1 /*- 2 * Copyright (c) 1992 Terrence R. Lambert. 3 * Copyright (C) 1994, David Greenman 4 * Copyright (c) 1982, 1987, 1990, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * William Jolitz. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * from: @(#)machdep.c 7.4 (Berkeley) 6/3/91 39 * $FreeBSD: src/sys/i386/i386/machdep.c,v 1.385.2.30 2003/05/31 08:48:05 alc Exp $ 40 */ 41 42 #include "opt_compat.h" 43 #include "opt_ddb.h" 44 #include "opt_directio.h" 45 #include "opt_inet.h" 46 #include "opt_msgbuf.h" 47 #include "opt_swap.h" 48 49 #include <sys/param.h> 50 #include <sys/systm.h> 51 #include <sys/sysproto.h> 52 #include <sys/signalvar.h> 53 #include <sys/kernel.h> 54 #include <sys/linker.h> 55 #include <sys/malloc.h> 56 #include <sys/proc.h> 57 #include <sys/buf.h> 58 #include <sys/reboot.h> 59 #include <sys/mbuf.h> 60 #include <sys/msgbuf.h> 61 #include <sys/sysent.h> 62 #include <sys/sysctl.h> 63 #include <sys/vmmeter.h> 64 #include <sys/bus.h> 65 #include <sys/usched.h> 66 #include <sys/reg.h> 67 68 #include <vm/vm.h> 69 #include <vm/vm_param.h> 70 #include <sys/lock.h> 71 #include <vm/vm_kern.h> 72 #include <vm/vm_object.h> 73 #include <vm/vm_page.h> 74 #include <vm/vm_map.h> 75 #include <vm/vm_pager.h> 76 #include <vm/vm_extern.h> 77 78 #include <sys/thread2.h> 79 #include <sys/mplock2.h> 80 81 #include <sys/user.h> 82 #include <sys/exec.h> 83 #include <sys/cons.h> 84 85 #include <ddb/ddb.h> 86 87 #include <machine/cpu.h> 88 #include <machine/clock.h> 89 #include <machine/specialreg.h> 90 #include <machine/md_var.h> 91 #include <machine/pcb_ext.h> /* pcb.h included via sys/user.h */ 92 #include <machine/globaldata.h> /* CPU_prvspace */ 93 #include <machine/smp.h> 94 #ifdef PERFMON 95 #include <machine/perfmon.h> 96 #endif 97 #include <machine/cputypes.h> 98 99 #include <bus/isa/rtc.h> 100 #include <sys/random.h> 101 #include <sys/ptrace.h> 102 #include <machine/sigframe.h> 103 #include <unistd.h> /* umtx_* functions */ 104 #include <pthread.h> /* pthread_yield() */ 105 106 extern void dblfault_handler (void); 107 108 static void set_fpregs_xmm (struct save87 *, struct savexmm *); 109 static void fill_fpregs_xmm (struct savexmm *, struct save87 *); 110 #ifdef DIRECTIO 111 extern void ffs_rawread_setup(void); 112 #endif /* DIRECTIO */ 113 114 int64_t tsc_offsets[MAXCPU]; 115 116 #if defined(SWTCH_OPTIM_STATS) 117 extern int swtch_optim_stats; 118 SYSCTL_INT(_debug, OID_AUTO, swtch_optim_stats, 119 CTLFLAG_RD, &swtch_optim_stats, 0, ""); 120 SYSCTL_INT(_debug, OID_AUTO, tlb_flush_count, 121 CTLFLAG_RD, &tlb_flush_count, 0, ""); 122 #endif 123 124 static int 125 sysctl_hw_physmem(SYSCTL_HANDLER_ARGS) 126 { 127 u_long pmem = ctob(physmem); 128 129 int error = sysctl_handle_long(oidp, &pmem, 0, req); 130 return (error); 131 } 132 133 SYSCTL_PROC(_hw, HW_PHYSMEM, physmem, CTLTYPE_ULONG|CTLFLAG_RD, 134 0, 0, sysctl_hw_physmem, "LU", "Total system memory in bytes (number of pages * page size)"); 135 136 static int 137 sysctl_hw_usermem(SYSCTL_HANDLER_ARGS) 138 { 139 /* JG */ 140 int error = sysctl_handle_int(oidp, 0, 141 ctob((int)Maxmem - vmstats.v_wire_count), req); 142 return (error); 143 } 144 145 SYSCTL_PROC(_hw, HW_USERMEM, usermem, CTLTYPE_INT|CTLFLAG_RD, 146 0, 0, sysctl_hw_usermem, "IU", ""); 147 148 SYSCTL_ULONG(_hw, OID_AUTO, availpages, CTLFLAG_RD, &Maxmem, 0, ""); 149 150 #if 0 151 152 static int 153 sysctl_machdep_msgbuf(SYSCTL_HANDLER_ARGS) 154 { 155 int error; 156 157 /* Unwind the buffer, so that it's linear (possibly starting with 158 * some initial nulls). 159 */ 160 error=sysctl_handle_opaque(oidp,msgbufp->msg_ptr+msgbufp->msg_bufr, 161 msgbufp->msg_size-msgbufp->msg_bufr,req); 162 if(error) return(error); 163 if(msgbufp->msg_bufr>0) { 164 error=sysctl_handle_opaque(oidp,msgbufp->msg_ptr, 165 msgbufp->msg_bufr,req); 166 } 167 return(error); 168 } 169 170 SYSCTL_PROC(_machdep, OID_AUTO, msgbuf, CTLTYPE_STRING|CTLFLAG_RD, 171 0, 0, sysctl_machdep_msgbuf, "A","Contents of kernel message buffer"); 172 173 static int msgbuf_clear; 174 175 static int 176 sysctl_machdep_msgbuf_clear(SYSCTL_HANDLER_ARGS) 177 { 178 int error; 179 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, 180 req); 181 if (!error && req->newptr) { 182 /* Clear the buffer and reset write pointer */ 183 bzero(msgbufp->msg_ptr,msgbufp->msg_size); 184 msgbufp->msg_bufr=msgbufp->msg_bufx=0; 185 msgbuf_clear=0; 186 } 187 return (error); 188 } 189 190 SYSCTL_PROC(_machdep, OID_AUTO, msgbuf_clear, CTLTYPE_INT|CTLFLAG_RW, 191 &msgbuf_clear, 0, sysctl_machdep_msgbuf_clear, "I", 192 "Clear kernel message buffer"); 193 194 #endif 195 196 /* 197 * Send an interrupt to process. 198 * 199 * Stack is set up to allow sigcode stored 200 * at top to call routine, followed by kcall 201 * to sigreturn routine below. After sigreturn 202 * resets the signal mask, the stack, and the 203 * frame pointer, it returns to the user 204 * specified pc, psl. 205 */ 206 void 207 sendsig(sig_t catcher, int sig, sigset_t *mask, u_long code) 208 { 209 struct lwp *lp = curthread->td_lwp; 210 struct proc *p = lp->lwp_proc; 211 struct trapframe *regs; 212 struct sigacts *psp = p->p_sigacts; 213 struct sigframe sf, *sfp; 214 int oonstack; 215 char *sp; 216 217 regs = lp->lwp_md.md_regs; 218 oonstack = (lp->lwp_sigstk.ss_flags & SS_ONSTACK) ? 1 : 0; 219 220 /* Save user context */ 221 bzero(&sf, sizeof(struct sigframe)); 222 sf.sf_uc.uc_sigmask = *mask; 223 sf.sf_uc.uc_stack = lp->lwp_sigstk; 224 sf.sf_uc.uc_mcontext.mc_onstack = oonstack; 225 KKASSERT(__offsetof(struct trapframe, tf_rdi) == 0); 226 bcopy(regs, &sf.sf_uc.uc_mcontext.mc_rdi, sizeof(struct trapframe)); 227 228 /* Make the size of the saved context visible to userland */ 229 sf.sf_uc.uc_mcontext.mc_len = sizeof(sf.sf_uc.uc_mcontext); 230 231 /* Allocate and validate space for the signal handler context. */ 232 if ((lp->lwp_flags & LWP_ALTSTACK) != 0 && !oonstack && 233 SIGISMEMBER(psp->ps_sigonstack, sig)) { 234 sp = (char *)(lp->lwp_sigstk.ss_sp + lp->lwp_sigstk.ss_size - 235 sizeof(struct sigframe)); 236 lp->lwp_sigstk.ss_flags |= SS_ONSTACK; 237 } else { 238 /* We take red zone into account */ 239 sp = (char *)regs->tf_rsp - sizeof(struct sigframe) - 128; 240 } 241 242 /* Align to 16 bytes */ 243 sfp = (struct sigframe *)((intptr_t)sp & ~0xFUL); 244 245 /* Translate the signal is appropriate */ 246 if (p->p_sysent->sv_sigtbl) { 247 if (sig <= p->p_sysent->sv_sigsize) 248 sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)]; 249 } 250 251 /* 252 * Build the argument list for the signal handler. 253 * 254 * Arguments are in registers (%rdi, %rsi, %rdx, %rcx) 255 */ 256 regs->tf_rdi = sig; /* argument 1 */ 257 regs->tf_rdx = (register_t)&sfp->sf_uc; /* argument 3 */ 258 259 if (SIGISMEMBER(psp->ps_siginfo, sig)) { 260 /* 261 * Signal handler installed with SA_SIGINFO. 262 * 263 * action(signo, siginfo, ucontext) 264 */ 265 regs->tf_rsi = (register_t)&sfp->sf_si; /* argument 2 */ 266 regs->tf_rcx = (register_t)regs->tf_err; /* argument 4 */ 267 sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher; 268 269 /* fill siginfo structure */ 270 sf.sf_si.si_signo = sig; 271 sf.sf_si.si_code = code; 272 sf.sf_si.si_addr = (void *)regs->tf_addr; 273 } else { 274 /* 275 * Old FreeBSD-style arguments. 276 * 277 * handler (signo, code, [uc], addr) 278 */ 279 regs->tf_rsi = (register_t)code; /* argument 2 */ 280 regs->tf_rcx = (register_t)regs->tf_addr; /* argument 4 */ 281 sf.sf_ahu.sf_handler = catcher; 282 } 283 284 #if 0 285 /* 286 * If we're a vm86 process, we want to save the segment registers. 287 * We also change eflags to be our emulated eflags, not the actual 288 * eflags. 289 */ 290 if (regs->tf_eflags & PSL_VM) { 291 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs; 292 struct vm86_kernel *vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86; 293 294 sf.sf_uc.uc_mcontext.mc_gs = tf->tf_vm86_gs; 295 sf.sf_uc.uc_mcontext.mc_fs = tf->tf_vm86_fs; 296 sf.sf_uc.uc_mcontext.mc_es = tf->tf_vm86_es; 297 sf.sf_uc.uc_mcontext.mc_ds = tf->tf_vm86_ds; 298 299 if (vm86->vm86_has_vme == 0) 300 sf.sf_uc.uc_mcontext.mc_eflags = 301 (tf->tf_eflags & ~(PSL_VIF | PSL_VIP)) | 302 (vm86->vm86_eflags & (PSL_VIF | PSL_VIP)); 303 304 /* 305 * Clear PSL_NT to inhibit T_TSSFLT faults on return from 306 * syscalls made by the signal handler. This just avoids 307 * wasting time for our lazy fixup of such faults. PSL_NT 308 * does nothing in vm86 mode, but vm86 programs can set it 309 * almost legitimately in probes for old cpu types. 310 */ 311 tf->tf_eflags &= ~(PSL_VM | PSL_NT | PSL_VIF | PSL_VIP); 312 } 313 #endif 314 315 /* 316 * Save the FPU state and reinit the FP unit 317 */ 318 npxpush(&sf.sf_uc.uc_mcontext); 319 320 /* 321 * Copy the sigframe out to the user's stack. 322 */ 323 if (copyout(&sf, sfp, sizeof(struct sigframe)) != 0) { 324 /* 325 * Something is wrong with the stack pointer. 326 * ...Kill the process. 327 */ 328 sigexit(lp, SIGILL); 329 } 330 331 regs->tf_rsp = (register_t)sfp; 332 regs->tf_rip = PS_STRINGS - *(p->p_sysent->sv_szsigcode); 333 334 /* 335 * i386 abi specifies that the direction flag must be cleared 336 * on function entry 337 */ 338 regs->tf_rflags &= ~(PSL_T|PSL_D); 339 340 /* 341 * 64 bit mode has a code and stack selector but 342 * no data or extra selector. %fs and %gs are not 343 * stored in-context. 344 */ 345 regs->tf_cs = _ucodesel; 346 regs->tf_ss = _udatasel; 347 } 348 349 /* 350 * Sanitize the trapframe for a virtual kernel passing control to a custom 351 * VM context. Remove any items that would otherwise create a privilage 352 * issue. 353 * 354 * XXX at the moment we allow userland to set the resume flag. Is this a 355 * bad idea? 356 */ 357 int 358 cpu_sanitize_frame(struct trapframe *frame) 359 { 360 frame->tf_cs = _ucodesel; 361 frame->tf_ss = _udatasel; 362 /* XXX VM (8086) mode not supported? */ 363 frame->tf_rflags &= (PSL_RF | PSL_USERCHANGE | PSL_VM_UNSUPP); 364 frame->tf_rflags |= PSL_RESERVED_DEFAULT | PSL_I; 365 366 return(0); 367 } 368 369 /* 370 * Sanitize the tls so loading the descriptor does not blow up 371 * on us. For x86_64 we don't have to do anything. 372 */ 373 int 374 cpu_sanitize_tls(struct savetls *tls) 375 { 376 return(0); 377 } 378 379 /* 380 * sigreturn(ucontext_t *sigcntxp) 381 * 382 * System call to cleanup state after a signal 383 * has been taken. Reset signal mask and 384 * stack state from context left by sendsig (above). 385 * Return to previous pc and psl as specified by 386 * context left by sendsig. Check carefully to 387 * make sure that the user has not modified the 388 * state to gain improper privileges. 389 */ 390 #define EFL_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 391 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 392 393 int 394 sys_sigreturn(struct sigreturn_args *uap) 395 { 396 struct lwp *lp = curthread->td_lwp; 397 struct trapframe *regs; 398 ucontext_t uc; 399 ucontext_t *ucp; 400 register_t rflags; 401 int cs; 402 int error; 403 404 /* 405 * We have to copy the information into kernel space so userland 406 * can't modify it while we are sniffing it. 407 */ 408 regs = lp->lwp_md.md_regs; 409 error = copyin(uap->sigcntxp, &uc, sizeof(uc)); 410 if (error) 411 return (error); 412 ucp = &uc; 413 rflags = ucp->uc_mcontext.mc_rflags; 414 415 /* VM (8086) mode not supported */ 416 rflags &= ~PSL_VM_UNSUPP; 417 418 #if 0 419 if (eflags & PSL_VM) { 420 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs; 421 struct vm86_kernel *vm86; 422 423 /* 424 * if pcb_ext == 0 or vm86_inited == 0, the user hasn't 425 * set up the vm86 area, and we can't enter vm86 mode. 426 */ 427 if (lp->lwp_thread->td_pcb->pcb_ext == 0) 428 return (EINVAL); 429 vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86; 430 if (vm86->vm86_inited == 0) 431 return (EINVAL); 432 433 /* go back to user mode if both flags are set */ 434 if ((eflags & PSL_VIP) && (eflags & PSL_VIF)) 435 trapsignal(lp->lwp_proc, SIGBUS, 0); 436 437 if (vm86->vm86_has_vme) { 438 eflags = (tf->tf_eflags & ~VME_USERCHANGE) | 439 (eflags & VME_USERCHANGE) | PSL_VM; 440 } else { 441 vm86->vm86_eflags = eflags; /* save VIF, VIP */ 442 eflags = (tf->tf_eflags & ~VM_USERCHANGE) | (eflags & VM_USERCHANGE) | PSL_VM; 443 } 444 bcopy(&ucp.uc_mcontext.mc_gs, tf, sizeof(struct trapframe)); 445 tf->tf_eflags = eflags; 446 tf->tf_vm86_ds = tf->tf_ds; 447 tf->tf_vm86_es = tf->tf_es; 448 tf->tf_vm86_fs = tf->tf_fs; 449 tf->tf_vm86_gs = tf->tf_gs; 450 tf->tf_ds = _udatasel; 451 tf->tf_es = _udatasel; 452 #if 0 453 tf->tf_fs = _udatasel; 454 tf->tf_gs = _udatasel; 455 #endif 456 } else 457 #endif 458 { 459 /* 460 * Don't allow users to change privileged or reserved flags. 461 */ 462 /* 463 * XXX do allow users to change the privileged flag PSL_RF. 464 * The cpu sets PSL_RF in tf_eflags for faults. Debuggers 465 * should sometimes set it there too. tf_eflags is kept in 466 * the signal context during signal handling and there is no 467 * other place to remember it, so the PSL_RF bit may be 468 * corrupted by the signal handler without us knowing. 469 * Corruption of the PSL_RF bit at worst causes one more or 470 * one less debugger trap, so allowing it is fairly harmless. 471 */ 472 if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) { 473 kprintf("sigreturn: rflags = 0x%lx\n", (long)rflags); 474 return(EINVAL); 475 } 476 477 /* 478 * Don't allow users to load a valid privileged %cs. Let the 479 * hardware check for invalid selectors, excess privilege in 480 * other selectors, invalid %eip's and invalid %esp's. 481 */ 482 cs = ucp->uc_mcontext.mc_cs; 483 if (!CS_SECURE(cs)) { 484 kprintf("sigreturn: cs = 0x%x\n", cs); 485 trapsignal(lp, SIGBUS, T_PROTFLT); 486 return(EINVAL); 487 } 488 bcopy(&ucp->uc_mcontext.mc_rdi, regs, sizeof(struct trapframe)); 489 } 490 491 /* 492 * Restore the FPU state from the frame 493 */ 494 npxpop(&ucp->uc_mcontext); 495 496 if (ucp->uc_mcontext.mc_onstack & 1) 497 lp->lwp_sigstk.ss_flags |= SS_ONSTACK; 498 else 499 lp->lwp_sigstk.ss_flags &= ~SS_ONSTACK; 500 501 lp->lwp_sigmask = ucp->uc_sigmask; 502 SIG_CANTMASK(lp->lwp_sigmask); 503 return(EJUSTRETURN); 504 } 505 506 /* 507 * cpu_idle() represents the idle LWKT. You cannot return from this function 508 * (unless you want to blow things up!). Instead we look for runnable threads 509 * and loop or halt as appropriate. Giant is not held on entry to the thread. 510 * 511 * The main loop is entered with a critical section held, we must release 512 * the critical section before doing anything else. lwkt_switch() will 513 * check for pending interrupts due to entering and exiting its own 514 * critical section. 515 * 516 * Note on cpu_idle_hlt: On an SMP system we rely on a scheduler IPI 517 * to wake a HLTed cpu up. 518 */ 519 static int cpu_idle_hlt = 1; 520 static int cpu_idle_hltcnt; 521 static int cpu_idle_spincnt; 522 SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hlt, CTLFLAG_RW, 523 &cpu_idle_hlt, 0, "Idle loop HLT enable"); 524 SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hltcnt, CTLFLAG_RW, 525 &cpu_idle_hltcnt, 0, "Idle loop entry halts"); 526 SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_spincnt, CTLFLAG_RW, 527 &cpu_idle_spincnt, 0, "Idle loop entry spins"); 528 529 void 530 cpu_idle(void) 531 { 532 struct thread *td = curthread; 533 struct mdglobaldata *gd = mdcpu; 534 int reqflags; 535 536 crit_exit(); 537 KKASSERT(td->td_critcount == 0); 538 cpu_enable_intr(); 539 540 for (;;) { 541 /* 542 * See if there are any LWKTs ready to go. 543 */ 544 lwkt_switch(); 545 546 /* 547 * The idle loop halts only if no threads are scheduleable 548 * and no signals have occured. 549 */ 550 if (cpu_idle_hlt && 551 (td->td_gd->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0) { 552 splz(); 553 if ((td->td_gd->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0) { 554 #ifdef DEBUGIDLE 555 struct timeval tv1, tv2; 556 gettimeofday(&tv1, NULL); 557 #endif 558 reqflags = gd->mi.gd_reqflags & 559 ~RQF_IDLECHECK_WK_MASK; 560 KKASSERT(gd->mi.gd_processing_ipiq == 0); 561 umtx_sleep(&gd->mi.gd_reqflags, reqflags, 562 1000000); 563 #ifdef DEBUGIDLE 564 gettimeofday(&tv2, NULL); 565 if (tv2.tv_usec - tv1.tv_usec + 566 (tv2.tv_sec - tv1.tv_sec) * 1000000 567 > 500000) { 568 kprintf("cpu %d idlelock %08x %08x\n", 569 gd->mi.gd_cpuid, 570 gd->mi.gd_reqflags, 571 gd->gd_fpending); 572 } 573 #endif 574 } 575 ++cpu_idle_hltcnt; 576 } else { 577 splz(); 578 __asm __volatile("pause"); 579 ++cpu_idle_spincnt; 580 } 581 } 582 } 583 584 /* 585 * Called by the spinlock code with or without a critical section held 586 * when a spinlock is found to be seriously constested. 587 * 588 * We need to enter a critical section to prevent signals from recursing 589 * into pthreads. 590 */ 591 void 592 cpu_spinlock_contested(void) 593 { 594 cpu_pause(); 595 } 596 597 /* 598 * Clear registers on exec 599 */ 600 void 601 exec_setregs(u_long entry, u_long stack, u_long ps_strings) 602 { 603 struct thread *td = curthread; 604 struct lwp *lp = td->td_lwp; 605 struct pcb *pcb = td->td_pcb; 606 struct trapframe *regs = lp->lwp_md.md_regs; 607 608 /* was i386_user_cleanup() in NetBSD */ 609 user_ldt_free(pcb); 610 611 bzero((char *)regs, sizeof(struct trapframe)); 612 regs->tf_rip = entry; 613 regs->tf_rsp = ((stack - 8) & ~0xFul) + 8; /* align the stack */ 614 regs->tf_rdi = stack; /* argv */ 615 regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T); 616 regs->tf_ss = _udatasel; 617 regs->tf_cs = _ucodesel; 618 regs->tf_rbx = ps_strings; 619 620 /* 621 * Reset the hardware debug registers if they were in use. 622 * They won't have any meaning for the newly exec'd process. 623 */ 624 if (pcb->pcb_flags & PCB_DBREGS) { 625 pcb->pcb_dr0 = 0; 626 pcb->pcb_dr1 = 0; 627 pcb->pcb_dr2 = 0; 628 pcb->pcb_dr3 = 0; 629 pcb->pcb_dr6 = 0; 630 pcb->pcb_dr7 = 0; /* JG set bit 10? */ 631 if (pcb == td->td_pcb) { 632 /* 633 * Clear the debug registers on the running 634 * CPU, otherwise they will end up affecting 635 * the next process we switch to. 636 */ 637 reset_dbregs(); 638 } 639 pcb->pcb_flags &= ~PCB_DBREGS; 640 } 641 642 /* 643 * Initialize the math emulator (if any) for the current process. 644 * Actually, just clear the bit that says that the emulator has 645 * been initialized. Initialization is delayed until the process 646 * traps to the emulator (if it is done at all) mainly because 647 * emulators don't provide an entry point for initialization. 648 */ 649 pcb->pcb_flags &= ~FP_SOFTFP; 650 651 /* 652 * NOTE: do not set CR0_TS here. npxinit() must do it after clearing 653 * gd_npxthread. Otherwise a preemptive interrupt thread 654 * may panic in npxdna(). 655 */ 656 crit_enter(); 657 #if 0 658 load_cr0(rcr0() | CR0_MP); 659 #endif 660 661 /* 662 * NOTE: The MSR values must be correct so we can return to 663 * userland. gd_user_fs/gs must be correct so the switch 664 * code knows what the current MSR values are. 665 */ 666 pcb->pcb_fsbase = 0; /* Values loaded from PCB on switch */ 667 pcb->pcb_gsbase = 0; 668 /* Initialize the npx (if any) for the current process. */ 669 npxinit(); 670 crit_exit(); 671 672 /* 673 * note: linux emulator needs edx to be 0x0 on entry, which is 674 * handled in execve simply by setting the 64 bit syscall 675 * return value to 0. 676 */ 677 } 678 679 void 680 cpu_setregs(void) 681 { 682 #if 0 683 unsigned int cr0; 684 685 cr0 = rcr0(); 686 cr0 |= CR0_NE; /* Done by npxinit() */ 687 cr0 |= CR0_MP | CR0_TS; /* Done at every execve() too. */ 688 cr0 |= CR0_WP | CR0_AM; 689 load_cr0(cr0); 690 load_gs(_udatasel); 691 #endif 692 } 693 694 static int 695 sysctl_machdep_adjkerntz(SYSCTL_HANDLER_ARGS) 696 { 697 int error; 698 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, 699 req); 700 if (!error && req->newptr) 701 resettodr(); 702 return (error); 703 } 704 705 SYSCTL_PROC(_machdep, CPU_ADJKERNTZ, adjkerntz, CTLTYPE_INT|CTLFLAG_RW, 706 &adjkerntz, 0, sysctl_machdep_adjkerntz, "I", ""); 707 708 extern u_long bootdev; /* not a cdev_t - encoding is different */ 709 SYSCTL_ULONG(_machdep, OID_AUTO, guessed_bootdev, 710 CTLFLAG_RD, &bootdev, 0, "Boot device (not in cdev_t format)"); 711 712 /* 713 * Initialize 386 and configure to run kernel 714 */ 715 716 /* 717 * Initialize segments & interrupt table 718 */ 719 720 extern struct user *proc0paddr; 721 722 #if 0 723 724 extern inthand_t 725 IDTVEC(div), IDTVEC(dbg), IDTVEC(nmi), IDTVEC(bpt), IDTVEC(ofl), 726 IDTVEC(bnd), IDTVEC(ill), IDTVEC(dna), IDTVEC(fpusegm), 727 IDTVEC(tss), IDTVEC(missing), IDTVEC(stk), IDTVEC(prot), 728 IDTVEC(page), IDTVEC(mchk), IDTVEC(rsvd), IDTVEC(fpu), IDTVEC(align), 729 IDTVEC(xmm), IDTVEC(dblfault), 730 IDTVEC(fast_syscall), IDTVEC(fast_syscall32); 731 #endif 732 733 #ifdef DEBUG_INTERRUPTS 734 extern inthand_t *Xrsvdary[256]; 735 #endif 736 737 int 738 ptrace_set_pc(struct lwp *lp, unsigned long addr) 739 { 740 lp->lwp_md.md_regs->tf_rip = addr; 741 return (0); 742 } 743 744 int 745 ptrace_single_step(struct lwp *lp) 746 { 747 lp->lwp_md.md_regs->tf_rflags |= PSL_T; 748 return (0); 749 } 750 751 int 752 fill_regs(struct lwp *lp, struct reg *regs) 753 { 754 struct trapframe *tp; 755 756 if ((tp = lp->lwp_md.md_regs) == NULL) 757 return EINVAL; 758 bcopy(&tp->tf_rdi, ®s->r_rdi, sizeof(*regs)); 759 return (0); 760 } 761 762 int 763 set_regs(struct lwp *lp, struct reg *regs) 764 { 765 struct trapframe *tp; 766 767 tp = lp->lwp_md.md_regs; 768 if (!EFL_SECURE(regs->r_rflags, tp->tf_rflags) || 769 !CS_SECURE(regs->r_cs)) 770 return (EINVAL); 771 bcopy(®s->r_rdi, &tp->tf_rdi, sizeof(*regs)); 772 return (0); 773 } 774 775 static void 776 fill_fpregs_xmm(struct savexmm *sv_xmm, struct save87 *sv_87) 777 { 778 struct env87 *penv_87 = &sv_87->sv_env; 779 struct envxmm *penv_xmm = &sv_xmm->sv_env; 780 int i; 781 782 /* FPU control/status */ 783 penv_87->en_cw = penv_xmm->en_cw; 784 penv_87->en_sw = penv_xmm->en_sw; 785 penv_87->en_tw = penv_xmm->en_tw; 786 penv_87->en_fip = penv_xmm->en_fip; 787 penv_87->en_fcs = penv_xmm->en_fcs; 788 penv_87->en_opcode = penv_xmm->en_opcode; 789 penv_87->en_foo = penv_xmm->en_foo; 790 penv_87->en_fos = penv_xmm->en_fos; 791 792 /* FPU registers */ 793 for (i = 0; i < 8; ++i) 794 sv_87->sv_ac[i] = sv_xmm->sv_fp[i].fp_acc; 795 } 796 797 static void 798 set_fpregs_xmm(struct save87 *sv_87, struct savexmm *sv_xmm) 799 { 800 struct env87 *penv_87 = &sv_87->sv_env; 801 struct envxmm *penv_xmm = &sv_xmm->sv_env; 802 int i; 803 804 /* FPU control/status */ 805 penv_xmm->en_cw = penv_87->en_cw; 806 penv_xmm->en_sw = penv_87->en_sw; 807 penv_xmm->en_tw = penv_87->en_tw; 808 penv_xmm->en_fip = penv_87->en_fip; 809 penv_xmm->en_fcs = penv_87->en_fcs; 810 penv_xmm->en_opcode = penv_87->en_opcode; 811 penv_xmm->en_foo = penv_87->en_foo; 812 penv_xmm->en_fos = penv_87->en_fos; 813 814 /* FPU registers */ 815 for (i = 0; i < 8; ++i) 816 sv_xmm->sv_fp[i].fp_acc = sv_87->sv_ac[i]; 817 } 818 819 int 820 fill_fpregs(struct lwp *lp, struct fpreg *fpregs) 821 { 822 if (lp->lwp_thread == NULL || lp->lwp_thread->td_pcb == NULL) 823 return EINVAL; 824 if (cpu_fxsr) { 825 fill_fpregs_xmm(&lp->lwp_thread->td_pcb->pcb_save.sv_xmm, 826 (struct save87 *)fpregs); 827 return (0); 828 } 829 bcopy(&lp->lwp_thread->td_pcb->pcb_save.sv_87, fpregs, sizeof *fpregs); 830 return (0); 831 } 832 833 int 834 set_fpregs(struct lwp *lp, struct fpreg *fpregs) 835 { 836 if (cpu_fxsr) { 837 set_fpregs_xmm((struct save87 *)fpregs, 838 &lp->lwp_thread->td_pcb->pcb_save.sv_xmm); 839 return (0); 840 } 841 bcopy(fpregs, &lp->lwp_thread->td_pcb->pcb_save.sv_87, sizeof *fpregs); 842 return (0); 843 } 844 845 int 846 fill_dbregs(struct lwp *lp, struct dbreg *dbregs) 847 { 848 return (ENOSYS); 849 } 850 851 int 852 set_dbregs(struct lwp *lp, struct dbreg *dbregs) 853 { 854 return (ENOSYS); 855 } 856 857 #if 0 858 /* 859 * Return > 0 if a hardware breakpoint has been hit, and the 860 * breakpoint was in user space. Return 0, otherwise. 861 */ 862 int 863 user_dbreg_trap(void) 864 { 865 u_int32_t dr7, dr6; /* debug registers dr6 and dr7 */ 866 u_int32_t bp; /* breakpoint bits extracted from dr6 */ 867 int nbp; /* number of breakpoints that triggered */ 868 caddr_t addr[4]; /* breakpoint addresses */ 869 int i; 870 871 dr7 = rdr7(); 872 if ((dr7 & 0x000000ff) == 0) { 873 /* 874 * all GE and LE bits in the dr7 register are zero, 875 * thus the trap couldn't have been caused by the 876 * hardware debug registers 877 */ 878 return 0; 879 } 880 881 nbp = 0; 882 dr6 = rdr6(); 883 bp = dr6 & 0x0000000f; 884 885 if (!bp) { 886 /* 887 * None of the breakpoint bits are set meaning this 888 * trap was not caused by any of the debug registers 889 */ 890 return 0; 891 } 892 893 /* 894 * at least one of the breakpoints were hit, check to see 895 * which ones and if any of them are user space addresses 896 */ 897 898 if (bp & 0x01) { 899 addr[nbp++] = (caddr_t)rdr0(); 900 } 901 if (bp & 0x02) { 902 addr[nbp++] = (caddr_t)rdr1(); 903 } 904 if (bp & 0x04) { 905 addr[nbp++] = (caddr_t)rdr2(); 906 } 907 if (bp & 0x08) { 908 addr[nbp++] = (caddr_t)rdr3(); 909 } 910 911 for (i=0; i<nbp; i++) { 912 if (addr[i] < 913 (caddr_t)VM_MAX_USER_ADDRESS) { 914 /* 915 * addr[i] is in user space 916 */ 917 return nbp; 918 } 919 } 920 921 /* 922 * None of the breakpoints are in user space. 923 */ 924 return 0; 925 } 926 927 #endif 928 929 void 930 identcpu(void) 931 { 932 int regs[4]; 933 934 do_cpuid(1, regs); 935 cpu_feature = regs[3]; 936 } 937 938 939 #ifndef DDB 940 void 941 Debugger(const char *msg) 942 { 943 kprintf("Debugger(\"%s\") called.\n", msg); 944 } 945 #endif /* no DDB */ 946