void 1342 fork_return(struct lwp *lp, struct trapframe *frame) 1343 { 1344 frame->tf_rax = 0; /* Child returns zero */ 1345 frame->tf_rflags &= ~PSL_C; /* success */ 1346 frame->tf_rdx = 1; 1347 1348 generic_lwp_return(lp, frame); 1349 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid); 1350 } 1351 1352 /* 1353 * Simplified back end of syscall(), used when returning from fork() 1354 * directly into user mode. 1355 * 1356 * This code will return back into the fork trampoline code which then 1357 * runs doreti. 1358 * 1359 * NOTE: The mplock is not held at any point. 1360 */ 1361 void 1362 generic_lwp_return(struct lwp *lp, struct trapframe *frame) 1363 { 1364 struct proc *p = lp->lwp_proc; 1365 1366 /* 1367 * Newly forked processes are given a kernel priority. We have to 1368 * adjust the priority to a normal user priority and fake entry 1369 * into the kernel (call userenter()) to install a passive release 1370 * function just in case userret() decides to stop the process. This 1371 * can occur when ^Z races a fork. If we do not install the passive 1372 * release function the current process designation will not be 1373 * released when the thread goes to sleep. 1374 */ 1375 lwkt_setpri_self(TDPRI_USER_NORM); 1376 userenter(lp->lwp_thread, p); 1377 userret(lp, frame, 0); 1378 #ifdef KTRACE 1379 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET)) 1380 ktrsysret(lp, SYS_fork, 0, 0); 1381 #endif 1382 lp->lwp_flags |= LWP_PASSIVE_ACQ; 1383 userexit(lp); 1384 lp->lwp_flags &= ~LWP_PASSIVE_ACQ; 1385 } 1386 1387 /* 1388 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA 1389 * fault (which is then passed back to the virtual kernel) if an attempt is 1390 * made to use the FP unit. 1391 * 1392 * XXX this is a fairly big hack. 1393 */ 1394 void 1395 set_vkernel_fp(struct trapframe *frame) 1396 { 1397 struct thread *td = curthread; 1398 1399 if (frame->tf_xflags & PGEX_FPFAULT) { 1400 td->td_pcb->pcb_flags |= FP_VIRTFP; 1401 if (mdcpu->gd_npxthread == td) 1402 npxexit(); 1403 } else { 1404 td->td_pcb->pcb_flags &= ~FP_VIRTFP; 1405 } 1406 } 1407 1408 /* 1409 * Called from vkernel_trap() to fixup the vkernel's syscall 1410 * frame for vmspace_ctl() return. 1411 */ 1412 void 1413 cpu_vkernel_trap(struct trapframe *frame, int error) 1414 { 1415 frame->tf_rax = error; 1416 if (error) 1417 frame->tf_rflags |= PSL_C; 1418 else 1419 frame->tf_rflags &= ~PSL_C; 1420 } 1421