1 /*- 2 * Copyright (C) 1994, David Greenman 3 * Copyright (c) 1990, 1993 4 * The Regents of the University of California. All rights reserved. 5 * 6 * This code is derived from software contributed to Berkeley by 7 * the University of Utah, and William Jolitz. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. All advertising materials mentioning features or use of this software 18 * must display the following acknowledgement: 19 * This product includes software developed by the University of 20 * California, Berkeley and its contributors. 21 * 4. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 * 37 * from: @(#)trap.c 7.4 (Berkeley) 5/13/91 38 * $FreeBSD: src/sys/i386/i386/trap.c,v 1.147.2.11 2003/02/27 19:09:59 luoqi Exp $ 39 */ 40 41 /* 42 * x86_64 Trap and System call handling 43 */ 44 45 #include "use_isa.h" 46 47 #include "opt_ddb.h" 48 #include "opt_ktrace.h" 49 50 #include <sys/param.h> 51 #include <sys/systm.h> 52 #include <sys/proc.h> 53 #include <sys/pioctl.h> 54 #include <sys/kernel.h> 55 #include <sys/resourcevar.h> 56 #include <sys/signalvar.h> 57 #include <sys/signal2.h> 58 #include <sys/syscall.h> 59 #include <sys/sysctl.h> 60 #include <sys/sysent.h> 61 #include <sys/uio.h> 62 #include <sys/vmmeter.h> 63 #include <sys/malloc.h> 64 #ifdef KTRACE 65 #include <sys/ktrace.h> 66 #endif 67 #include <sys/ktr.h> 68 #include <sys/vkernel.h> 69 #include <sys/sysproto.h> 70 #include <sys/sysunion.h> 71 #include <sys/vmspace.h> 72 73 #include <vm/vm.h> 74 #include <vm/vm_param.h> 75 #include <sys/lock.h> 76 #include <vm/pmap.h> 77 #include <vm/vm_kern.h> 78 #include <vm/vm_map.h> 79 #include <vm/vm_page.h> 80 #include <vm/vm_extern.h> 81 82 #include <machine/cpu.h> 83 #include <machine/md_var.h> 84 #include <machine/pcb.h> 85 #include <machine/smp.h> 86 #include <machine/tss.h> 87 #include <machine/globaldata.h> 88 89 #include <ddb/ddb.h> 90 91 #include <sys/msgport2.h> 92 #include <sys/thread2.h> 93 #include <sys/mplock2.h> 94 95 #define MAKEMPSAFE(have_mplock) \ 96 if (have_mplock == 0) { \ 97 get_mplock(); \ 98 have_mplock = 1; \ 99 } 100 101 int (*pmath_emulate) (struct trapframe *); 102 103 extern int trapwrite (unsigned addr); 104 105 static int trap_pfault (struct trapframe *, int, vm_offset_t); 106 static void trap_fatal (struct trapframe *, int, vm_offset_t); 107 void dblfault_handler (void); 108 extern int vmm_enabled; 109 110 #if 0 111 extern inthand_t IDTVEC(syscall); 112 #endif 113 114 #define MAX_TRAP_MSG 30 115 static char *trap_msg[] = { 116 "", /* 0 unused */ 117 "privileged instruction fault", /* 1 T_PRIVINFLT */ 118 "", /* 2 unused */ 119 "breakpoint instruction fault", /* 3 T_BPTFLT */ 120 "", /* 4 unused */ 121 "", /* 5 unused */ 122 "arithmetic trap", /* 6 T_ARITHTRAP */ 123 "system forced exception", /* 7 T_ASTFLT */ 124 "", /* 8 unused */ 125 "general protection fault", /* 9 T_PROTFLT */ 126 "trace trap", /* 10 T_TRCTRAP */ 127 "", /* 11 unused */ 128 "page fault", /* 12 T_PAGEFLT */ 129 "", /* 13 unused */ 130 "alignment fault", /* 14 T_ALIGNFLT */ 131 "", /* 15 unused */ 132 "", /* 16 unused */ 133 "", /* 17 unused */ 134 "integer divide fault", /* 18 T_DIVIDE */ 135 "non-maskable interrupt trap", /* 19 T_NMI */ 136 "overflow trap", /* 20 T_OFLOW */ 137 "FPU bounds check fault", /* 21 T_BOUND */ 138 "FPU device not available", /* 22 T_DNA */ 139 "double fault", /* 23 T_DOUBLEFLT */ 140 "FPU operand fetch fault", /* 24 T_FPOPFLT */ 141 "invalid TSS fault", /* 25 T_TSSFLT */ 142 "segment not present fault", /* 26 T_SEGNPFLT */ 143 "stack fault", /* 27 T_STKFLT */ 144 "machine check trap", /* 28 T_MCHK */ 145 "SIMD floating-point exception", /* 29 T_XMMFLT */ 146 "reserved (unknown) fault", /* 30 T_RESERVED */ 147 }; 148 149 #ifdef DDB 150 static int ddb_on_nmi = 1; 151 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW, 152 &ddb_on_nmi, 0, "Go to DDB on NMI"); 153 #endif 154 static int panic_on_nmi = 1; 155 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW, 156 &panic_on_nmi, 0, "Panic on NMI"); 157 static int fast_release; 158 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW, 159 &fast_release, 0, "Passive Release was optimal"); 160 static int slow_release; 161 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW, 162 &slow_release, 0, "Passive Release was nonoptimal"); 163 164 /* 165 * Passively intercepts the thread switch function to increase 166 * the thread priority from a user priority to a kernel priority, reducing 167 * syscall and trap overhead for the case where no switch occurs. 168 * 169 * Synchronizes td_ucred with p_ucred. This is used by system calls, 170 * signal handling, faults, AST traps, and anything else that enters the 171 * kernel from userland and provides the kernel with a stable read-only 172 * copy of the process ucred. 173 */ 174 static __inline void 175 userenter(struct thread *curtd, struct proc *curp) 176 { 177 struct ucred *ocred; 178 struct ucred *ncred; 179 180 curtd->td_release = lwkt_passive_release; 181 182 if (curtd->td_ucred != curp->p_ucred) { 183 ncred = crhold(curp->p_ucred); 184 ocred = curtd->td_ucred; 185 curtd->td_ucred = ncred; 186 if (ocred) 187 crfree(ocred); 188 } 189 } 190 191 /* 192 * Handle signals, profiling, and other AST's and/or tasks that 193 * must be completed before we can return to or try to return to userland. 194 * 195 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64 196 * arithmatic on the delta calculation so the absolute tick values are 197 * truncated to an integer. 198 */ 199 static void 200 userret(struct lwp *lp, struct trapframe *frame, int sticks) 201 { 202 struct proc *p = lp->lwp_proc; 203 int sig; 204 205 /* 206 * Charge system time if profiling. Note: times are in microseconds. 207 * This may do a copyout and block, so do it first even though it 208 * means some system time will be charged as user time. 209 */ 210 if (p->p_flags & P_PROFIL) { 211 addupc_task(p, frame->tf_rip, 212 (u_int)((int)lp->lwp_thread->td_sticks - sticks)); 213 } 214 215 recheck: 216 /* 217 * Specific on-return-to-usermode checks (LWP_MP_WEXIT, 218 * LWP_MP_VNLRU, etc). 219 */ 220 if (lp->lwp_mpflags & LWP_MP_URETMASK) 221 lwpuserret(lp); 222 223 /* 224 * Block here if we are in a stopped state. 225 */ 226 if (p->p_stat == SSTOP || p->p_stat == SCORE) { 227 lwkt_gettoken(&p->p_token); 228 tstop(); 229 lwkt_reltoken(&p->p_token); 230 goto recheck; 231 } 232 233 /* 234 * Post any pending upcalls. If running a virtual kernel be sure 235 * to restore the virtual kernel's vmspace before posting the upcall. 236 */ 237 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF)) { 238 lwkt_gettoken(&p->p_token); 239 if (p->p_flags & P_SIGVTALRM) { 240 p->p_flags &= ~P_SIGVTALRM; 241 ksignal(p, SIGVTALRM); 242 } 243 if (p->p_flags & P_SIGPROF) { 244 p->p_flags &= ~P_SIGPROF; 245 ksignal(p, SIGPROF); 246 } 247 lwkt_reltoken(&p->p_token); 248 goto recheck; 249 } 250 251 /* 252 * Post any pending signals 253 * 254 * WARNING! postsig() can exit and not return. 255 */ 256 if ((sig = CURSIG_TRACE(lp)) != 0) { 257 lwkt_gettoken(&p->p_token); 258 postsig(sig); 259 lwkt_reltoken(&p->p_token); 260 goto recheck; 261 } 262 263 /* 264 * block here if we are swapped out, but still process signals 265 * (such as SIGKILL). proc0 (the swapin scheduler) is already 266 * aware of our situation, we do not have to wake it up. 267 */ 268 if (p->p_flags & P_SWAPPEDOUT) { 269 lwkt_gettoken(&p->p_token); 270 get_mplock(); 271 p->p_flags |= P_SWAPWAIT; 272 swapin_request(); 273 if (p->p_flags & P_SWAPWAIT) 274 tsleep(p, PCATCH, "SWOUT", 0); 275 p->p_flags &= ~P_SWAPWAIT; 276 rel_mplock(); 277 lwkt_reltoken(&p->p_token); 278 goto recheck; 279 } 280 281 /* 282 * In a multi-threaded program it is possible for a thread to change 283 * signal state during a system call which temporarily changes the 284 * signal mask. In this case postsig() might not be run and we 285 * have to restore the mask ourselves. 286 */ 287 if (lp->lwp_flags & LWP_OLDMASK) { 288 lp->lwp_flags &= ~LWP_OLDMASK; 289 lp->lwp_sigmask = lp->lwp_oldsigmask; 290 goto recheck; 291 } 292 } 293 294 /* 295 * Cleanup from userenter and any passive release that might have occured. 296 * We must reclaim the current-process designation before we can return 297 * to usermode. We also handle both LWKT and USER reschedule requests. 298 */ 299 static __inline void 300 userexit(struct lwp *lp) 301 { 302 struct thread *td = lp->lwp_thread; 303 /* globaldata_t gd = td->td_gd; */ 304 305 /* 306 * Handle stop requests at kernel priority. Any requests queued 307 * after this loop will generate another AST. 308 */ 309 while (lp->lwp_proc->p_stat == SSTOP || 310 lp->lwp_proc->p_stat == SCORE) { 311 lwkt_gettoken(&lp->lwp_proc->p_token); 312 tstop(); 313 lwkt_reltoken(&lp->lwp_proc->p_token); 314 } 315 316 /* 317 * Reduce our priority in preparation for a return to userland. If 318 * our passive release function was still in place, our priority was 319 * never raised and does not need to be reduced. 320 */ 321 lwkt_passive_recover(td); 322 323 /* 324 * Become the current user scheduled process if we aren't already, 325 * and deal with reschedule requests and other factors. 326 */ 327 lp->lwp_proc->p_usched->acquire_curproc(lp); 328 /* WARNING: we may have migrated cpu's */ 329 /* gd = td->td_gd; */ 330 } 331 332 #if !defined(KTR_KERNENTRY) 333 #define KTR_KERNENTRY KTR_ALL 334 #endif 335 KTR_INFO_MASTER(kernentry); 336 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, 337 "TRAP(pid %hd, tid %hd, trapno %ld, eva %lu)", 338 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva); 339 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %hd, tid %hd)", 340 pid_t pid, lwpid_t tid); 341 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %hd, tid %hd, nr %ld)", 342 pid_t pid, lwpid_t tid, register_t trapno); 343 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %hd, tid %hd, err %d)", 344 pid_t pid, lwpid_t tid, int err); 345 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %hd, tid %hd)", 346 pid_t pid, lwpid_t tid); 347 348 /* 349 * Exception, fault, and trap interface to the kernel. 350 * This common code is called from assembly language IDT gate entry 351 * routines that prepare a suitable stack frame, and restore this 352 * frame after the exception has been processed. 353 * 354 * This function is also called from doreti in an interlock to handle ASTs. 355 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap 356 * 357 * NOTE! We have to retrieve the fault address prior to obtaining the 358 * MP lock because get_mplock() may switch out. YYY cr2 really ought 359 * to be retrieved by the assembly code, not here. 360 * 361 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing 362 * if an attempt is made to switch from a fast interrupt or IPI. This is 363 * necessary to properly take fatal kernel traps on SMP machines if 364 * get_mplock() has to block. 365 */ 366 367 void 368 user_trap(struct trapframe *frame) 369 { 370 struct globaldata *gd = mycpu; 371 struct thread *td = gd->gd_curthread; 372 struct lwp *lp = td->td_lwp; 373 struct proc *p; 374 int sticks = 0; 375 int i = 0, ucode = 0, type, code; 376 int have_mplock = 0; 377 #ifdef INVARIANTS 378 int crit_count = td->td_critcount; 379 lwkt_tokref_t curstop = td->td_toks_stop; 380 #endif 381 vm_offset_t eva; 382 383 p = td->td_proc; 384 385 if (frame->tf_trapno == T_PAGEFLT) 386 eva = frame->tf_addr; 387 else 388 eva = 0; 389 #if 0 390 kprintf("USER_TRAP AT %08lx xflags %ld trapno %ld eva %08lx\n", 391 frame->tf_rip, frame->tf_xflags, frame->tf_trapno, eva); 392 #endif 393 394 /* 395 * Everything coming from user mode runs through user_trap, 396 * including system calls. 397 */ 398 if (frame->tf_trapno == T_FAST_SYSCALL) { 399 syscall2(frame); 400 return; 401 } 402 403 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid, 404 frame->tf_trapno, eva); 405 406 #ifdef DDB 407 if (db_active) { 408 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0); 409 ++gd->gd_trap_nesting_level; 410 MAKEMPSAFE(have_mplock); 411 trap_fatal(frame, TRUE, eva); 412 --gd->gd_trap_nesting_level; 413 goto out2; 414 } 415 #endif 416 417 type = frame->tf_trapno; 418 code = frame->tf_err; 419 420 userenter(td, p); 421 422 sticks = (int)td->td_sticks; 423 lp->lwp_md.md_regs = frame; 424 425 switch (type) { 426 case T_PRIVINFLT: /* privileged instruction fault */ 427 i = SIGILL; 428 ucode = ILL_PRVOPC; 429 break; 430 431 case T_BPTFLT: /* bpt instruction fault */ 432 case T_TRCTRAP: /* trace trap */ 433 frame->tf_rflags &= ~PSL_T; 434 i = SIGTRAP; 435 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT); 436 break; 437 438 case T_ARITHTRAP: /* arithmetic trap */ 439 ucode = code; 440 i = SIGFPE; 441 break; 442 443 case T_ASTFLT: /* Allow process switch */ 444 mycpu->gd_cnt.v_soft++; 445 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) { 446 atomic_clear_int(&mycpu->gd_reqflags, RQF_AST_OWEUPC); 447 addupc_task(p, p->p_prof.pr_addr, p->p_prof.pr_ticks); 448 } 449 goto out; 450 451 /* 452 * The following two traps can happen in 453 * vm86 mode, and, if so, we want to handle 454 * them specially. 455 */ 456 case T_PROTFLT: /* general protection fault */ 457 case T_STKFLT: /* stack fault */ 458 #if 0 459 if (frame->tf_eflags & PSL_VM) { 460 i = vm86_emulate((struct vm86frame *)frame); 461 if (i == 0) 462 goto out; 463 break; 464 } 465 #endif 466 /* FALL THROUGH */ 467 468 case T_SEGNPFLT: /* segment not present fault */ 469 case T_TSSFLT: /* invalid TSS fault */ 470 case T_DOUBLEFLT: /* double fault */ 471 default: 472 i = SIGBUS; 473 ucode = code + BUS_SEGM_FAULT ; 474 break; 475 476 case T_PAGEFLT: /* page fault */ 477 i = trap_pfault(frame, TRUE, eva); 478 if (i == -1 || i == 0) 479 goto out; 480 481 482 if (i == SIGSEGV) 483 ucode = SEGV_MAPERR; 484 else { 485 i = SIGSEGV; 486 ucode = SEGV_ACCERR; 487 } 488 break; 489 490 case T_DIVIDE: /* integer divide fault */ 491 ucode = FPE_INTDIV; 492 i = SIGFPE; 493 break; 494 495 #if NISA > 0 496 case T_NMI: 497 MAKEMPSAFE(have_mplock); 498 /* machine/parity/power fail/"kitchen sink" faults */ 499 if (isa_nmi(code) == 0) { 500 #ifdef DDB 501 /* 502 * NMI can be hooked up to a pushbutton 503 * for debugging. 504 */ 505 if (ddb_on_nmi) { 506 kprintf ("NMI ... going to debugger\n"); 507 kdb_trap(type, 0, frame); 508 } 509 #endif /* DDB */ 510 goto out2; 511 } else if (panic_on_nmi) 512 panic("NMI indicates hardware failure"); 513 break; 514 #endif /* NISA > 0 */ 515 516 case T_OFLOW: /* integer overflow fault */ 517 ucode = FPE_INTOVF; 518 i = SIGFPE; 519 break; 520 521 case T_BOUND: /* bounds check fault */ 522 ucode = FPE_FLTSUB; 523 i = SIGFPE; 524 break; 525 526 case T_DNA: 527 /* 528 * Virtual kernel intercept - pass the DNA exception 529 * to the (emulated) virtual kernel if it asked to handle 530 * it. This occurs when the virtual kernel is holding 531 * onto the FP context for a different emulated 532 * process then the one currently running. 533 * 534 * We must still call npxdna() since we may have 535 * saved FP state that the (emulated) virtual kernel 536 * needs to hand over to a different emulated process. 537 */ 538 if (lp->lwp_vkernel && lp->lwp_vkernel->ve && 539 (td->td_pcb->pcb_flags & FP_VIRTFP) 540 ) { 541 npxdna(frame); 542 break; 543 } 544 545 /* 546 * The kernel may have switched out the FP unit's 547 * state, causing the user process to take a fault 548 * when it tries to use the FP unit. Restore the 549 * state here 550 */ 551 if (npxdna(frame)) 552 goto out; 553 if (!pmath_emulate) { 554 i = SIGFPE; 555 ucode = FPE_FPU_NP_TRAP; 556 break; 557 } 558 i = (*pmath_emulate)(frame); 559 if (i == 0) { 560 if (!(frame->tf_rflags & PSL_T)) 561 goto out2; 562 frame->tf_rflags &= ~PSL_T; 563 i = SIGTRAP; 564 } 565 /* else ucode = emulator_only_knows() XXX */ 566 break; 567 568 case T_FPOPFLT: /* FPU operand fetch fault */ 569 ucode = T_FPOPFLT; 570 i = SIGILL; 571 break; 572 573 case T_XMMFLT: /* SIMD floating-point exception */ 574 ucode = 0; /* XXX */ 575 i = SIGFPE; 576 break; 577 } 578 579 /* 580 * Virtual kernel intercept - if the fault is directly related to a 581 * VM context managed by a virtual kernel then let the virtual kernel 582 * handle it. 583 */ 584 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 585 vkernel_trap(lp, frame); 586 goto out; 587 } 588 589 /* 590 * Translate fault for emulators (e.g. Linux) 591 */ 592 if (*p->p_sysent->sv_transtrap) 593 i = (*p->p_sysent->sv_transtrap)(i, type); 594 595 MAKEMPSAFE(have_mplock); 596 trapsignal(lp, i, ucode); 597 598 #ifdef DEBUG 599 if (type <= MAX_TRAP_MSG) { 600 uprintf("fatal process exception: %s", 601 trap_msg[type]); 602 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 603 uprintf(", fault VA = 0x%lx", (u_long)eva); 604 uprintf("\n"); 605 } 606 #endif 607 608 out: 609 userret(lp, frame, sticks); 610 userexit(lp); 611 out2: ; 612 if (have_mplock) 613 rel_mplock(); 614 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid); 615 #ifdef INVARIANTS 616 KASSERT(crit_count == td->td_critcount, 617 ("trap: critical section count mismatch! %d/%d", 618 crit_count, td->td_pri)); 619 KASSERT(curstop == td->td_toks_stop, 620 ("trap: extra tokens held after trap! %ld/%ld", 621 curstop - &td->td_toks_base, 622 td->td_toks_stop - &td->td_toks_base)); 623 #endif 624 } 625 626 void 627 kern_trap(struct trapframe *frame) 628 { 629 struct globaldata *gd = mycpu; 630 struct thread *td = gd->gd_curthread; 631 struct lwp *lp; 632 struct proc *p; 633 int i = 0, ucode = 0, type, code; 634 int have_mplock = 0; 635 #ifdef INVARIANTS 636 int crit_count = td->td_critcount; 637 lwkt_tokref_t curstop = td->td_toks_stop; 638 #endif 639 vm_offset_t eva; 640 641 lp = td->td_lwp; 642 p = td->td_proc; 643 644 if (frame->tf_trapno == T_PAGEFLT) 645 eva = frame->tf_addr; 646 else 647 eva = 0; 648 649 #ifdef DDB 650 if (db_active) { 651 ++gd->gd_trap_nesting_level; 652 MAKEMPSAFE(have_mplock); 653 trap_fatal(frame, FALSE, eva); 654 --gd->gd_trap_nesting_level; 655 goto out2; 656 } 657 #endif 658 659 type = frame->tf_trapno; 660 code = frame->tf_err; 661 662 #if 0 663 kernel_trap: 664 #endif 665 /* kernel trap */ 666 667 switch (type) { 668 case T_PAGEFLT: /* page fault */ 669 trap_pfault(frame, FALSE, eva); 670 goto out2; 671 672 case T_DNA: 673 /* 674 * The kernel may be using npx for copying or other 675 * purposes. 676 */ 677 panic("kernel NPX should not happen"); 678 if (npxdna(frame)) 679 goto out2; 680 break; 681 682 case T_PROTFLT: /* general protection fault */ 683 case T_SEGNPFLT: /* segment not present fault */ 684 /* 685 * Invalid segment selectors and out of bounds 686 * %eip's and %esp's can be set up in user mode. 687 * This causes a fault in kernel mode when the 688 * kernel tries to return to user mode. We want 689 * to get this fault so that we can fix the 690 * problem here and not have to check all the 691 * selectors and pointers when the user changes 692 * them. 693 */ 694 if (mycpu->gd_intr_nesting_level == 0) { 695 if (td->td_pcb->pcb_onfault) { 696 frame->tf_rip = 697 (register_t)td->td_pcb->pcb_onfault; 698 goto out2; 699 } 700 } 701 break; 702 703 case T_TSSFLT: 704 /* 705 * PSL_NT can be set in user mode and isn't cleared 706 * automatically when the kernel is entered. This 707 * causes a TSS fault when the kernel attempts to 708 * `iret' because the TSS link is uninitialized. We 709 * want to get this fault so that we can fix the 710 * problem here and not every time the kernel is 711 * entered. 712 */ 713 if (frame->tf_rflags & PSL_NT) { 714 frame->tf_rflags &= ~PSL_NT; 715 goto out2; 716 } 717 break; 718 719 case T_TRCTRAP: /* trace trap */ 720 #if 0 721 if (frame->tf_eip == (int)IDTVEC(syscall)) { 722 /* 723 * We've just entered system mode via the 724 * syscall lcall. Continue single stepping 725 * silently until the syscall handler has 726 * saved the flags. 727 */ 728 goto out2; 729 } 730 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) { 731 /* 732 * The syscall handler has now saved the 733 * flags. Stop single stepping it. 734 */ 735 frame->tf_eflags &= ~PSL_T; 736 goto out2; 737 } 738 #endif 739 #if 0 740 /* 741 * Ignore debug register trace traps due to 742 * accesses in the user's address space, which 743 * can happen under several conditions such as 744 * if a user sets a watchpoint on a buffer and 745 * then passes that buffer to a system call. 746 * We still want to get TRCTRAPS for addresses 747 * in kernel space because that is useful when 748 * debugging the kernel. 749 */ 750 if (user_dbreg_trap()) { 751 /* 752 * Reset breakpoint bits because the 753 * processor doesn't 754 */ 755 load_dr6(rdr6() & 0xfffffff0); 756 goto out2; 757 } 758 #endif 759 /* 760 * Fall through (TRCTRAP kernel mode, kernel address) 761 */ 762 case T_BPTFLT: 763 /* 764 * If DDB is enabled, let it handle the debugger trap. 765 * Otherwise, debugger traps "can't happen". 766 */ 767 #ifdef DDB 768 MAKEMPSAFE(have_mplock); 769 if (kdb_trap (type, 0, frame)) 770 goto out2; 771 #endif 772 break; 773 case T_DIVIDE: 774 MAKEMPSAFE(have_mplock); 775 trap_fatal(frame, FALSE, eva); 776 goto out2; 777 case T_NMI: 778 MAKEMPSAFE(have_mplock); 779 trap_fatal(frame, FALSE, eva); 780 goto out2; 781 case T_SYSCALL80: 782 case T_FAST_SYSCALL: 783 /* 784 * Ignore this trap generated from a spurious SIGTRAP. 785 * 786 * single stepping in / syscalls leads to spurious / SIGTRAP 787 * so ignore 788 * 789 * Haiku (c) 2007 Simon 'corecode' Schubert 790 */ 791 goto out2; 792 } 793 794 /* 795 * Translate fault for emulators (e.g. Linux) 796 */ 797 if (*p->p_sysent->sv_transtrap) 798 i = (*p->p_sysent->sv_transtrap)(i, type); 799 800 MAKEMPSAFE(have_mplock); 801 trapsignal(lp, i, ucode); 802 803 #ifdef DEBUG 804 if (type <= MAX_TRAP_MSG) { 805 uprintf("fatal process exception: %s", 806 trap_msg[type]); 807 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 808 uprintf(", fault VA = 0x%lx", (u_long)eva); 809 uprintf("\n"); 810 } 811 #endif 812 813 out2: 814 ; 815 if (have_mplock) 816 rel_mplock(); 817 #ifdef INVARIANTS 818 KASSERT(crit_count == td->td_critcount, 819 ("trap: critical section count mismatch! %d/%d", 820 crit_count, td->td_pri)); 821 KASSERT(curstop == td->td_toks_stop, 822 ("trap: extra tokens held after trap! %ld/%ld", 823 curstop - &td->td_toks_base, 824 td->td_toks_stop - &td->td_toks_base)); 825 #endif 826 } 827 828 int 829 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva) 830 { 831 vm_offset_t va; 832 struct vmspace *vm = NULL; 833 vm_map_t map = 0; 834 int rv = 0; 835 vm_prot_t ftype; 836 thread_t td = curthread; 837 struct lwp *lp = td->td_lwp; 838 int fault_flags; 839 840 va = trunc_page(eva); 841 if (usermode == FALSE) { 842 /* 843 * This is a fault on kernel virtual memory. 844 */ 845 map = &kernel_map; 846 } else { 847 /* 848 * This is a fault on non-kernel virtual memory. 849 * vm is initialized above to NULL. If curproc is NULL 850 * or curproc->p_vmspace is NULL the fault is fatal. 851 */ 852 if (lp != NULL) 853 vm = lp->lwp_vmspace; 854 855 if (vm == NULL) 856 goto nogo; 857 858 map = &vm->vm_map; 859 } 860 861 if (frame->tf_err & PGEX_W) 862 ftype = VM_PROT_READ | VM_PROT_WRITE; 863 else 864 ftype = VM_PROT_READ; 865 866 if (map != &kernel_map) { 867 /* 868 * Keep swapout from messing with us during this 869 * critical time. 870 */ 871 PHOLD(lp->lwp_proc); 872 873 /* 874 * Grow the stack if necessary 875 */ 876 /* grow_stack returns false only if va falls into 877 * a growable stack region and the stack growth 878 * fails. It returns true if va was not within 879 * a growable stack region, or if the stack 880 * growth succeeded. 881 */ 882 if (!grow_stack (lp->lwp_proc, va)) { 883 rv = KERN_FAILURE; 884 PRELE(lp->lwp_proc); 885 goto nogo; 886 } 887 888 fault_flags = 0; 889 if (usermode) 890 fault_flags |= VM_FAULT_BURST; 891 if (ftype & VM_PROT_WRITE) 892 fault_flags |= VM_FAULT_DIRTY; 893 else 894 fault_flags |= VM_FAULT_NORMAL; 895 rv = vm_fault(map, va, ftype, fault_flags); 896 897 PRELE(lp->lwp_proc); 898 } else { 899 /* 900 * Don't have to worry about process locking or stacks in the kernel. 901 */ 902 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 903 } 904 905 if (rv == KERN_SUCCESS) 906 return (0); 907 nogo: 908 if (!usermode) { 909 if (td->td_gd->gd_intr_nesting_level == 0 && 910 td->td_pcb->pcb_onfault) { 911 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault; 912 return (0); 913 } 914 trap_fatal(frame, usermode, eva); 915 return (-1); 916 } 917 918 /* 919 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no 920 * kludge is needed to pass the fault address to signal handlers. 921 */ 922 struct proc *p = td->td_proc; 923 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n", 924 (void *)va, (void *)frame->tf_rip, p->p_pid, p->p_comm); 925 /* Debugger("seg-fault"); */ 926 927 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 928 } 929 930 static void 931 trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva) 932 { 933 int code, type, ss; 934 long rsp; 935 936 code = frame->tf_xflags; 937 type = frame->tf_trapno; 938 939 if (type <= MAX_TRAP_MSG) { 940 kprintf("\n\nFatal trap %d: %s while in %s mode\n", 941 type, trap_msg[type], 942 (usermode ? "user" : "kernel")); 943 } 944 /* two separate prints in case of a trap on an unmapped page */ 945 kprintf("cpuid = %d\n", mycpu->gd_cpuid); 946 if (type == T_PAGEFLT) { 947 kprintf("fault virtual address = %p\n", (void *)eva); 948 kprintf("fault code = %s %s, %s\n", 949 usermode ? "user" : "supervisor", 950 code & PGEX_W ? "write" : "read", 951 code & PGEX_P ? "protection violation" : "page not present"); 952 } 953 kprintf("instruction pointer = 0x%lx:0x%lx\n", 954 frame->tf_cs & 0xffff, frame->tf_rip); 955 if (usermode) { 956 ss = frame->tf_ss & 0xffff; 957 rsp = frame->tf_rsp; 958 } else { 959 ss = GSEL(GDATA_SEL, SEL_KPL); 960 rsp = (long)&frame->tf_rsp; 961 } 962 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp); 963 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp); 964 kprintf("processor eflags = "); 965 if (frame->tf_rflags & PSL_T) 966 kprintf("trace trap, "); 967 if (frame->tf_rflags & PSL_I) 968 kprintf("interrupt enabled, "); 969 if (frame->tf_rflags & PSL_NT) 970 kprintf("nested task, "); 971 if (frame->tf_rflags & PSL_RF) 972 kprintf("resume, "); 973 #if 0 974 if (frame->tf_eflags & PSL_VM) 975 kprintf("vm86, "); 976 #endif 977 kprintf("IOPL = %jd\n", (intmax_t)((frame->tf_rflags & PSL_IOPL) >> 12)); 978 kprintf("current process = "); 979 if (curproc) { 980 kprintf("%lu (%s)\n", 981 (u_long)curproc->p_pid, curproc->p_comm ? 982 curproc->p_comm : ""); 983 } else { 984 kprintf("Idle\n"); 985 } 986 kprintf("current thread = pri %d ", curthread->td_pri); 987 if (curthread->td_critcount) 988 kprintf("(CRIT)"); 989 kprintf("\n"); 990 /** 991 * XXX FIXME: 992 * we probably SHOULD have stopped the other CPUs before now! 993 * another CPU COULD have been touching cpl at this moment... 994 */ 995 kprintf(" <- SMP: XXX"); 996 kprintf("\n"); 997 998 #ifdef KDB 999 if (kdb_trap(&psl)) 1000 return; 1001 #endif 1002 #ifdef DDB 1003 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame)) 1004 return; 1005 #endif 1006 kprintf("trap number = %d\n", type); 1007 if (type <= MAX_TRAP_MSG) 1008 panic("%s", trap_msg[type]); 1009 else 1010 panic("unknown/reserved trap"); 1011 } 1012 1013 /* 1014 * Double fault handler. Called when a fault occurs while writing 1015 * a frame for a trap/exception onto the stack. This usually occurs 1016 * when the stack overflows (such is the case with infinite recursion, 1017 * for example). 1018 * 1019 * XXX Note that the current PTD gets replaced by IdlePTD when the 1020 * task switch occurs. This means that the stack that was active at 1021 * the time of the double fault is not available at <kstack> unless 1022 * the machine was idle when the double fault occurred. The downside 1023 * of this is that "trace <ebp>" in ddb won't work. 1024 */ 1025 void 1026 dblfault_handler(void) 1027 { 1028 #if JG 1029 struct mdglobaldata *gd = mdcpu; 1030 #endif 1031 1032 kprintf("\nFatal double fault:\n"); 1033 #if JG 1034 kprintf("rip = 0x%lx\n", gd->gd_common_tss.tss_rip); 1035 kprintf("rsp = 0x%lx\n", gd->gd_common_tss.tss_rsp); 1036 kprintf("rbp = 0x%lx\n", gd->gd_common_tss.tss_rbp); 1037 #endif 1038 /* two separate prints in case of a trap on an unmapped page */ 1039 kprintf("cpuid = %d\n", mycpu->gd_cpuid); 1040 panic("double fault"); 1041 } 1042 1043 /* 1044 * Compensate for 386 brain damage (missing URKR). 1045 * This is a little simpler than the pagefault handler in trap() because 1046 * it the page tables have already been faulted in and high addresses 1047 * are thrown out early for other reasons. 1048 */ 1049 int 1050 trapwrite(unsigned addr) 1051 { 1052 struct lwp *lp; 1053 vm_offset_t va; 1054 struct vmspace *vm; 1055 int rv; 1056 1057 va = trunc_page((vm_offset_t)addr); 1058 /* 1059 * XXX - MAX is END. Changed > to >= for temp. fix. 1060 */ 1061 if (va >= VM_MAX_USER_ADDRESS) 1062 return (1); 1063 1064 lp = curthread->td_lwp; 1065 vm = lp->lwp_vmspace; 1066 1067 PHOLD(lp->lwp_proc); 1068 1069 if (!grow_stack (lp->lwp_proc, va)) { 1070 PRELE(lp->lwp_proc); 1071 return (1); 1072 } 1073 1074 /* 1075 * fault the data page 1076 */ 1077 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY); 1078 1079 PRELE(lp->lwp_proc); 1080 1081 if (rv != KERN_SUCCESS) 1082 return 1; 1083 1084 return (0); 1085 } 1086 1087 /* 1088 * syscall2 - MP aware system call request C handler 1089 * 1090 * A system call is essentially treated as a trap except that the 1091 * MP lock is not held on entry or return. We are responsible for 1092 * obtaining the MP lock if necessary and for handling ASTs 1093 * (e.g. a task switch) prior to return. 1094 * 1095 * In general, only simple access and manipulation of curproc and 1096 * the current stack is allowed without having to hold MP lock. 1097 * 1098 * MPSAFE - note that large sections of this routine are run without 1099 * the MP lock. 1100 */ 1101 void 1102 syscall2(struct trapframe *frame) 1103 { 1104 struct thread *td = curthread; 1105 struct proc *p = td->td_proc; 1106 struct lwp *lp = td->td_lwp; 1107 caddr_t params; 1108 struct sysent *callp; 1109 register_t orig_tf_rflags; 1110 int sticks; 1111 int error; 1112 int narg; 1113 #ifdef INVARIANTS 1114 int crit_count = td->td_critcount; 1115 lwkt_tokref_t curstop = td->td_toks_stop; 1116 #endif 1117 int have_mplock = 0; 1118 register_t *argp; 1119 u_int code; 1120 int reg, regcnt; 1121 union sysunion args; 1122 register_t *argsdst; 1123 1124 mycpu->gd_cnt.v_syscall++; 1125 1126 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid, 1127 frame->tf_rax); 1128 1129 userenter(td, p); /* lazy raise our priority */ 1130 1131 reg = 0; 1132 regcnt = 6; 1133 /* 1134 * Misc 1135 */ 1136 sticks = (int)td->td_sticks; 1137 orig_tf_rflags = frame->tf_rflags; 1138 1139 /* 1140 * Virtual kernel intercept - if a VM context managed by a virtual 1141 * kernel issues a system call the virtual kernel handles it, not us. 1142 * Restore the virtual kernel context and return from its system 1143 * call. The current frame is copied out to the virtual kernel. 1144 */ 1145 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 1146 vkernel_trap(lp, frame); 1147 error = EJUSTRETURN; 1148 goto out; 1149 } 1150 1151 /* 1152 * Get the system call parameters and account for time 1153 */ 1154 lp->lwp_md.md_regs = frame; 1155 params = (caddr_t)frame->tf_rsp + sizeof(register_t); 1156 code = frame->tf_rax; 1157 1158 if (p->p_sysent->sv_prepsyscall) { 1159 (*p->p_sysent->sv_prepsyscall)( 1160 frame, (int *)(&args.nosys.sysmsg + 1), 1161 &code, ¶ms); 1162 } else { 1163 if (code == SYS_syscall || code == SYS___syscall) { 1164 code = frame->tf_rdi; 1165 reg++; 1166 regcnt--; 1167 } 1168 } 1169 1170 if (p->p_sysent->sv_mask) 1171 code &= p->p_sysent->sv_mask; 1172 1173 if (code >= p->p_sysent->sv_size) 1174 callp = &p->p_sysent->sv_table[0]; 1175 else 1176 callp = &p->p_sysent->sv_table[code]; 1177 1178 narg = callp->sy_narg & SYF_ARGMASK; 1179 1180 /* 1181 * On x86_64 we get up to six arguments in registers. The rest are 1182 * on the stack. The first six members of 'struct trapframe' happen 1183 * to be the registers used to pass arguments, in exactly the right 1184 * order. 1185 */ 1186 argp = &frame->tf_rdi; 1187 argp += reg; 1188 argsdst = (register_t *)(&args.nosys.sysmsg + 1); 1189 /* 1190 * JG can we overflow the space pointed to by 'argsdst' 1191 * either with 'bcopy' or with 'copyin'? 1192 */ 1193 bcopy(argp, argsdst, sizeof(register_t) * regcnt); 1194 /* 1195 * copyin is MP aware, but the tracing code is not 1196 */ 1197 if (narg > regcnt) { 1198 KASSERT(params != NULL, ("copyin args with no params!")); 1199 error = copyin(params, &argsdst[regcnt], 1200 (narg - regcnt) * sizeof(register_t)); 1201 if (error) { 1202 #ifdef KTRACE 1203 if (KTRPOINT(td, KTR_SYSCALL)) { 1204 MAKEMPSAFE(have_mplock); 1205 1206 ktrsyscall(lp, code, narg, 1207 (void *)(&args.nosys.sysmsg + 1)); 1208 } 1209 #endif 1210 goto bad; 1211 } 1212 } 1213 1214 #ifdef KTRACE 1215 if (KTRPOINT(td, KTR_SYSCALL)) { 1216 MAKEMPSAFE(have_mplock); 1217 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1)); 1218 } 1219 #endif 1220 1221 /* 1222 * Default return value is 0 (will be copied to %rax). Double-value 1223 * returns use %rax and %rdx. %rdx is left unchanged for system 1224 * calls which return only one result. 1225 */ 1226 args.sysmsg_fds[0] = 0; 1227 args.sysmsg_fds[1] = frame->tf_rdx; 1228 1229 /* 1230 * The syscall might manipulate the trap frame. If it does it 1231 * will probably return EJUSTRETURN. 1232 */ 1233 args.sysmsg_frame = frame; 1234 1235 STOPEVENT(p, S_SCE, narg); /* MP aware */ 1236 1237 /* 1238 * NOTE: All system calls run MPSAFE now. The system call itself 1239 * is responsible for getting the MP lock. 1240 */ 1241 error = (*callp->sy_call)(&args); 1242 1243 #if 0 1244 kprintf("system call %d returned %d\n", code, error); 1245 #endif 1246 1247 out: 1248 /* 1249 * MP SAFE (we may or may not have the MP lock at this point) 1250 */ 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 frame->tf_rip -= frame->tf_err; 1271 frame->tf_r10 = frame->tf_rcx; 1272 break; 1273 case EJUSTRETURN: 1274 break; 1275 case EASYNC: 1276 panic("Unexpected EASYNC return value (for now)"); 1277 default: 1278 bad: 1279 if (p->p_sysent->sv_errsize) { 1280 if (error >= p->p_sysent->sv_errsize) 1281 error = -1; /* XXX */ 1282 else 1283 error = p->p_sysent->sv_errtbl[error]; 1284 } 1285 frame->tf_rax = error; 1286 frame->tf_rflags |= PSL_C; 1287 break; 1288 } 1289 1290 /* 1291 * Traced syscall. trapsignal() is not MP aware. 1292 */ 1293 if (orig_tf_rflags & PSL_T) { 1294 MAKEMPSAFE(have_mplock); 1295 frame->tf_rflags &= ~PSL_T; 1296 trapsignal(lp, SIGTRAP, 0); 1297 } 1298 1299 /* 1300 * Handle reschedule and other end-of-syscall issues 1301 */ 1302 userret(lp, frame, sticks); 1303 1304 #ifdef KTRACE 1305 if (KTRPOINT(td, KTR_SYSRET)) { 1306 MAKEMPSAFE(have_mplock); 1307 ktrsysret(lp, code, error, args.sysmsg_result); 1308 } 1309 #endif 1310 1311 /* 1312 * This works because errno is findable through the 1313 * register set. If we ever support an emulation where this 1314 * is not the case, this code will need to be revisited. 1315 */ 1316 STOPEVENT(p, S_SCX, code); 1317 1318 userexit(lp); 1319 /* 1320 * Release the MP lock if we had to get it 1321 */ 1322 if (have_mplock) 1323 rel_mplock(); 1324 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error); 1325 #ifdef INVARIANTS 1326 KASSERT(&td->td_toks_base == td->td_toks_stop, 1327 ("syscall: critical section count mismatch! %d/%d", 1328 crit_count, td->td_pri)); 1329 KASSERT(curstop == td->td_toks_stop, 1330 ("syscall: extra tokens held after trap! %ld", 1331 td->td_toks_stop - &td->td_toks_base)); 1332 #endif 1333 } 1334 1335 /* 1336 * NOTE: mplock not held at any point 1337 */ 1338 void 1339 fork_return(struct lwp *lp, struct trapframe *frame) 1340 { 1341 frame->tf_rax = 0; /* Child returns zero */ 1342 frame->tf_rflags &= ~PSL_C; /* success */ 1343 frame->tf_rdx = 1; 1344 1345 generic_lwp_return(lp, frame); 1346 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid); 1347 } 1348 1349 /* 1350 * Simplified back end of syscall(), used when returning from fork() 1351 * directly into user mode. 1352 * 1353 * This code will return back into the fork trampoline code which then 1354 * runs doreti. 1355 * 1356 * NOTE: The mplock is not held at any point. 1357 */ 1358 void 1359 generic_lwp_return(struct lwp *lp, struct trapframe *frame) 1360 { 1361 struct proc *p = lp->lwp_proc; 1362 1363 /* 1364 * Check for exit-race. If one lwp exits the process concurrent with 1365 * another lwp creating a new thread, the two operations may cross 1366 * each other resulting in the newly-created lwp not receiving a 1367 * KILL signal. 1368 */ 1369 if (p->p_flags & P_WEXIT) { 1370 kprintf("pid %d (%s) exit race handled\n", 1371 p->p_pid, p->p_comm); 1372 lwpsignal(p, lp, SIGKILL); 1373 } 1374 1375 /* 1376 * Newly forked processes are given a kernel priority. We have to 1377 * adjust the priority to a normal user priority and fake entry 1378 * into the kernel (call userenter()) to install a passive release 1379 * function just in case userret() decides to stop the process. This 1380 * can occur when ^Z races a fork. If we do not install the passive 1381 * release function the current process designation will not be 1382 * released when the thread goes to sleep. 1383 */ 1384 lwkt_setpri_self(TDPRI_USER_NORM); 1385 userenter(lp->lwp_thread, p); 1386 userret(lp, frame, 0); 1387 #ifdef KTRACE 1388 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET)) 1389 ktrsysret(lp, SYS_fork, 0, 0); 1390 #endif 1391 lp->lwp_flags |= LWP_PASSIVE_ACQ; 1392 userexit(lp); 1393 lp->lwp_flags &= ~LWP_PASSIVE_ACQ; 1394 } 1395 1396 /* 1397 * doreti has turned into this. The frame is directly on the stack. We 1398 * pull everything else we need (fpu and tls context) from the current 1399 * thread. 1400 * 1401 * Note on fpu interactions: In a virtual kernel, the fpu context for 1402 * an emulated user mode process is not shared with the virtual kernel's 1403 * fpu context, so we only have to 'stack' fpu contexts within the virtual 1404 * kernel itself, and not even then since the signal() contexts that we care 1405 * about save and restore the FPU state (I think anyhow). 1406 * 1407 * vmspace_ctl() returns an error only if it had problems instaling the 1408 * context we supplied or problems copying data to/from our VM space. 1409 */ 1410 void 1411 go_user(struct intrframe *frame) 1412 { 1413 struct trapframe *tf = (void *)&frame->if_rdi; 1414 int r; 1415 void *id; 1416 1417 /* 1418 * Interrupts may be disabled on entry, make sure all signals 1419 * can be received before beginning our loop. 1420 */ 1421 sigsetmask(0); 1422 1423 /* 1424 * Switch to the current simulated user process, then call 1425 * user_trap() when we break out of it (usually due to a signal). 1426 */ 1427 for (;;) { 1428 #if 1 1429 /* 1430 * Always make the FPU state correct. This should generally 1431 * be faster because the cost of taking a #NM fault through 1432 * the vkernel to the real kernel is astronomical. 1433 */ 1434 tf->tf_xflags &= ~PGEX_FPFAULT; 1435 if (mdcpu->gd_npxthread != curthread) { 1436 if (mdcpu->gd_npxthread) 1437 npxsave(mdcpu->gd_npxthread->td_savefpu); 1438 npxdna(tf); 1439 } 1440 #else 1441 /* 1442 * Tell the real kernel whether it is ok to use the FP 1443 * unit or not, allowing us to take a T_DNA exception 1444 * if the context tries to use the FP. 1445 */ 1446 if (mdcpu->gd_npxthread == curthread) { 1447 tf->tf_xflags &= ~PGEX_FPFAULT; 1448 } else { 1449 tf->tf_xflags |= PGEX_FPFAULT; 1450 } 1451 #endif 1452 1453 /* 1454 * Run emulated user process context. This call interlocks 1455 * with new mailbox signals. 1456 * 1457 * Set PGEX_U unconditionally, indicating a user frame (the 1458 * bit is normally set only by T_PAGEFLT). 1459 */ 1460 if (vmm_enabled) 1461 id = (void *)vtophys(curproc->p_vmspace->vm_pmap.pm_pml4); 1462 else 1463 id = &curproc->p_vmspace->vm_pmap; 1464 1465 r = vmspace_ctl(id, VMSPACE_CTL_RUN, tf, &curthread->td_savevext); 1466 1467 frame->if_xflags |= PGEX_U; 1468 #if 0 1469 kprintf("GO USER %d trap %ld EVA %08lx RIP %08lx RSP %08lx XFLAGS %02lx/%02lx\n", 1470 r, tf->tf_trapno, tf->tf_addr, tf->tf_rip, tf->tf_rsp, 1471 tf->tf_xflags, frame->if_xflags); 1472 #endif 1473 if (r < 0) { 1474 if (errno != EINTR) 1475 panic("vmspace_ctl failed error %d", errno); 1476 } else { 1477 if (tf->tf_trapno) { 1478 user_trap(tf); 1479 } 1480 } 1481 if (mycpu->gd_reqflags & RQF_AST_MASK) { 1482 tf->tf_trapno = T_ASTFLT; 1483 user_trap(tf); 1484 } 1485 tf->tf_trapno = 0; 1486 } 1487 } 1488 1489 /* 1490 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA 1491 * fault (which is then passed back to the virtual kernel) if an attempt is 1492 * made to use the FP unit. 1493 * 1494 * XXX this is a fairly big hack. 1495 */ 1496 void 1497 set_vkernel_fp(struct trapframe *frame) 1498 { 1499 struct thread *td = curthread; 1500 1501 if (frame->tf_xflags & PGEX_FPFAULT) { 1502 td->td_pcb->pcb_flags |= FP_VIRTFP; 1503 if (mdcpu->gd_npxthread == td) 1504 npxexit(); 1505 } else { 1506 td->td_pcb->pcb_flags &= ~FP_VIRTFP; 1507 } 1508 } 1509 1510 /* 1511 * Called from vkernel_trap() to fixup the vkernel's syscall 1512 * frame for vmspace_ctl() return. 1513 */ 1514 void 1515 cpu_vkernel_trap(struct trapframe *frame, int error) 1516 { 1517 frame->tf_rax = error; 1518 if (error) 1519 frame->tf_rflags |= PSL_C; 1520 else 1521 frame->tf_rflags &= ~PSL_C; 1522 } 1523