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/upcall.h> 69 #include <sys/vkernel.h> 70 #include <sys/sysproto.h> 71 #include <sys/sysunion.h> 72 #include <sys/vmspace.h> 73 74 #include <vm/vm.h> 75 #include <vm/vm_param.h> 76 #include <sys/lock.h> 77 #include <vm/pmap.h> 78 #include <vm/vm_kern.h> 79 #include <vm/vm_map.h> 80 #include <vm/vm_page.h> 81 #include <vm/vm_extern.h> 82 83 #include <machine/cpu.h> 84 #include <machine/md_var.h> 85 #include <machine/pcb.h> 86 #include <machine/smp.h> 87 #include <machine/tss.h> 88 #include <machine/globaldata.h> 89 90 #include <ddb/ddb.h> 91 92 #include <sys/msgport2.h> 93 #include <sys/thread2.h> 94 #include <sys/mplock2.h> 95 96 #ifdef SMP 97 98 #define MAKEMPSAFE(have_mplock) \ 99 if (have_mplock == 0) { \ 100 get_mplock(); \ 101 have_mplock = 1; \ 102 } 103 104 #else 105 106 #define MAKEMPSAFE(have_mplock) 107 108 #endif 109 110 int (*pmath_emulate) (struct trapframe *); 111 112 extern int trapwrite (unsigned addr); 113 114 static int trap_pfault (struct trapframe *, int, vm_offset_t); 115 static void trap_fatal (struct trapframe *, int, vm_offset_t); 116 void dblfault_handler (void); 117 118 #if 0 119 extern inthand_t IDTVEC(syscall); 120 #endif 121 122 #define MAX_TRAP_MSG 30 123 static char *trap_msg[] = { 124 "", /* 0 unused */ 125 "privileged instruction fault", /* 1 T_PRIVINFLT */ 126 "", /* 2 unused */ 127 "breakpoint instruction fault", /* 3 T_BPTFLT */ 128 "", /* 4 unused */ 129 "", /* 5 unused */ 130 "arithmetic trap", /* 6 T_ARITHTRAP */ 131 "system forced exception", /* 7 T_ASTFLT */ 132 "", /* 8 unused */ 133 "general protection fault", /* 9 T_PROTFLT */ 134 "trace trap", /* 10 T_TRCTRAP */ 135 "", /* 11 unused */ 136 "page fault", /* 12 T_PAGEFLT */ 137 "", /* 13 unused */ 138 "alignment fault", /* 14 T_ALIGNFLT */ 139 "", /* 15 unused */ 140 "", /* 16 unused */ 141 "", /* 17 unused */ 142 "integer divide fault", /* 18 T_DIVIDE */ 143 "non-maskable interrupt trap", /* 19 T_NMI */ 144 "overflow trap", /* 20 T_OFLOW */ 145 "FPU bounds check fault", /* 21 T_BOUND */ 146 "FPU device not available", /* 22 T_DNA */ 147 "double fault", /* 23 T_DOUBLEFLT */ 148 "FPU operand fetch fault", /* 24 T_FPOPFLT */ 149 "invalid TSS fault", /* 25 T_TSSFLT */ 150 "segment not present fault", /* 26 T_SEGNPFLT */ 151 "stack fault", /* 27 T_STKFLT */ 152 "machine check trap", /* 28 T_MCHK */ 153 "SIMD floating-point exception", /* 29 T_XMMFLT */ 154 "reserved (unknown) fault", /* 30 T_RESERVED */ 155 }; 156 157 #ifdef DDB 158 static int ddb_on_nmi = 1; 159 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW, 160 &ddb_on_nmi, 0, "Go to DDB on NMI"); 161 #endif 162 static int panic_on_nmi = 1; 163 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW, 164 &panic_on_nmi, 0, "Panic on NMI"); 165 static int fast_release; 166 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW, 167 &fast_release, 0, "Passive Release was optimal"); 168 static int slow_release; 169 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW, 170 &slow_release, 0, "Passive Release was nonoptimal"); 171 172 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure"); 173 extern int max_sysmsg; 174 175 /* 176 * Passively intercepts the thread switch function to increase the thread 177 * priority from a user priority to a kernel priority, reducing 178 * syscall and trap overhead for the case where no switch occurs. 179 * 180 * Synchronizes td_ucred with p_ucred. This is used by system calls, 181 * signal handling, faults, AST traps, and anything else that enters the 182 * kernel from userland and provides the kernel with a stable read-only 183 * copy of the process ucred. 184 */ 185 static __inline void 186 userenter(struct thread *curtd, struct proc *curp) 187 { 188 struct ucred *ocred; 189 struct ucred *ncred; 190 191 curtd->td_release = lwkt_passive_release; 192 193 if (curtd->td_ucred != curp->p_ucred) { 194 ncred = crhold(curp->p_ucred); 195 ocred = curtd->td_ucred; 196 curtd->td_ucred = ncred; 197 if (ocred) 198 crfree(ocred); 199 } 200 } 201 202 /* 203 * Handle signals, upcalls, profiling, and other AST's and/or tasks that 204 * must be completed before we can return to or try to return to userland. 205 * 206 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64 207 * arithmatic on the delta calculation so the absolute tick values are 208 * truncated to an integer. 209 */ 210 static void 211 userret(struct lwp *lp, struct trapframe *frame, int sticks) 212 { 213 struct proc *p = lp->lwp_proc; 214 int sig; 215 216 /* 217 * Charge system time if profiling. Note: times are in microseconds. 218 * This may do a copyout and block, so do it first even though it 219 * means some system time will be charged as user time. 220 */ 221 if (p->p_flags & P_PROFIL) { 222 addupc_task(p, frame->tf_rip, 223 (u_int)((int)lp->lwp_thread->td_sticks - sticks)); 224 } 225 226 recheck: 227 /* 228 * If the jungle wants us dead, so be it. 229 */ 230 if (lp->lwp_mpflags & LWP_MP_WEXIT) { 231 lwkt_gettoken(&p->p_token); 232 lwp_exit(0); 233 lwkt_reltoken(&p->p_token); /* NOT REACHED */ 234 } 235 236 /* 237 * Block here if we are in a stopped state. 238 */ 239 if (p->p_stat == SSTOP) { 240 lwkt_gettoken(&p->p_token); 241 tstop(); 242 lwkt_reltoken(&p->p_token); 243 goto recheck; 244 } 245 246 /* 247 * Post any pending upcalls. If running a virtual kernel be sure 248 * to restore the virtual kernel's vmspace before posting the upcall. 249 */ 250 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF | P_UPCALLPEND)) { 251 lwkt_gettoken(&p->p_token); 252 if (p->p_flags & P_SIGVTALRM) { 253 p->p_flags &= ~P_SIGVTALRM; 254 ksignal(p, SIGVTALRM); 255 } 256 if (p->p_flags & P_SIGPROF) { 257 p->p_flags &= ~P_SIGPROF; 258 ksignal(p, SIGPROF); 259 } 260 if (p->p_flags & P_UPCALLPEND) { 261 p->p_flags &= ~P_UPCALLPEND; 262 postupcall(lp); 263 } 264 lwkt_reltoken(&p->p_token); 265 goto recheck; 266 } 267 268 /* 269 * Post any pending signals 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 * Make sure postsig() handled request to restore old signal mask after 300 * running signal handler. 301 */ 302 KKASSERT((lp->lwp_flags & LWP_OLDMASK) == 0); 303 } 304 305 /* 306 * Cleanup from userenter and any passive release that might have occured. 307 * We must reclaim the current-process designation before we can return 308 * to usermode. We also handle both LWKT and USER reschedule requests. 309 */ 310 static __inline void 311 userexit(struct lwp *lp) 312 { 313 struct thread *td = lp->lwp_thread; 314 /* globaldata_t gd = td->td_gd; */ 315 316 /* 317 * Handle stop requests at kernel priority. Any requests queued 318 * after this loop will generate another AST. 319 */ 320 while (lp->lwp_proc->p_stat == SSTOP) { 321 lwkt_gettoken(&lp->lwp_proc->p_token); 322 tstop(); 323 lwkt_reltoken(&lp->lwp_proc->p_token); 324 } 325 326 /* 327 * Reduce our priority in preparation for a return to userland. If 328 * our passive release function was still in place, our priority was 329 * never raised and does not need to be reduced. 330 */ 331 lwkt_passive_recover(td); 332 333 /* 334 * Become the current user scheduled process if we aren't already, 335 * and deal with reschedule requests and other factors. 336 */ 337 lp->lwp_proc->p_usched->acquire_curproc(lp); 338 /* WARNING: we may have migrated cpu's */ 339 /* gd = td->td_gd; */ 340 } 341 342 #if !defined(KTR_KERNENTRY) 343 #define KTR_KERNENTRY KTR_ALL 344 #endif 345 KTR_INFO_MASTER(kernentry); 346 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p", 347 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t)); 348 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d", 349 sizeof(int) + sizeof(int)); 350 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d", 351 sizeof(int) + sizeof(int) + sizeof(int)); 352 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d", 353 sizeof(int) + sizeof(int) + sizeof(int)); 354 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d", 355 sizeof(int) + sizeof(int)); 356 357 /* 358 * Exception, fault, and trap interface to the kernel. 359 * This common code is called from assembly language IDT gate entry 360 * routines that prepare a suitable stack frame, and restore this 361 * frame after the exception has been processed. 362 * 363 * This function is also called from doreti in an interlock to handle ASTs. 364 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap 365 * 366 * NOTE! We have to retrieve the fault address prior to obtaining the 367 * MP lock because get_mplock() may switch out. YYY cr2 really ought 368 * to be retrieved by the assembly code, not here. 369 * 370 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing 371 * if an attempt is made to switch from a fast interrupt or IPI. This is 372 * necessary to properly take fatal kernel traps on SMP machines if 373 * get_mplock() has to block. 374 */ 375 376 void 377 user_trap(struct trapframe *frame) 378 { 379 struct globaldata *gd = mycpu; 380 struct thread *td = gd->gd_curthread; 381 struct lwp *lp = td->td_lwp; 382 struct proc *p; 383 int sticks = 0; 384 int i = 0, ucode = 0, type, code; 385 #ifdef SMP 386 int have_mplock = 0; 387 #endif 388 #ifdef INVARIANTS 389 int crit_count = td->td_critcount; 390 lwkt_tokref_t curstop = td->td_toks_stop; 391 #endif 392 vm_offset_t eva; 393 394 p = td->td_proc; 395 396 if (frame->tf_trapno == T_PAGEFLT) 397 eva = frame->tf_addr; 398 else 399 eva = 0; 400 #if 0 401 kprintf("USER_TRAP AT %08lx xflags %ld trapno %ld eva %08lx\n", 402 frame->tf_rip, frame->tf_xflags, frame->tf_trapno, eva); 403 #endif 404 405 /* 406 * Everything coming from user mode runs through user_trap, 407 * including system calls. 408 */ 409 if (frame->tf_trapno == T_FAST_SYSCALL) { 410 syscall2(frame); 411 return; 412 } 413 414 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid, 415 frame->tf_trapno, eva); 416 417 #ifdef DDB 418 if (db_active) { 419 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0); 420 ++gd->gd_trap_nesting_level; 421 MAKEMPSAFE(have_mplock); 422 trap_fatal(frame, TRUE, eva); 423 --gd->gd_trap_nesting_level; 424 goto out2; 425 } 426 #endif 427 428 type = frame->tf_trapno; 429 code = frame->tf_err; 430 431 userenter(td, p); 432 433 sticks = (int)td->td_sticks; 434 lp->lwp_md.md_regs = frame; 435 436 switch (type) { 437 case T_PRIVINFLT: /* privileged instruction fault */ 438 ucode = type; 439 i = SIGILL; 440 break; 441 442 case T_BPTFLT: /* bpt instruction fault */ 443 case T_TRCTRAP: /* trace trap */ 444 frame->tf_rflags &= ~PSL_T; 445 i = SIGTRAP; 446 break; 447 448 case T_ARITHTRAP: /* arithmetic trap */ 449 ucode = code; 450 i = SIGFPE; 451 break; 452 453 case T_ASTFLT: /* Allow process switch */ 454 mycpu->gd_cnt.v_soft++; 455 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) { 456 atomic_clear_int(&mycpu->gd_reqflags, RQF_AST_OWEUPC); 457 addupc_task(p, p->p_prof.pr_addr, p->p_prof.pr_ticks); 458 } 459 goto out; 460 461 /* 462 * The following two traps can happen in 463 * vm86 mode, and, if so, we want to handle 464 * them specially. 465 */ 466 case T_PROTFLT: /* general protection fault */ 467 case T_STKFLT: /* stack fault */ 468 #if 0 469 if (frame->tf_eflags & PSL_VM) { 470 i = vm86_emulate((struct vm86frame *)frame); 471 if (i == 0) 472 goto out; 473 break; 474 } 475 #endif 476 /* FALL THROUGH */ 477 478 case T_SEGNPFLT: /* segment not present fault */ 479 case T_TSSFLT: /* invalid TSS fault */ 480 case T_DOUBLEFLT: /* double fault */ 481 default: 482 ucode = code + BUS_SEGM_FAULT ; 483 i = SIGBUS; 484 break; 485 486 case T_PAGEFLT: /* page fault */ 487 MAKEMPSAFE(have_mplock); 488 i = trap_pfault(frame, TRUE, eva); 489 if (i == -1 || i == 0) 490 goto out; 491 492 ucode = T_PAGEFLT; 493 break; 494 495 case T_DIVIDE: /* integer divide fault */ 496 ucode = FPE_INTDIV; 497 i = SIGFPE; 498 break; 499 500 #if NISA > 0 501 case T_NMI: 502 MAKEMPSAFE(have_mplock); 503 /* machine/parity/power fail/"kitchen sink" faults */ 504 if (isa_nmi(code) == 0) { 505 #ifdef DDB 506 /* 507 * NMI can be hooked up to a pushbutton 508 * for debugging. 509 */ 510 if (ddb_on_nmi) { 511 kprintf ("NMI ... going to debugger\n"); 512 kdb_trap (type, 0, frame); 513 } 514 #endif /* DDB */ 515 goto out2; 516 } else if (panic_on_nmi) 517 panic("NMI indicates hardware failure"); 518 break; 519 #endif /* NISA > 0 */ 520 521 case T_OFLOW: /* integer overflow fault */ 522 ucode = FPE_INTOVF; 523 i = SIGFPE; 524 break; 525 526 case T_BOUND: /* bounds check fault */ 527 ucode = FPE_FLTSUB; 528 i = SIGFPE; 529 break; 530 531 case T_DNA: 532 /* 533 * Virtual kernel intercept - pass the DNA exception 534 * to the (emulated) virtual kernel if it asked to handle 535 * it. This occurs when the virtual kernel is holding 536 * onto the FP context for a different emulated 537 * process then the one currently running. 538 * 539 * We must still call npxdna() since we may have 540 * saved FP state that the (emulated) virtual kernel 541 * needs to hand over to a different emulated process. 542 */ 543 if (lp->lwp_vkernel && lp->lwp_vkernel->ve && 544 (td->td_pcb->pcb_flags & FP_VIRTFP) 545 ) { 546 npxdna(frame); 547 break; 548 } 549 /* 550 * The kernel may have switched out the FP unit's 551 * state, causing the user process to take a fault 552 * when it tries to use the FP unit. Restore the 553 * state here 554 */ 555 if (npxdna(frame)) 556 goto out; 557 if (!pmath_emulate) { 558 i = SIGFPE; 559 ucode = FPE_FPU_NP_TRAP; 560 break; 561 } 562 i = (*pmath_emulate)(frame); 563 if (i == 0) { 564 if (!(frame->tf_rflags & PSL_T)) 565 goto out2; 566 frame->tf_rflags &= ~PSL_T; 567 i = SIGTRAP; 568 } 569 /* else ucode = emulator_only_knows() XXX */ 570 break; 571 572 case T_FPOPFLT: /* FPU operand fetch fault */ 573 ucode = T_FPOPFLT; 574 i = SIGILL; 575 break; 576 577 case T_XMMFLT: /* SIMD floating-point exception */ 578 ucode = 0; /* XXX */ 579 i = SIGFPE; 580 break; 581 } 582 583 /* 584 * Virtual kernel intercept - if the fault is directly related to a 585 * VM context managed by a virtual kernel then let the virtual kernel 586 * handle it. 587 */ 588 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 589 vkernel_trap(lp, frame); 590 goto out; 591 } 592 593 /* 594 * Translate fault for emulators (e.g. Linux) 595 */ 596 if (*p->p_sysent->sv_transtrap) 597 i = (*p->p_sysent->sv_transtrap)(i, type); 598 599 MAKEMPSAFE(have_mplock); 600 trapsignal(lp, i, ucode); 601 602 #ifdef DEBUG 603 if (type <= MAX_TRAP_MSG) { 604 uprintf("fatal process exception: %s", 605 trap_msg[type]); 606 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 607 uprintf(", fault VA = 0x%lx", (u_long)eva); 608 uprintf("\n"); 609 } 610 #endif 611 612 out: 613 userret(lp, frame, sticks); 614 userexit(lp); 615 out2: ; 616 #ifdef SMP 617 if (have_mplock) 618 rel_mplock(); 619 #endif 620 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid); 621 #ifdef INVARIANTS 622 KASSERT(crit_count == td->td_critcount, 623 ("trap: critical section count mismatch! %d/%d", 624 crit_count, td->td_pri)); 625 KASSERT(curstop == td->td_toks_stop, 626 ("trap: extra tokens held after trap! %ld/%ld", 627 curstop - &td->td_toks_base, 628 td->td_toks_stop - &td->td_toks_base)); 629 #endif 630 } 631 632 void 633 kern_trap(struct trapframe *frame) 634 { 635 struct globaldata *gd = mycpu; 636 struct thread *td = gd->gd_curthread; 637 struct lwp *lp; 638 struct proc *p; 639 int i = 0, ucode = 0, type, code; 640 #ifdef SMP 641 int have_mplock = 0; 642 #endif 643 #ifdef INVARIANTS 644 int crit_count = td->td_critcount; 645 lwkt_tokref_t curstop = td->td_toks_stop; 646 #endif 647 vm_offset_t eva; 648 649 lp = td->td_lwp; 650 p = td->td_proc; 651 652 if (frame->tf_trapno == T_PAGEFLT) 653 eva = frame->tf_addr; 654 else 655 eva = 0; 656 657 #ifdef DDB 658 if (db_active) { 659 ++gd->gd_trap_nesting_level; 660 MAKEMPSAFE(have_mplock); 661 trap_fatal(frame, FALSE, eva); 662 --gd->gd_trap_nesting_level; 663 goto out2; 664 } 665 #endif 666 667 type = frame->tf_trapno; 668 code = frame->tf_err; 669 670 #if 0 671 kernel_trap: 672 #endif 673 /* kernel trap */ 674 675 switch (type) { 676 case T_PAGEFLT: /* page fault */ 677 MAKEMPSAFE(have_mplock); 678 trap_pfault(frame, FALSE, eva); 679 goto out2; 680 681 case T_DNA: 682 /* 683 * The kernel may be using npx for copying or other 684 * purposes. 685 */ 686 panic("kernel NPX should not happen"); 687 if (npxdna(frame)) 688 goto out2; 689 break; 690 691 case T_PROTFLT: /* general protection fault */ 692 case T_SEGNPFLT: /* segment not present fault */ 693 /* 694 * Invalid segment selectors and out of bounds 695 * %eip's and %esp's can be set up in user mode. 696 * This causes a fault in kernel mode when the 697 * kernel tries to return to user mode. We want 698 * to get this fault so that we can fix the 699 * problem here and not have to check all the 700 * selectors and pointers when the user changes 701 * them. 702 */ 703 if (mycpu->gd_intr_nesting_level == 0) { 704 if (td->td_pcb->pcb_onfault) { 705 frame->tf_rip = 706 (register_t)td->td_pcb->pcb_onfault; 707 goto out2; 708 } 709 } 710 break; 711 712 case T_TSSFLT: 713 /* 714 * PSL_NT can be set in user mode and isn't cleared 715 * automatically when the kernel is entered. This 716 * causes a TSS fault when the kernel attempts to 717 * `iret' because the TSS link is uninitialized. We 718 * want to get this fault so that we can fix the 719 * problem here and not every time the kernel is 720 * entered. 721 */ 722 if (frame->tf_rflags & PSL_NT) { 723 frame->tf_rflags &= ~PSL_NT; 724 goto out2; 725 } 726 break; 727 728 case T_TRCTRAP: /* trace trap */ 729 #if 0 730 if (frame->tf_eip == (int)IDTVEC(syscall)) { 731 /* 732 * We've just entered system mode via the 733 * syscall lcall. Continue single stepping 734 * silently until the syscall handler has 735 * saved the flags. 736 */ 737 goto out2; 738 } 739 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) { 740 /* 741 * The syscall handler has now saved the 742 * flags. Stop single stepping it. 743 */ 744 frame->tf_eflags &= ~PSL_T; 745 goto out2; 746 } 747 #endif 748 #if 0 749 /* 750 * Ignore debug register trace traps due to 751 * accesses in the user's address space, which 752 * can happen under several conditions such as 753 * if a user sets a watchpoint on a buffer and 754 * then passes that buffer to a system call. 755 * We still want to get TRCTRAPS for addresses 756 * in kernel space because that is useful when 757 * debugging the kernel. 758 */ 759 if (user_dbreg_trap()) { 760 /* 761 * Reset breakpoint bits because the 762 * processor doesn't 763 */ 764 load_dr6(rdr6() & 0xfffffff0); 765 goto out2; 766 } 767 #endif 768 /* 769 * Fall through (TRCTRAP kernel mode, kernel address) 770 */ 771 case T_BPTFLT: 772 /* 773 * If DDB is enabled, let it handle the debugger trap. 774 * Otherwise, debugger traps "can't happen". 775 */ 776 #ifdef DDB 777 MAKEMPSAFE(have_mplock); 778 if (kdb_trap (type, 0, frame)) 779 goto out2; 780 #endif 781 break; 782 case T_DIVIDE: 783 MAKEMPSAFE(have_mplock); 784 trap_fatal(frame, FALSE, eva); 785 goto out2; 786 case T_NMI: 787 MAKEMPSAFE(have_mplock); 788 trap_fatal(frame, FALSE, eva); 789 goto out2; 790 case T_SYSCALL80: 791 case T_FAST_SYSCALL: 792 /* 793 * Ignore this trap generated from a spurious SIGTRAP. 794 * 795 * single stepping in / syscalls leads to spurious / SIGTRAP 796 * so ignore 797 * 798 * Haiku (c) 2007 Simon 'corecode' Schubert 799 */ 800 goto out2; 801 } 802 803 /* 804 * Translate fault for emulators (e.g. Linux) 805 */ 806 if (*p->p_sysent->sv_transtrap) 807 i = (*p->p_sysent->sv_transtrap)(i, type); 808 809 MAKEMPSAFE(have_mplock); 810 trapsignal(lp, i, ucode); 811 812 #ifdef DEBUG 813 if (type <= MAX_TRAP_MSG) { 814 uprintf("fatal process exception: %s", 815 trap_msg[type]); 816 if ((type == T_PAGEFLT) || (type == T_PROTFLT)) 817 uprintf(", fault VA = 0x%lx", (u_long)eva); 818 uprintf("\n"); 819 } 820 #endif 821 822 out2: 823 ; 824 #ifdef SMP 825 if (have_mplock) 826 rel_mplock(); 827 #endif 828 #ifdef INVARIANTS 829 KASSERT(crit_count == td->td_critcount, 830 ("trap: critical section count mismatch! %d/%d", 831 crit_count, td->td_pri)); 832 KASSERT(curstop == td->td_toks_stop, 833 ("trap: extra tokens held after trap! %ld/%ld", 834 curstop - &td->td_toks_base, 835 td->td_toks_stop - &td->td_toks_base)); 836 #endif 837 } 838 839 int 840 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva) 841 { 842 vm_offset_t va; 843 struct vmspace *vm = NULL; 844 vm_map_t map = 0; 845 int rv = 0; 846 vm_prot_t ftype; 847 thread_t td = curthread; 848 struct lwp *lp = td->td_lwp; 849 int fault_flags; 850 851 va = trunc_page(eva); 852 if (usermode == FALSE) { 853 /* 854 * This is a fault on kernel virtual memory. 855 */ 856 map = &kernel_map; 857 } else { 858 /* 859 * This is a fault on non-kernel virtual memory. 860 * vm is initialized above to NULL. If curproc is NULL 861 * or curproc->p_vmspace is NULL the fault is fatal. 862 */ 863 if (lp != NULL) 864 vm = lp->lwp_vmspace; 865 866 if (vm == NULL) 867 goto nogo; 868 869 map = &vm->vm_map; 870 } 871 872 if (frame->tf_err & PGEX_W) 873 ftype = VM_PROT_READ | VM_PROT_WRITE; 874 else 875 ftype = VM_PROT_READ; 876 877 if (map != &kernel_map) { 878 /* 879 * Keep swapout from messing with us during this 880 * critical time. 881 */ 882 PHOLD(lp->lwp_proc); 883 884 /* 885 * Grow the stack if necessary 886 */ 887 /* grow_stack returns false only if va falls into 888 * a growable stack region and the stack growth 889 * fails. It returns true if va was not within 890 * a growable stack region, or if the stack 891 * growth succeeded. 892 */ 893 if (!grow_stack (lp->lwp_proc, va)) { 894 rv = KERN_FAILURE; 895 PRELE(lp->lwp_proc); 896 goto nogo; 897 } 898 899 fault_flags = 0; 900 if (usermode) 901 fault_flags |= VM_FAULT_BURST; 902 if (ftype & VM_PROT_WRITE) 903 fault_flags |= VM_FAULT_DIRTY; 904 else 905 fault_flags |= VM_FAULT_NORMAL; 906 907 rv = vm_fault(map, va, ftype, fault_flags); 908 PRELE(lp->lwp_proc); 909 } else { 910 /* 911 * Don't have to worry about process locking or stacks in the kernel. 912 */ 913 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL); 914 } 915 916 if (rv == KERN_SUCCESS) 917 return (0); 918 nogo: 919 if (!usermode) { 920 if (td->td_gd->gd_intr_nesting_level == 0 && 921 td->td_pcb->pcb_onfault) { 922 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault; 923 return (0); 924 } 925 trap_fatal(frame, usermode, eva); 926 return (-1); 927 } 928 929 /* 930 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no 931 * kludge is needed to pass the fault address to signal handlers. 932 */ 933 struct proc *p = td->td_proc; 934 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n", 935 (void *)va, (void *)frame->tf_rip, p->p_pid, p->p_comm); 936 /* Debugger("seg-fault"); */ 937 938 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV); 939 } 940 941 static void 942 trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva) 943 { 944 int code, type, ss; 945 long rsp; 946 947 code = frame->tf_xflags; 948 type = frame->tf_trapno; 949 950 if (type <= MAX_TRAP_MSG) { 951 kprintf("\n\nFatal trap %d: %s while in %s mode\n", 952 type, trap_msg[type], 953 (usermode ? "user" : "kernel")); 954 } 955 #ifdef SMP 956 /* two separate prints in case of a trap on an unmapped page */ 957 kprintf("cpuid = %d\n", mycpu->gd_cpuid); 958 #endif 959 if (type == T_PAGEFLT) { 960 kprintf("fault virtual address = %p\n", (void *)eva); 961 kprintf("fault code = %s %s, %s\n", 962 usermode ? "user" : "supervisor", 963 code & PGEX_W ? "write" : "read", 964 code & PGEX_P ? "protection violation" : "page not present"); 965 } 966 kprintf("instruction pointer = 0x%lx:0x%lx\n", 967 frame->tf_cs & 0xffff, frame->tf_rip); 968 if (usermode) { 969 ss = frame->tf_ss & 0xffff; 970 rsp = frame->tf_rsp; 971 } else { 972 ss = GSEL(GDATA_SEL, SEL_KPL); 973 rsp = (long)&frame->tf_rsp; 974 } 975 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp); 976 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp); 977 kprintf("processor eflags = "); 978 if (frame->tf_rflags & PSL_T) 979 kprintf("trace trap, "); 980 if (frame->tf_rflags & PSL_I) 981 kprintf("interrupt enabled, "); 982 if (frame->tf_rflags & PSL_NT) 983 kprintf("nested task, "); 984 if (frame->tf_rflags & PSL_RF) 985 kprintf("resume, "); 986 #if 0 987 if (frame->tf_eflags & PSL_VM) 988 kprintf("vm86, "); 989 #endif 990 kprintf("IOPL = %jd\n", (intmax_t)((frame->tf_rflags & PSL_IOPL) >> 12)); 991 kprintf("current process = "); 992 if (curproc) { 993 kprintf("%lu (%s)\n", 994 (u_long)curproc->p_pid, curproc->p_comm ? 995 curproc->p_comm : ""); 996 } else { 997 kprintf("Idle\n"); 998 } 999 kprintf("current thread = pri %d ", curthread->td_pri); 1000 if (curthread->td_critcount) 1001 kprintf("(CRIT)"); 1002 kprintf("\n"); 1003 #ifdef SMP 1004 /** 1005 * XXX FIXME: 1006 * we probably SHOULD have stopped the other CPUs before now! 1007 * another CPU COULD have been touching cpl at this moment... 1008 */ 1009 kprintf(" <- SMP: XXX"); 1010 #endif 1011 kprintf("\n"); 1012 1013 #ifdef KDB 1014 if (kdb_trap(&psl)) 1015 return; 1016 #endif 1017 #ifdef DDB 1018 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame)) 1019 return; 1020 #endif 1021 kprintf("trap number = %d\n", type); 1022 if (type <= MAX_TRAP_MSG) 1023 panic("%s", trap_msg[type]); 1024 else 1025 panic("unknown/reserved trap"); 1026 } 1027 1028 /* 1029 * Double fault handler. Called when a fault occurs while writing 1030 * a frame for a trap/exception onto the stack. This usually occurs 1031 * when the stack overflows (such is the case with infinite recursion, 1032 * for example). 1033 * 1034 * XXX Note that the current PTD gets replaced by IdlePTD when the 1035 * task switch occurs. This means that the stack that was active at 1036 * the time of the double fault is not available at <kstack> unless 1037 * the machine was idle when the double fault occurred. The downside 1038 * of this is that "trace <ebp>" in ddb won't work. 1039 */ 1040 void 1041 dblfault_handler(void) 1042 { 1043 #if JG 1044 struct mdglobaldata *gd = mdcpu; 1045 #endif 1046 1047 kprintf("\nFatal double fault:\n"); 1048 #if JG 1049 kprintf("rip = 0x%lx\n", gd->gd_common_tss.tss_rip); 1050 kprintf("rsp = 0x%lx\n", gd->gd_common_tss.tss_rsp); 1051 kprintf("rbp = 0x%lx\n", gd->gd_common_tss.tss_rbp); 1052 #endif 1053 #ifdef SMP 1054 /* two separate prints in case of a trap on an unmapped page */ 1055 kprintf("cpuid = %d\n", mycpu->gd_cpuid); 1056 #endif 1057 panic("double fault"); 1058 } 1059 1060 /* 1061 * Compensate for 386 brain damage (missing URKR). 1062 * This is a little simpler than the pagefault handler in trap() because 1063 * it the page tables have already been faulted in and high addresses 1064 * are thrown out early for other reasons. 1065 */ 1066 int 1067 trapwrite(unsigned addr) 1068 { 1069 struct lwp *lp; 1070 vm_offset_t va; 1071 struct vmspace *vm; 1072 int rv; 1073 1074 va = trunc_page((vm_offset_t)addr); 1075 /* 1076 * XXX - MAX is END. Changed > to >= for temp. fix. 1077 */ 1078 if (va >= VM_MAX_USER_ADDRESS) 1079 return (1); 1080 1081 lp = curthread->td_lwp; 1082 vm = lp->lwp_vmspace; 1083 1084 PHOLD(lp->lwp_proc); 1085 1086 if (!grow_stack (lp->lwp_proc, va)) { 1087 PRELE(lp->lwp_proc); 1088 return (1); 1089 } 1090 1091 /* 1092 * fault the data page 1093 */ 1094 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY); 1095 1096 PRELE(lp->lwp_proc); 1097 1098 if (rv != KERN_SUCCESS) 1099 return 1; 1100 1101 return (0); 1102 } 1103 1104 /* 1105 * syscall2 - MP aware system call request C handler 1106 * 1107 * A system call is essentially treated as a trap except that the 1108 * MP lock is not held on entry or return. We are responsible for 1109 * obtaining the MP lock if necessary and for handling ASTs 1110 * (e.g. a task switch) prior to return. 1111 * 1112 * In general, only simple access and manipulation of curproc and 1113 * the current stack is allowed without having to hold MP lock. 1114 * 1115 * MPSAFE - note that large sections of this routine are run without 1116 * the MP lock. 1117 */ 1118 void 1119 syscall2(struct trapframe *frame) 1120 { 1121 struct thread *td = curthread; 1122 struct proc *p = td->td_proc; 1123 struct lwp *lp = td->td_lwp; 1124 caddr_t params; 1125 struct sysent *callp; 1126 register_t orig_tf_rflags; 1127 int sticks; 1128 int error; 1129 int narg; 1130 #ifdef INVARIANTS 1131 int crit_count = td->td_critcount; 1132 lwkt_tokref_t curstop = td->td_toks_stop; 1133 #endif 1134 #ifdef SMP 1135 int have_mplock = 0; 1136 #endif 1137 register_t *argp; 1138 u_int code; 1139 int reg, regcnt; 1140 union sysunion args; 1141 register_t *argsdst; 1142 1143 mycpu->gd_cnt.v_syscall++; 1144 1145 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid, 1146 frame->tf_eax); 1147 1148 userenter(td, p); /* lazy raise our priority */ 1149 1150 reg = 0; 1151 regcnt = 6; 1152 /* 1153 * Misc 1154 */ 1155 sticks = (int)td->td_sticks; 1156 orig_tf_rflags = frame->tf_rflags; 1157 1158 /* 1159 * Virtual kernel intercept - if a VM context managed by a virtual 1160 * kernel issues a system call the virtual kernel handles it, not us. 1161 * Restore the virtual kernel context and return from its system 1162 * call. The current frame is copied out to the virtual kernel. 1163 */ 1164 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 1165 vkernel_trap(lp, frame); 1166 error = EJUSTRETURN; 1167 goto out; 1168 } 1169 1170 /* 1171 * Get the system call parameters and account for time 1172 */ 1173 lp->lwp_md.md_regs = frame; 1174 params = (caddr_t)frame->tf_rsp + sizeof(register_t); 1175 code = frame->tf_rax; 1176 1177 if (p->p_sysent->sv_prepsyscall) { 1178 (*p->p_sysent->sv_prepsyscall)( 1179 frame, (int *)(&args.nosys.sysmsg + 1), 1180 &code, ¶ms); 1181 } else { 1182 if (code == SYS_syscall || code == SYS___syscall) { 1183 code = frame->tf_rdi; 1184 reg++; 1185 regcnt--; 1186 } 1187 } 1188 1189 if (p->p_sysent->sv_mask) 1190 code &= p->p_sysent->sv_mask; 1191 1192 if (code >= p->p_sysent->sv_size) 1193 callp = &p->p_sysent->sv_table[0]; 1194 else 1195 callp = &p->p_sysent->sv_table[code]; 1196 1197 narg = callp->sy_narg & SYF_ARGMASK; 1198 1199 /* 1200 * On x86_64 we get up to six arguments in registers. The rest are 1201 * on the stack. The first six members of 'struct trapframe' happen 1202 * to be the registers used to pass arguments, in exactly the right 1203 * order. 1204 */ 1205 argp = &frame->tf_rdi; 1206 argp += reg; 1207 argsdst = (register_t *)(&args.nosys.sysmsg + 1); 1208 /* 1209 * JG can we overflow the space pointed to by 'argsdst' 1210 * either with 'bcopy' or with 'copyin'? 1211 */ 1212 bcopy(argp, argsdst, sizeof(register_t) * regcnt); 1213 /* 1214 * copyin is MP aware, but the tracing code is not 1215 */ 1216 if (narg > regcnt) { 1217 KASSERT(params != NULL, ("copyin args with no params!")); 1218 error = copyin(params, &argsdst[regcnt], 1219 (narg - regcnt) * sizeof(register_t)); 1220 if (error) { 1221 #ifdef KTRACE 1222 if (KTRPOINT(td, KTR_SYSCALL)) { 1223 MAKEMPSAFE(have_mplock); 1224 1225 ktrsyscall(lp, code, narg, 1226 (void *)(&args.nosys.sysmsg + 1)); 1227 } 1228 #endif 1229 goto bad; 1230 } 1231 } 1232 1233 #ifdef KTRACE 1234 if (KTRPOINT(td, KTR_SYSCALL)) { 1235 MAKEMPSAFE(have_mplock); 1236 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1)); 1237 } 1238 #endif 1239 1240 /* 1241 * Default return value is 0 (will be copied to %rax). Double-value 1242 * returns use %rax and %rdx. %rdx is left unchanged for system 1243 * calls which return only one result. 1244 */ 1245 args.sysmsg_fds[0] = 0; 1246 args.sysmsg_fds[1] = frame->tf_rdx; 1247 1248 /* 1249 * The syscall might manipulate the trap frame. If it does it 1250 * will probably return EJUSTRETURN. 1251 */ 1252 args.sysmsg_frame = frame; 1253 1254 STOPEVENT(p, S_SCE, narg); /* MP aware */ 1255 1256 /* 1257 * NOTE: All system calls run MPSAFE now. The system call itself 1258 * is responsible for getting the MP lock. 1259 */ 1260 error = (*callp->sy_call)(&args); 1261 1262 #if 0 1263 kprintf("system call %d returned %d\n", code, error); 1264 #endif 1265 1266 out: 1267 /* 1268 * MP SAFE (we may or may not have the MP lock at this point) 1269 */ 1270 switch (error) { 1271 case 0: 1272 /* 1273 * Reinitialize proc pointer `p' as it may be different 1274 * if this is a child returning from fork syscall. 1275 */ 1276 p = curproc; 1277 lp = curthread->td_lwp; 1278 frame->tf_rax = args.sysmsg_fds[0]; 1279 frame->tf_rdx = args.sysmsg_fds[1]; 1280 frame->tf_rflags &= ~PSL_C; 1281 break; 1282 case ERESTART: 1283 /* 1284 * Reconstruct pc, we know that 'syscall' is 2 bytes. 1285 * We have to do a full context restore so that %r10 1286 * (which was holding the value of %rcx) is restored for 1287 * the next iteration. 1288 */ 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, 0); 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 #ifdef SMP 1339 /* 1340 * Release the MP lock if we had to get it 1341 */ 1342 if (have_mplock) 1343 rel_mplock(); 1344 #endif 1345 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error); 1346 #ifdef INVARIANTS 1347 KASSERT(&td->td_toks_base == td->td_toks_stop, 1348 ("syscall: critical section count mismatch! %d/%d", 1349 crit_count, td->td_pri)); 1350 KASSERT(curstop == td->td_toks_stop, 1351 ("syscall: extra tokens held after trap! %ld", 1352 td->td_toks_stop - &td->td_toks_base)); 1353 #endif 1354 } 1355 1356 /* 1357 * NOTE: mplock not held at any point 1358 */ 1359 void 1360 fork_return(struct lwp *lp, struct trapframe *frame) 1361 { 1362 frame->tf_rax = 0; /* Child returns zero */ 1363 frame->tf_rflags &= ~PSL_C; /* success */ 1364 frame->tf_rdx = 1; 1365 1366 generic_lwp_return(lp, frame); 1367 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid); 1368 } 1369 1370 /* 1371 * Simplified back end of syscall(), used when returning from fork() 1372 * directly into user mode. 1373 * 1374 * This code will return back into the fork trampoline code which then 1375 * runs doreti. 1376 * 1377 * NOTE: The mplock is not held at any point. 1378 */ 1379 void 1380 generic_lwp_return(struct lwp *lp, struct trapframe *frame) 1381 { 1382 struct proc *p = lp->lwp_proc; 1383 1384 /* 1385 * Newly forked processes are given a kernel priority. We have to 1386 * adjust the priority to a normal user priority and fake entry 1387 * into the kernel (call userenter()) to install a passive release 1388 * function just in case userret() decides to stop the process. This 1389 * can occur when ^Z races a fork. If we do not install the passive 1390 * release function the current process designation will not be 1391 * released when the thread goes to sleep. 1392 */ 1393 lwkt_setpri_self(TDPRI_USER_NORM); 1394 userenter(lp->lwp_thread, p); 1395 userret(lp, frame, 0); 1396 #ifdef KTRACE 1397 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET)) 1398 ktrsysret(lp, SYS_fork, 0, 0); 1399 #endif 1400 lp->lwp_flags |= LWP_PASSIVE_ACQ; 1401 userexit(lp); 1402 lp->lwp_flags &= ~LWP_PASSIVE_ACQ; 1403 } 1404 1405 /* 1406 * doreti has turned into this. The frame is directly on the stack. We 1407 * pull everything else we need (fpu and tls context) from the current 1408 * thread. 1409 * 1410 * Note on fpu interactions: In a virtual kernel, the fpu context for 1411 * an emulated user mode process is not shared with the virtual kernel's 1412 * fpu context, so we only have to 'stack' fpu contexts within the virtual 1413 * kernel itself, and not even then since the signal() contexts that we care 1414 * about save and restore the FPU state (I think anyhow). 1415 * 1416 * vmspace_ctl() returns an error only if it had problems instaling the 1417 * context we supplied or problems copying data to/from our VM space. 1418 */ 1419 void 1420 go_user(struct intrframe *frame) 1421 { 1422 struct trapframe *tf = (void *)&frame->if_rdi; 1423 int r; 1424 1425 /* 1426 * Interrupts may be disabled on entry, make sure all signals 1427 * can be received before beginning our loop. 1428 */ 1429 sigsetmask(0); 1430 1431 /* 1432 * Switch to the current simulated user process, then call 1433 * user_trap() when we break out of it (usually due to a signal). 1434 */ 1435 for (;;) { 1436 /* 1437 * Tell the real kernel whether it is ok to use the FP 1438 * unit or not. 1439 */ 1440 if (mdcpu->gd_npxthread == curthread) { 1441 tf->tf_xflags &= ~PGEX_FPFAULT; 1442 } else { 1443 tf->tf_xflags |= PGEX_FPFAULT; 1444 } 1445 1446 /* 1447 * Run emulated user process context. This call interlocks 1448 * with new mailbox signals. 1449 * 1450 * Set PGEX_U unconditionally, indicating a user frame (the 1451 * bit is normally set only by T_PAGEFLT). 1452 */ 1453 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN, 1454 tf, &curthread->td_savevext); 1455 frame->if_xflags |= PGEX_U; 1456 #if 0 1457 kprintf("GO USER %d trap %ld EVA %08lx RIP %08lx RSP %08lx XFLAGS %02lx/%02lx\n", 1458 r, tf->tf_trapno, tf->tf_addr, tf->tf_rip, tf->tf_rsp, 1459 tf->tf_xflags, frame->if_xflags); 1460 #endif 1461 if (r < 0) { 1462 if (errno != EINTR) 1463 panic("vmspace_ctl failed error %d", errno); 1464 } else { 1465 if (tf->tf_trapno) { 1466 user_trap(tf); 1467 } 1468 } 1469 if (mycpu->gd_reqflags & RQF_AST_MASK) { 1470 tf->tf_trapno = T_ASTFLT; 1471 user_trap(tf); 1472 } 1473 tf->tf_trapno = 0; 1474 } 1475 } 1476 1477 /* 1478 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA 1479 * fault (which is then passed back to the virtual kernel) if an attempt is 1480 * made to use the FP unit. 1481 * 1482 * XXX this is a fairly big hack. 1483 */ 1484 void 1485 set_vkernel_fp(struct trapframe *frame) 1486 { 1487 struct thread *td = curthread; 1488 1489 if (frame->tf_xflags & PGEX_FPFAULT) { 1490 td->td_pcb->pcb_flags |= FP_VIRTFP; 1491 if (mdcpu->gd_npxthread == td) 1492 npxexit(); 1493 } else { 1494 td->td_pcb->pcb_flags &= ~FP_VIRTFP; 1495 } 1496 } 1497 1498 /* 1499 * Called from vkernel_trap() to fixup the vkernel's syscall 1500 * frame for vmspace_ctl() return. 1501 */ 1502 void 1503 cpu_vkernel_trap(struct trapframe *frame, int error) 1504 { 1505 frame->tf_rax = error; 1506 if (error) 1507 frame->tf_rflags |= PSL_C; 1508 else 1509 frame->tf_rflags &= ~PSL_C; 1510 } 1511