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_atalk.h" 43 #include "opt_compat.h" 44 #include "opt_ddb.h" 45 #include "opt_directio.h" 46 #include "opt_inet.h" 47 #include "opt_ipx.h" 48 #include "opt_msgbuf.h" 49 #include "opt_swap.h" 50 51 #include <sys/param.h> 52 #include <sys/systm.h> 53 #include <sys/sysproto.h> 54 #include <sys/signalvar.h> 55 #include <sys/kernel.h> 56 #include <sys/linker.h> 57 #include <sys/malloc.h> 58 #include <sys/proc.h> 59 #include <sys/buf.h> 60 #include <sys/reboot.h> 61 #include <sys/mbuf.h> 62 #include <sys/msgbuf.h> 63 #include <sys/sysent.h> 64 #include <sys/sysctl.h> 65 #include <sys/vmmeter.h> 66 #include <sys/bus.h> 67 #include <sys/upcall.h> 68 #include <sys/usched.h> 69 #include <sys/reg.h> 70 71 #include <vm/vm.h> 72 #include <vm/vm_param.h> 73 #include <sys/lock.h> 74 #include <vm/vm_kern.h> 75 #include <vm/vm_object.h> 76 #include <vm/vm_page.h> 77 #include <vm/vm_map.h> 78 #include <vm/vm_pager.h> 79 #include <vm/vm_extern.h> 80 81 #include <sys/thread2.h> 82 #include <sys/mplock2.h> 83 84 #include <sys/user.h> 85 #include <sys/exec.h> 86 #include <sys/cons.h> 87 88 #include <ddb/ddb.h> 89 90 #include <machine/cpu.h> 91 #include <machine/clock.h> 92 #include <machine/specialreg.h> 93 #include <machine/md_var.h> 94 #include <machine/pcb_ext.h> /* pcb.h included via sys/user.h */ 95 #include <machine/globaldata.h> /* CPU_prvspace */ 96 #include <machine/smp.h> 97 #ifdef PERFMON 98 #include <machine/perfmon.h> 99 #endif 100 #include <machine/cputypes.h> 101 102 #include <bus/isa/rtc.h> 103 #include <sys/random.h> 104 #include <sys/ptrace.h> 105 #include <machine/sigframe.h> 106 #include <unistd.h> /* umtx_* functions */ 107 #include <pthread.h> /* pthread_yield() */ 108 109 extern void dblfault_handler (void); 110 111 #ifndef CPU_DISABLE_SSE 112 static void set_fpregs_xmm (struct save87 *, struct savexmm *); 113 static void fill_fpregs_xmm (struct savexmm *, struct save87 *); 114 #endif /* CPU_DISABLE_SSE */ 115 #ifdef DIRECTIO 116 extern void ffs_rawread_setup(void); 117 #endif /* DIRECTIO */ 118 119 #ifdef SMP 120 int64_t tsc_offsets[MAXCPU]; 121 #else 122 int64_t tsc_offsets[1]; 123 #endif 124 125 #if defined(SWTCH_OPTIM_STATS) 126 extern int swtch_optim_stats; 127 SYSCTL_INT(_debug, OID_AUTO, swtch_optim_stats, 128 CTLFLAG_RD, &swtch_optim_stats, 0, ""); 129 SYSCTL_INT(_debug, OID_AUTO, tlb_flush_count, 130 CTLFLAG_RD, &tlb_flush_count, 0, ""); 131 #endif 132 133 static int 134 sysctl_hw_physmem(SYSCTL_HANDLER_ARGS) 135 { 136 u_long pmem = ctob(physmem); 137 138 int error = sysctl_handle_long(oidp, &pmem, 0, req); 139 return (error); 140 } 141 142 SYSCTL_PROC(_hw, HW_PHYSMEM, physmem, CTLTYPE_ULONG|CTLFLAG_RD, 143 0, 0, sysctl_hw_physmem, "LU", "Total system memory in bytes (number of pages * page size)"); 144 145 static int 146 sysctl_hw_usermem(SYSCTL_HANDLER_ARGS) 147 { 148 /* JG */ 149 int error = sysctl_handle_int(oidp, 0, 150 ctob((int)Maxmem - vmstats.v_wire_count), req); 151 return (error); 152 } 153 154 SYSCTL_PROC(_hw, HW_USERMEM, usermem, CTLTYPE_INT|CTLFLAG_RD, 155 0, 0, sysctl_hw_usermem, "IU", ""); 156 157 SYSCTL_ULONG(_hw, OID_AUTO, availpages, CTLFLAG_RD, &Maxmem, 0, ""); 158 159 #if 0 160 161 static int 162 sysctl_machdep_msgbuf(SYSCTL_HANDLER_ARGS) 163 { 164 int error; 165 166 /* Unwind the buffer, so that it's linear (possibly starting with 167 * some initial nulls). 168 */ 169 error=sysctl_handle_opaque(oidp,msgbufp->msg_ptr+msgbufp->msg_bufr, 170 msgbufp->msg_size-msgbufp->msg_bufr,req); 171 if(error) return(error); 172 if(msgbufp->msg_bufr>0) { 173 error=sysctl_handle_opaque(oidp,msgbufp->msg_ptr, 174 msgbufp->msg_bufr,req); 175 } 176 return(error); 177 } 178 179 SYSCTL_PROC(_machdep, OID_AUTO, msgbuf, CTLTYPE_STRING|CTLFLAG_RD, 180 0, 0, sysctl_machdep_msgbuf, "A","Contents of kernel message buffer"); 181 182 static int msgbuf_clear; 183 184 static int 185 sysctl_machdep_msgbuf_clear(SYSCTL_HANDLER_ARGS) 186 { 187 int error; 188 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, 189 req); 190 if (!error && req->newptr) { 191 /* Clear the buffer and reset write pointer */ 192 bzero(msgbufp->msg_ptr,msgbufp->msg_size); 193 msgbufp->msg_bufr=msgbufp->msg_bufx=0; 194 msgbuf_clear=0; 195 } 196 return (error); 197 } 198 199 SYSCTL_PROC(_machdep, OID_AUTO, msgbuf_clear, CTLTYPE_INT|CTLFLAG_RW, 200 &msgbuf_clear, 0, sysctl_machdep_msgbuf_clear, "I", 201 "Clear kernel message buffer"); 202 203 #endif 204 205 /* 206 * Send an interrupt to process. 207 * 208 * Stack is set up to allow sigcode stored 209 * at top to call routine, followed by kcall 210 * to sigreturn routine below. After sigreturn 211 * resets the signal mask, the stack, and the 212 * frame pointer, it returns to the user 213 * specified pc, psl. 214 */ 215 void 216 sendsig(sig_t catcher, int sig, sigset_t *mask, u_long code) 217 { 218 struct lwp *lp = curthread->td_lwp; 219 struct proc *p = lp->lwp_proc; 220 struct trapframe *regs; 221 struct sigacts *psp = p->p_sigacts; 222 struct sigframe sf, *sfp; 223 int oonstack; 224 char *sp; 225 226 regs = lp->lwp_md.md_regs; 227 oonstack = (lp->lwp_sigstk.ss_flags & SS_ONSTACK) ? 1 : 0; 228 229 /* Save user context */ 230 bzero(&sf, sizeof(struct sigframe)); 231 sf.sf_uc.uc_sigmask = *mask; 232 sf.sf_uc.uc_stack = lp->lwp_sigstk; 233 sf.sf_uc.uc_mcontext.mc_onstack = oonstack; 234 KKASSERT(__offsetof(struct trapframe, tf_rdi) == 0); 235 bcopy(regs, &sf.sf_uc.uc_mcontext.mc_rdi, sizeof(struct trapframe)); 236 237 /* Make the size of the saved context visible to userland */ 238 sf.sf_uc.uc_mcontext.mc_len = sizeof(sf.sf_uc.uc_mcontext); 239 240 /* Save mailbox pending state for syscall interlock semantics */ 241 if (p->p_flag & P_MAILBOX) 242 sf.sf_uc.uc_mcontext.mc_xflags |= PGEX_MAILBOX; 243 244 /* Allocate and validate space for the signal handler context. */ 245 if ((lp->lwp_flag & LWP_ALTSTACK) != 0 && !oonstack && 246 SIGISMEMBER(psp->ps_sigonstack, sig)) { 247 sp = (char *)(lp->lwp_sigstk.ss_sp + lp->lwp_sigstk.ss_size - 248 sizeof(struct sigframe)); 249 lp->lwp_sigstk.ss_flags |= SS_ONSTACK; 250 } else { 251 /* We take red zone into account */ 252 sp = (char *)regs->tf_rsp - sizeof(struct sigframe) - 128; 253 } 254 255 /* Align to 16 bytes */ 256 sfp = (struct sigframe *)((intptr_t)sp & ~0xFUL); 257 258 /* Translate the signal is appropriate */ 259 if (p->p_sysent->sv_sigtbl) { 260 if (sig <= p->p_sysent->sv_sigsize) 261 sig = p->p_sysent->sv_sigtbl[_SIG_IDX(sig)]; 262 } 263 264 /* 265 * Build the argument list for the signal handler. 266 * 267 * Arguments are in registers (%rdi, %rsi, %rdx, %rcx) 268 */ 269 regs->tf_rdi = sig; /* argument 1 */ 270 regs->tf_rdx = (register_t)&sfp->sf_uc; /* argument 3 */ 271 272 if (SIGISMEMBER(psp->ps_siginfo, sig)) { 273 /* 274 * Signal handler installed with SA_SIGINFO. 275 * 276 * action(signo, siginfo, ucontext) 277 */ 278 regs->tf_rsi = (register_t)&sfp->sf_si; /* argument 2 */ 279 regs->tf_rcx = (register_t)regs->tf_err; /* argument 4 */ 280 sf.sf_ahu.sf_action = (__siginfohandler_t *)catcher; 281 282 /* fill siginfo structure */ 283 sf.sf_si.si_signo = sig; 284 sf.sf_si.si_code = code; 285 sf.sf_si.si_addr = (void *)regs->tf_err; 286 } else { 287 /* 288 * Old FreeBSD-style arguments. 289 * 290 * handler (signo, code, [uc], addr) 291 */ 292 regs->tf_rsi = (register_t)code; /* argument 2 */ 293 regs->tf_rcx = (register_t)regs->tf_err; /* argument 4 */ 294 sf.sf_ahu.sf_handler = catcher; 295 } 296 297 #if 0 298 /* 299 * If we're a vm86 process, we want to save the segment registers. 300 * We also change eflags to be our emulated eflags, not the actual 301 * eflags. 302 */ 303 if (regs->tf_eflags & PSL_VM) { 304 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs; 305 struct vm86_kernel *vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86; 306 307 sf.sf_uc.uc_mcontext.mc_gs = tf->tf_vm86_gs; 308 sf.sf_uc.uc_mcontext.mc_fs = tf->tf_vm86_fs; 309 sf.sf_uc.uc_mcontext.mc_es = tf->tf_vm86_es; 310 sf.sf_uc.uc_mcontext.mc_ds = tf->tf_vm86_ds; 311 312 if (vm86->vm86_has_vme == 0) 313 sf.sf_uc.uc_mcontext.mc_eflags = 314 (tf->tf_eflags & ~(PSL_VIF | PSL_VIP)) | 315 (vm86->vm86_eflags & (PSL_VIF | PSL_VIP)); 316 317 /* 318 * Clear PSL_NT to inhibit T_TSSFLT faults on return from 319 * syscalls made by the signal handler. This just avoids 320 * wasting time for our lazy fixup of such faults. PSL_NT 321 * does nothing in vm86 mode, but vm86 programs can set it 322 * almost legitimately in probes for old cpu types. 323 */ 324 tf->tf_eflags &= ~(PSL_VM | PSL_NT | PSL_VIF | PSL_VIP); 325 } 326 #endif 327 328 /* 329 * Save the FPU state and reinit the FP unit 330 */ 331 npxpush(&sf.sf_uc.uc_mcontext); 332 333 /* 334 * Copy the sigframe out to the user's stack. 335 */ 336 if (copyout(&sf, sfp, sizeof(struct sigframe)) != 0) { 337 /* 338 * Something is wrong with the stack pointer. 339 * ...Kill the process. 340 */ 341 sigexit(lp, SIGILL); 342 } 343 344 regs->tf_rsp = (register_t)sfp; 345 regs->tf_rip = PS_STRINGS - *(p->p_sysent->sv_szsigcode); 346 347 /* 348 * i386 abi specifies that the direction flag must be cleared 349 * on function entry 350 */ 351 regs->tf_rflags &= ~(PSL_T|PSL_D); 352 353 /* 354 * 64 bit mode has a code and stack selector but 355 * no data or extra selector. %fs and %gs are not 356 * stored in-context. 357 */ 358 regs->tf_cs = _ucodesel; 359 regs->tf_ss = _udatasel; 360 } 361 362 /* 363 * Sanitize the trapframe for a virtual kernel passing control to a custom 364 * VM context. Remove any items that would otherwise create a privilage 365 * issue. 366 * 367 * XXX at the moment we allow userland to set the resume flag. Is this a 368 * bad idea? 369 */ 370 int 371 cpu_sanitize_frame(struct trapframe *frame) 372 { 373 frame->tf_cs = _ucodesel; 374 frame->tf_ss = _udatasel; 375 /* XXX VM (8086) mode not supported? */ 376 frame->tf_rflags &= (PSL_RF | PSL_USERCHANGE | PSL_VM_UNSUPP); 377 frame->tf_rflags |= PSL_RESERVED_DEFAULT | PSL_I; 378 379 return(0); 380 } 381 382 /* 383 * Sanitize the tls so loading the descriptor does not blow up 384 * on us. For x86_64 we don't have to do anything. 385 */ 386 int 387 cpu_sanitize_tls(struct savetls *tls) 388 { 389 return(0); 390 } 391 392 /* 393 * sigreturn(ucontext_t *sigcntxp) 394 * 395 * System call to cleanup state after a signal 396 * has been taken. Reset signal mask and 397 * stack state from context left by sendsig (above). 398 * Return to previous pc and psl as specified by 399 * context left by sendsig. Check carefully to 400 * make sure that the user has not modified the 401 * state to gain improper privileges. 402 */ 403 #define EFL_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) 404 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) 405 406 int 407 sys_sigreturn(struct sigreturn_args *uap) 408 { 409 struct lwp *lp = curthread->td_lwp; 410 struct proc *p = lp->lwp_proc; 411 struct trapframe *regs; 412 ucontext_t uc; 413 ucontext_t *ucp; 414 register_t rflags; 415 int cs; 416 int error; 417 418 /* 419 * We have to copy the information into kernel space so userland 420 * can't modify it while we are sniffing it. 421 */ 422 regs = lp->lwp_md.md_regs; 423 error = copyin(uap->sigcntxp, &uc, sizeof(uc)); 424 if (error) 425 return (error); 426 ucp = &uc; 427 rflags = ucp->uc_mcontext.mc_rflags; 428 429 /* VM (8086) mode not supported */ 430 rflags &= ~PSL_VM_UNSUPP; 431 432 #if 0 433 if (eflags & PSL_VM) { 434 struct trapframe_vm86 *tf = (struct trapframe_vm86 *)regs; 435 struct vm86_kernel *vm86; 436 437 /* 438 * if pcb_ext == 0 or vm86_inited == 0, the user hasn't 439 * set up the vm86 area, and we can't enter vm86 mode. 440 */ 441 if (lp->lwp_thread->td_pcb->pcb_ext == 0) 442 return (EINVAL); 443 vm86 = &lp->lwp_thread->td_pcb->pcb_ext->ext_vm86; 444 if (vm86->vm86_inited == 0) 445 return (EINVAL); 446 447 /* go back to user mode if both flags are set */ 448 if ((eflags & PSL_VIP) && (eflags & PSL_VIF)) 449 trapsignal(lp->lwp_proc, SIGBUS, 0); 450 451 if (vm86->vm86_has_vme) { 452 eflags = (tf->tf_eflags & ~VME_USERCHANGE) | 453 (eflags & VME_USERCHANGE) | PSL_VM; 454 } else { 455 vm86->vm86_eflags = eflags; /* save VIF, VIP */ 456 eflags = (tf->tf_eflags & ~VM_USERCHANGE) | (eflags & VM_USERCHANGE) | PSL_VM; 457 } 458 bcopy(&ucp.uc_mcontext.mc_gs, tf, sizeof(struct trapframe)); 459 tf->tf_eflags = eflags; 460 tf->tf_vm86_ds = tf->tf_ds; 461 tf->tf_vm86_es = tf->tf_es; 462 tf->tf_vm86_fs = tf->tf_fs; 463 tf->tf_vm86_gs = tf->tf_gs; 464 tf->tf_ds = _udatasel; 465 tf->tf_es = _udatasel; 466 #if 0 467 tf->tf_fs = _udatasel; 468 tf->tf_gs = _udatasel; 469 #endif 470 } else 471 #endif 472 { 473 /* 474 * Don't allow users to change privileged or reserved flags. 475 */ 476 /* 477 * XXX do allow users to change the privileged flag PSL_RF. 478 * The cpu sets PSL_RF in tf_eflags for faults. Debuggers 479 * should sometimes set it there too. tf_eflags is kept in 480 * the signal context during signal handling and there is no 481 * other place to remember it, so the PSL_RF bit may be 482 * corrupted by the signal handler without us knowing. 483 * Corruption of the PSL_RF bit at worst causes one more or 484 * one less debugger trap, so allowing it is fairly harmless. 485 */ 486 if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) { 487 kprintf("sigreturn: rflags = 0x%lx\n", (long)rflags); 488 return(EINVAL); 489 } 490 491 /* 492 * Don't allow users to load a valid privileged %cs. Let the 493 * hardware check for invalid selectors, excess privilege in 494 * other selectors, invalid %eip's and invalid %esp's. 495 */ 496 cs = ucp->uc_mcontext.mc_cs; 497 if (!CS_SECURE(cs)) { 498 kprintf("sigreturn: cs = 0x%x\n", cs); 499 trapsignal(lp, SIGBUS, T_PROTFLT); 500 return(EINVAL); 501 } 502 bcopy(&ucp->uc_mcontext.mc_rdi, regs, sizeof(struct trapframe)); 503 } 504 505 /* 506 * Restore the FPU state from the frame 507 */ 508 npxpop(&ucp->uc_mcontext); 509 510 /* 511 * Merge saved signal mailbox pending flag to maintain interlock 512 * semantics against system calls. 513 */ 514 if (ucp->uc_mcontext.mc_xflags & PGEX_MAILBOX) 515 p->p_flag |= P_MAILBOX; 516 517 if (ucp->uc_mcontext.mc_onstack & 1) 518 lp->lwp_sigstk.ss_flags |= SS_ONSTACK; 519 else 520 lp->lwp_sigstk.ss_flags &= ~SS_ONSTACK; 521 522 lp->lwp_sigmask = ucp->uc_sigmask; 523 SIG_CANTMASK(lp->lwp_sigmask); 524 return(EJUSTRETURN); 525 } 526 527 /* 528 * Stack frame on entry to function. %rax will contain the function vector, 529 * %rcx will contain the function data. flags, rcx, and rax will have 530 * already been pushed on the stack. 531 */ 532 struct upc_frame { 533 register_t rax; 534 register_t rcx; 535 register_t rdx; 536 register_t flags; 537 register_t oldip; 538 }; 539 540 void 541 sendupcall(struct vmupcall *vu, int morepending) 542 { 543 struct lwp *lp = curthread->td_lwp; 544 struct trapframe *regs; 545 struct upcall upcall; 546 struct upc_frame upc_frame; 547 int crit_count = 0; 548 549 /* 550 * If we are a virtual kernel running an emulated user process 551 * context, switch back to the virtual kernel context before 552 * trying to post the signal. 553 */ 554 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) { 555 lp->lwp_md.md_regs->tf_trapno = 0; 556 vkernel_trap(lp, lp->lwp_md.md_regs); 557 } 558 559 /* 560 * Get the upcall data structure 561 */ 562 if (copyin(lp->lwp_upcall, &upcall, sizeof(upcall)) || 563 copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int)) 564 ) { 565 vu->vu_pending = 0; 566 kprintf("bad upcall address\n"); 567 return; 568 } 569 570 /* 571 * If the data structure is already marked pending or has a critical 572 * section count, mark the data structure as pending and return 573 * without doing an upcall. vu_pending is left set. 574 */ 575 if (upcall.upc_pending || crit_count >= vu->vu_pending) { 576 if (upcall.upc_pending < vu->vu_pending) { 577 upcall.upc_pending = vu->vu_pending; 578 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending, 579 sizeof(upcall.upc_pending)); 580 } 581 return; 582 } 583 584 /* 585 * We can run this upcall now, clear vu_pending. 586 * 587 * Bump our critical section count and set or clear the 588 * user pending flag depending on whether more upcalls are 589 * pending. The user will be responsible for calling 590 * upc_dispatch(-1) to process remaining upcalls. 591 */ 592 vu->vu_pending = 0; 593 upcall.upc_pending = morepending; 594 ++crit_count; 595 copyout(&upcall.upc_pending, &lp->lwp_upcall->upc_pending, 596 sizeof(upcall.upc_pending)); 597 copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff, 598 sizeof(int)); 599 600 /* 601 * Construct a stack frame and issue the upcall 602 */ 603 regs = lp->lwp_md.md_regs; 604 upc_frame.rax = regs->tf_rax; 605 upc_frame.rcx = regs->tf_rcx; 606 upc_frame.rdx = regs->tf_rdx; 607 upc_frame.flags = regs->tf_rflags; 608 upc_frame.oldip = regs->tf_rip; 609 if (copyout(&upc_frame, (void *)(regs->tf_rsp - sizeof(upc_frame)), 610 sizeof(upc_frame)) != 0) { 611 kprintf("bad stack on upcall\n"); 612 } else { 613 regs->tf_rax = (register_t)vu->vu_func; 614 regs->tf_rcx = (register_t)vu->vu_data; 615 regs->tf_rdx = (register_t)lp->lwp_upcall; 616 regs->tf_rip = (register_t)vu->vu_ctx; 617 regs->tf_rsp -= sizeof(upc_frame); 618 } 619 } 620 621 /* 622 * fetchupcall occurs in the context of a system call, which means that 623 * we have to return EJUSTRETURN in order to prevent eax and edx from 624 * being overwritten by the syscall return value. 625 * 626 * if vu is not NULL we return the new context in %edx, the new data in %ecx, 627 * and the function pointer in %eax. 628 */ 629 int 630 fetchupcall(struct vmupcall *vu, int morepending, void *rsp) 631 { 632 struct upc_frame upc_frame; 633 struct lwp *lp = curthread->td_lwp; 634 struct trapframe *regs; 635 int error; 636 struct upcall upcall; 637 int crit_count; 638 639 regs = lp->lwp_md.md_regs; 640 641 error = copyout(&morepending, &lp->lwp_upcall->upc_pending, sizeof(int)); 642 if (error == 0) { 643 if (vu) { 644 /* 645 * This jumps us to the next ready context. 646 */ 647 vu->vu_pending = 0; 648 error = copyin(lp->lwp_upcall, &upcall, sizeof(upcall)); 649 crit_count = 0; 650 if (error == 0) 651 error = copyin((char *)upcall.upc_uthread + upcall.upc_critoff, &crit_count, sizeof(int)); 652 ++crit_count; 653 if (error == 0) 654 error = copyout(&crit_count, (char *)upcall.upc_uthread + upcall.upc_critoff, sizeof(int)); 655 regs->tf_rax = (register_t)vu->vu_func; 656 regs->tf_rcx = (register_t)vu->vu_data; 657 regs->tf_rdx = (register_t)lp->lwp_upcall; 658 regs->tf_rip = (register_t)vu->vu_ctx; 659 regs->tf_rsp = (register_t)rsp; 660 } else { 661 /* 662 * This returns us to the originally interrupted code. 663 */ 664 error = copyin(rsp, &upc_frame, sizeof(upc_frame)); 665 regs->tf_rax = upc_frame.rax; 666 regs->tf_rcx = upc_frame.rcx; 667 regs->tf_rdx = upc_frame.rdx; 668 regs->tf_rflags = (regs->tf_rflags & ~PSL_USERCHANGE) | 669 (upc_frame.flags & PSL_USERCHANGE); 670 regs->tf_rip = upc_frame.oldip; 671 regs->tf_rsp = (register_t)((char *)rsp + sizeof(upc_frame)); 672 } 673 } 674 if (error == 0) 675 error = EJUSTRETURN; 676 return(error); 677 } 678 679 /* 680 * cpu_idle() represents the idle LWKT. You cannot return from this function 681 * (unless you want to blow things up!). Instead we look for runnable threads 682 * and loop or halt as appropriate. Giant is not held on entry to the thread. 683 * 684 * The main loop is entered with a critical section held, we must release 685 * the critical section before doing anything else. lwkt_switch() will 686 * check for pending interrupts due to entering and exiting its own 687 * critical section. 688 * 689 * Note on cpu_idle_hlt: On an SMP system we rely on a scheduler IPI 690 * to wake a HLTed cpu up. However, there are cases where the idlethread 691 * will be entered with the possibility that no IPI will occur and in such 692 * cases lwkt_switch() sets RQF_WAKEUP and we nominally check 693 * RQF_IDLECHECK_WK_MASK. 694 */ 695 static int cpu_idle_hlt = 1; 696 static int cpu_idle_hltcnt; 697 static int cpu_idle_spincnt; 698 SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hlt, CTLFLAG_RW, 699 &cpu_idle_hlt, 0, "Idle loop HLT enable"); 700 SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_hltcnt, CTLFLAG_RW, 701 &cpu_idle_hltcnt, 0, "Idle loop entry halts"); 702 SYSCTL_INT(_machdep, OID_AUTO, cpu_idle_spincnt, CTLFLAG_RW, 703 &cpu_idle_spincnt, 0, "Idle loop entry spins"); 704 705 void 706 cpu_idle(void) 707 { 708 struct thread *td = curthread; 709 struct mdglobaldata *gd = mdcpu; 710 int reqflags; 711 712 crit_exit(); 713 KKASSERT(td->td_critcount == 0); 714 cpu_enable_intr(); 715 for (;;) { 716 /* 717 * See if there are any LWKTs ready to go. 718 */ 719 lwkt_switch(); 720 721 /* 722 * The idle loop halts only if no threads are scheduleable 723 * and no signals have occured. 724 */ 725 if (cpu_idle_hlt && 726 (td->td_gd->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0) { 727 splz(); 728 if ((td->td_gd->gd_reqflags & RQF_IDLECHECK_WK_MASK) == 0) { 729 #ifdef DEBUGIDLE 730 struct timeval tv1, tv2; 731 gettimeofday(&tv1, NULL); 732 #endif 733 reqflags = gd->mi.gd_reqflags & 734 ~RQF_IDLECHECK_WK_MASK; 735 umtx_sleep(&gd->mi.gd_reqflags, reqflags, 736 1000000); 737 #ifdef DEBUGIDLE 738 gettimeofday(&tv2, NULL); 739 if (tv2.tv_usec - tv1.tv_usec + 740 (tv2.tv_sec - tv1.tv_sec) * 1000000 741 > 500000) { 742 kprintf("cpu %d idlelock %08x %08x\n", 743 gd->mi.gd_cpuid, 744 gd->mi.gd_reqflags, 745 gd->gd_fpending); 746 } 747 #endif 748 } 749 ++cpu_idle_hltcnt; 750 } else { 751 splz(); 752 #ifdef SMP 753 __asm __volatile("pause"); 754 #endif 755 ++cpu_idle_spincnt; 756 } 757 } 758 } 759 760 #ifdef SMP 761 762 /* 763 * Called by the spinlock code with or without a critical section held 764 * when a spinlock is found to be seriously constested. 765 * 766 * We need to enter a critical section to prevent signals from recursing 767 * into pthreads. 768 */ 769 void 770 cpu_spinlock_contested(void) 771 { 772 cpu_pause(); 773 } 774 775 #endif 776 777 /* 778 * Clear registers on exec 779 */ 780 void 781 exec_setregs(u_long entry, u_long stack, u_long ps_strings) 782 { 783 struct thread *td = curthread; 784 struct lwp *lp = td->td_lwp; 785 struct pcb *pcb = td->td_pcb; 786 struct trapframe *regs = lp->lwp_md.md_regs; 787 788 /* was i386_user_cleanup() in NetBSD */ 789 user_ldt_free(pcb); 790 791 bzero((char *)regs, sizeof(struct trapframe)); 792 regs->tf_rip = entry; 793 regs->tf_rsp = ((stack - 8) & ~0xFul) + 8; /* align the stack */ 794 regs->tf_rdi = stack; /* argv */ 795 regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T); 796 regs->tf_ss = _udatasel; 797 regs->tf_cs = _ucodesel; 798 regs->tf_rbx = ps_strings; 799 800 /* 801 * Reset the hardware debug registers if they were in use. 802 * They won't have any meaning for the newly exec'd process. 803 */ 804 if (pcb->pcb_flags & PCB_DBREGS) { 805 pcb->pcb_dr0 = 0; 806 pcb->pcb_dr1 = 0; 807 pcb->pcb_dr2 = 0; 808 pcb->pcb_dr3 = 0; 809 pcb->pcb_dr6 = 0; 810 pcb->pcb_dr7 = 0; /* JG set bit 10? */ 811 if (pcb == td->td_pcb) { 812 /* 813 * Clear the debug registers on the running 814 * CPU, otherwise they will end up affecting 815 * the next process we switch to. 816 */ 817 reset_dbregs(); 818 } 819 pcb->pcb_flags &= ~PCB_DBREGS; 820 } 821 822 /* 823 * Initialize the math emulator (if any) for the current process. 824 * Actually, just clear the bit that says that the emulator has 825 * been initialized. Initialization is delayed until the process 826 * traps to the emulator (if it is done at all) mainly because 827 * emulators don't provide an entry point for initialization. 828 */ 829 pcb->pcb_flags &= ~FP_SOFTFP; 830 831 /* 832 * NOTE: do not set CR0_TS here. npxinit() must do it after clearing 833 * gd_npxthread. Otherwise a preemptive interrupt thread 834 * may panic in npxdna(). 835 */ 836 crit_enter(); 837 #if 0 838 load_cr0(rcr0() | CR0_MP); 839 #endif 840 841 /* 842 * NOTE: The MSR values must be correct so we can return to 843 * userland. gd_user_fs/gs must be correct so the switch 844 * code knows what the current MSR values are. 845 */ 846 pcb->pcb_fsbase = 0; /* Values loaded from PCB on switch */ 847 pcb->pcb_gsbase = 0; 848 /* Initialize the npx (if any) for the current process. */ 849 npxinit(__INITIAL_NPXCW__); 850 crit_exit(); 851 852 /* 853 * note: linux emulator needs edx to be 0x0 on entry, which is 854 * handled in execve simply by setting the 64 bit syscall 855 * return value to 0. 856 */ 857 } 858 859 void 860 cpu_setregs(void) 861 { 862 #if 0 863 unsigned int cr0; 864 865 cr0 = rcr0(); 866 cr0 |= CR0_NE; /* Done by npxinit() */ 867 cr0 |= CR0_MP | CR0_TS; /* Done at every execve() too. */ 868 cr0 |= CR0_WP | CR0_AM; 869 load_cr0(cr0); 870 load_gs(_udatasel); 871 #endif 872 } 873 874 static int 875 sysctl_machdep_adjkerntz(SYSCTL_HANDLER_ARGS) 876 { 877 int error; 878 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, 879 req); 880 if (!error && req->newptr) 881 resettodr(); 882 return (error); 883 } 884 885 SYSCTL_PROC(_machdep, CPU_ADJKERNTZ, adjkerntz, CTLTYPE_INT|CTLFLAG_RW, 886 &adjkerntz, 0, sysctl_machdep_adjkerntz, "I", ""); 887 888 extern u_long bootdev; /* not a cdev_t - encoding is different */ 889 SYSCTL_ULONG(_machdep, OID_AUTO, guessed_bootdev, 890 CTLFLAG_RD, &bootdev, 0, "Boot device (not in cdev_t format)"); 891 892 /* 893 * Initialize 386 and configure to run kernel 894 */ 895 896 /* 897 * Initialize segments & interrupt table 898 */ 899 900 extern struct user *proc0paddr; 901 902 #if 0 903 904 extern inthand_t 905 IDTVEC(div), IDTVEC(dbg), IDTVEC(nmi), IDTVEC(bpt), IDTVEC(ofl), 906 IDTVEC(bnd), IDTVEC(ill), IDTVEC(dna), IDTVEC(fpusegm), 907 IDTVEC(tss), IDTVEC(missing), IDTVEC(stk), IDTVEC(prot), 908 IDTVEC(page), IDTVEC(mchk), IDTVEC(rsvd), IDTVEC(fpu), IDTVEC(align), 909 IDTVEC(xmm), IDTVEC(dblfault), 910 IDTVEC(fast_syscall), IDTVEC(fast_syscall32); 911 #endif 912 913 #ifdef DEBUG_INTERRUPTS 914 extern inthand_t *Xrsvdary[256]; 915 #endif 916 917 int 918 ptrace_set_pc(struct lwp *lp, unsigned long addr) 919 { 920 lp->lwp_md.md_regs->tf_rip = addr; 921 return (0); 922 } 923 924 int 925 ptrace_single_step(struct lwp *lp) 926 { 927 lp->lwp_md.md_regs->tf_rflags |= PSL_T; 928 return (0); 929 } 930 931 int 932 fill_regs(struct lwp *lp, struct reg *regs) 933 { 934 struct trapframe *tp; 935 936 tp = lp->lwp_md.md_regs; 937 bcopy(&tp->tf_rdi, ®s->r_rdi, sizeof(*regs)); 938 return (0); 939 } 940 941 int 942 set_regs(struct lwp *lp, struct reg *regs) 943 { 944 struct trapframe *tp; 945 946 tp = lp->lwp_md.md_regs; 947 if (!EFL_SECURE(regs->r_rflags, tp->tf_rflags) || 948 !CS_SECURE(regs->r_cs)) 949 return (EINVAL); 950 bcopy(®s->r_rdi, &tp->tf_rdi, sizeof(*regs)); 951 return (0); 952 } 953 954 #ifndef CPU_DISABLE_SSE 955 static void 956 fill_fpregs_xmm(struct savexmm *sv_xmm, struct save87 *sv_87) 957 { 958 struct env87 *penv_87 = &sv_87->sv_env; 959 struct envxmm *penv_xmm = &sv_xmm->sv_env; 960 int i; 961 962 /* FPU control/status */ 963 penv_87->en_cw = penv_xmm->en_cw; 964 penv_87->en_sw = penv_xmm->en_sw; 965 penv_87->en_tw = penv_xmm->en_tw; 966 penv_87->en_fip = penv_xmm->en_fip; 967 penv_87->en_fcs = penv_xmm->en_fcs; 968 penv_87->en_opcode = penv_xmm->en_opcode; 969 penv_87->en_foo = penv_xmm->en_foo; 970 penv_87->en_fos = penv_xmm->en_fos; 971 972 /* FPU registers */ 973 for (i = 0; i < 8; ++i) 974 sv_87->sv_ac[i] = sv_xmm->sv_fp[i].fp_acc; 975 } 976 977 static void 978 set_fpregs_xmm(struct save87 *sv_87, struct savexmm *sv_xmm) 979 { 980 struct env87 *penv_87 = &sv_87->sv_env; 981 struct envxmm *penv_xmm = &sv_xmm->sv_env; 982 int i; 983 984 /* FPU control/status */ 985 penv_xmm->en_cw = penv_87->en_cw; 986 penv_xmm->en_sw = penv_87->en_sw; 987 penv_xmm->en_tw = penv_87->en_tw; 988 penv_xmm->en_fip = penv_87->en_fip; 989 penv_xmm->en_fcs = penv_87->en_fcs; 990 penv_xmm->en_opcode = penv_87->en_opcode; 991 penv_xmm->en_foo = penv_87->en_foo; 992 penv_xmm->en_fos = penv_87->en_fos; 993 994 /* FPU registers */ 995 for (i = 0; i < 8; ++i) 996 sv_xmm->sv_fp[i].fp_acc = sv_87->sv_ac[i]; 997 } 998 #endif /* CPU_DISABLE_SSE */ 999 1000 int 1001 fill_fpregs(struct lwp *lp, struct fpreg *fpregs) 1002 { 1003 #ifndef CPU_DISABLE_SSE 1004 if (cpu_fxsr) { 1005 fill_fpregs_xmm(&lp->lwp_thread->td_pcb->pcb_save.sv_xmm, 1006 (struct save87 *)fpregs); 1007 return (0); 1008 } 1009 #endif /* CPU_DISABLE_SSE */ 1010 bcopy(&lp->lwp_thread->td_pcb->pcb_save.sv_87, fpregs, sizeof *fpregs); 1011 return (0); 1012 } 1013 1014 int 1015 set_fpregs(struct lwp *lp, struct fpreg *fpregs) 1016 { 1017 #ifndef CPU_DISABLE_SSE 1018 if (cpu_fxsr) { 1019 set_fpregs_xmm((struct save87 *)fpregs, 1020 &lp->lwp_thread->td_pcb->pcb_save.sv_xmm); 1021 return (0); 1022 } 1023 #endif /* CPU_DISABLE_SSE */ 1024 bcopy(fpregs, &lp->lwp_thread->td_pcb->pcb_save.sv_87, sizeof *fpregs); 1025 return (0); 1026 } 1027 1028 int 1029 fill_dbregs(struct lwp *lp, struct dbreg *dbregs) 1030 { 1031 return (ENOSYS); 1032 } 1033 1034 int 1035 set_dbregs(struct lwp *lp, struct dbreg *dbregs) 1036 { 1037 return (ENOSYS); 1038 } 1039 1040 #if 0 1041 /* 1042 * Return > 0 if a hardware breakpoint has been hit, and the 1043 * breakpoint was in user space. Return 0, otherwise. 1044 */ 1045 int 1046 user_dbreg_trap(void) 1047 { 1048 u_int32_t dr7, dr6; /* debug registers dr6 and dr7 */ 1049 u_int32_t bp; /* breakpoint bits extracted from dr6 */ 1050 int nbp; /* number of breakpoints that triggered */ 1051 caddr_t addr[4]; /* breakpoint addresses */ 1052 int i; 1053 1054 dr7 = rdr7(); 1055 if ((dr7 & 0x000000ff) == 0) { 1056 /* 1057 * all GE and LE bits in the dr7 register are zero, 1058 * thus the trap couldn't have been caused by the 1059 * hardware debug registers 1060 */ 1061 return 0; 1062 } 1063 1064 nbp = 0; 1065 dr6 = rdr6(); 1066 bp = dr6 & 0x0000000f; 1067 1068 if (!bp) { 1069 /* 1070 * None of the breakpoint bits are set meaning this 1071 * trap was not caused by any of the debug registers 1072 */ 1073 return 0; 1074 } 1075 1076 /* 1077 * at least one of the breakpoints were hit, check to see 1078 * which ones and if any of them are user space addresses 1079 */ 1080 1081 if (bp & 0x01) { 1082 addr[nbp++] = (caddr_t)rdr0(); 1083 } 1084 if (bp & 0x02) { 1085 addr[nbp++] = (caddr_t)rdr1(); 1086 } 1087 if (bp & 0x04) { 1088 addr[nbp++] = (caddr_t)rdr2(); 1089 } 1090 if (bp & 0x08) { 1091 addr[nbp++] = (caddr_t)rdr3(); 1092 } 1093 1094 for (i=0; i<nbp; i++) { 1095 if (addr[i] < 1096 (caddr_t)VM_MAX_USER_ADDRESS) { 1097 /* 1098 * addr[i] is in user space 1099 */ 1100 return nbp; 1101 } 1102 } 1103 1104 /* 1105 * None of the breakpoints are in user space. 1106 */ 1107 return 0; 1108 } 1109 1110 #endif 1111 1112 void 1113 identcpu(void) 1114 { 1115 int regs[4]; 1116 1117 do_cpuid(1, regs); 1118 cpu_feature = regs[3]; 1119 } 1120 1121 1122 #ifndef DDB 1123 void 1124 Debugger(const char *msg) 1125 { 1126 kprintf("Debugger(\"%s\") called.\n", msg); 1127 } 1128 #endif /* no DDB */ 1129