1 /* $NetBSD: linux_machdep.c,v 1.26 2003/10/08 00:39:57 thorpej Exp $ */ 2 3 /*- 4 * Copyright (c) 1995, 2000, 2001 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Frank van der Linden and Emmanuel Dreyfus. 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 NetBSD 21 * Foundation, Inc. and its contributors. 22 * 4. Neither the name of The NetBSD Foundation nor the names of its 23 * contributors may be used to endorse or promote products derived 24 * from this software without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38 39 #include <sys/cdefs.h> 40 __KERNEL_RCSID(0, "$NetBSD: linux_machdep.c,v 1.26 2003/10/08 00:39:57 thorpej Exp $"); 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/signalvar.h> 45 #include <sys/kernel.h> 46 #include <sys/proc.h> 47 #include <sys/user.h> 48 #include <sys/buf.h> 49 #include <sys/reboot.h> 50 #include <sys/conf.h> 51 #include <sys/exec.h> 52 #include <sys/file.h> 53 #include <sys/callout.h> 54 #include <sys/malloc.h> 55 #include <sys/mbuf.h> 56 #include <sys/msgbuf.h> 57 #include <sys/mount.h> 58 #include <sys/vnode.h> 59 #include <sys/device.h> 60 #include <sys/sa.h> 61 #include <sys/syscallargs.h> 62 #include <sys/filedesc.h> 63 #include <sys/exec_elf.h> 64 #include <sys/disklabel.h> 65 #include <sys/ioctl.h> 66 #include <miscfs/specfs/specdev.h> 67 68 #include <compat/linux/common/linux_types.h> 69 #include <compat/linux/common/linux_signal.h> 70 #include <compat/linux/common/linux_util.h> 71 #include <compat/linux/common/linux_ioctl.h> 72 #include <compat/linux/common/linux_hdio.h> 73 #include <compat/linux/common/linux_exec.h> 74 #include <compat/linux/common/linux_machdep.h> 75 76 #include <compat/linux/linux_syscallargs.h> 77 78 #include <machine/cpu.h> 79 #include <machine/fpu.h> 80 #include <machine/psl.h> 81 #include <machine/reg.h> 82 #include <machine/vmparam.h> 83 84 /* 85 * To see whether wscons is configured (for virtual console ioctl calls). 86 */ 87 #if defined(_KERNEL_OPT) 88 #include "wsdisplay.h" 89 #endif 90 #if (NWSDISPLAY > 0) 91 #include <dev/wscons/wsconsio.h> 92 #include <dev/wscons/wsdisplay_usl_io.h> 93 #endif 94 95 /* 96 * Set set up registers on exec. 97 * XXX not used at the moment since in sys/kern/exec_conf, LINUX_COMPAT 98 * entry uses NetBSD's native setregs instead of linux_setregs 99 */ 100 void 101 linux_setregs(l, pack, stack) 102 struct lwp *l; 103 struct exec_package *pack; 104 u_long stack; 105 { 106 setregs(l, pack, stack); 107 } 108 109 /* 110 * Send an interrupt to process. 111 * 112 * Adapted from arch/powerpc/powerpc/sig_machdep.c:sendsig and 113 * compat/linux/arch/i386/linux_machdep.c:linux_sendsig 114 * 115 * XXX Does not work well yet with RT signals 116 * 117 */ 118 119 #ifdef __HAVE_SIGINFO 120 void 121 linux_sendsig(const ksiginfo_t *ksi, const sigset_t *mask) 122 #else 123 void 124 linux_sendsig(sig, mask, code) /* XXX Check me */ 125 int sig; 126 const sigset_t *mask; 127 u_long code; 128 #endif 129 { 130 #ifdef __HAVE_SIGINFO 131 const int sig = ksi->ksi_signo; 132 #endif 133 struct lwp *l = curlwp; 134 struct proc *p = l->l_proc; 135 struct trapframe *tf; 136 sig_t catcher = SIGACTION(p, sig).sa_handler; 137 struct linux_sigregs frame; 138 struct linux_pt_regs linux_regs; 139 struct linux_sigcontext sc; 140 register_t fp; 141 int onstack; 142 int i; 143 144 tf = trapframe(l); 145 146 /* 147 * Do we need to jump onto the signal stack? 148 */ 149 onstack = 150 (p->p_sigctx.ps_sigstk.ss_flags & (SS_DISABLE | SS_ONSTACK)) == 0 && 151 (SIGACTION(p, sig).sa_flags & SA_ONSTACK) != 0; 152 153 /* 154 * Signal stack is broken (see at the end of linux_sigreturn), so we do 155 * not use it yet. XXX fix this. 156 */ 157 onstack=0; 158 159 /* 160 * Allocate space for the signal handler context. 161 */ 162 if (onstack) { 163 fp = (register_t) 164 ((caddr_t)p->p_sigctx.ps_sigstk.ss_sp + 165 p->p_sigctx.ps_sigstk.ss_size); 166 } else { 167 fp = tf->fixreg[1]; 168 } 169 #ifdef DEBUG_LINUX 170 printf("fp at start of linux_sendsig = %x\n", fp); 171 #endif 172 fp -= sizeof(struct linux_sigregs); 173 fp &= ~0xf; 174 175 /* 176 * Prepare a sigcontext for later. 177 */ 178 memset(&sc, 0, sizeof sc); 179 sc.lsignal = (int)native_to_linux_signo[sig]; 180 sc.lhandler = (unsigned long)catcher; 181 native_to_linux_old_extra_sigset(&sc.lmask, &sc._unused[3], mask); 182 sc.lregs = (struct linux_pt_regs*)fp; 183 184 /* 185 * Setup the signal stack frame as Linux does it in 186 * arch/ppc/kernel/signal.c:setup_frame() 187 * 188 * Save register context. 189 */ 190 for (i = 0; i < 32; i++) 191 linux_regs.lgpr[i] = tf->fixreg[i]; 192 linux_regs.lnip = tf->srr0; 193 linux_regs.lmsr = tf->srr1; 194 linux_regs.lorig_gpr3 = tf->fixreg[3]; /* XXX Is that right? */ 195 linux_regs.lctr = tf->ctr; 196 linux_regs.llink = tf->lr; 197 linux_regs.lxer = tf->xer; 198 linux_regs.lccr = tf->cr; 199 linux_regs.lmq = 0; /* Unused, 601 only */ 200 linux_regs.ltrap = tf->exc; 201 linux_regs.ldar = tf->dar; 202 linux_regs.ldsisr = tf->dsisr; 203 linux_regs.lresult = 0; 204 205 memset(&frame, 0, sizeof(frame)); 206 memcpy(&frame.lgp_regs, &linux_regs, sizeof(linux_regs)); 207 208 save_fpu_lwp(curlwp); 209 memcpy(&frame.lfp_regs, curpcb->pcb_fpu.fpr, sizeof(frame.lfp_regs)); 210 211 /* 212 * Copy Linux's signal trampoline on the user stack It should not 213 * be used, but Linux binaries might expect it to be there. 214 */ 215 frame.ltramp[0] = 0x38997777; /* li r0, 0x7777 */ 216 frame.ltramp[1] = 0x44000002; /* sc */ 217 218 /* 219 * Move it to the user stack 220 * There is a little trick here, about the LINUX_ABIGAP: the 221 * linux_sigreg structure has a 56 int gap to support rs6000/xcoff 222 * binaries. But the Linux kernel seems to do without it, and it 223 * just skip it when building the stack frame. Hence the LINUX_ABIGAP. 224 */ 225 if (copyout(&frame, (caddr_t)fp, sizeof (frame) - LINUX_ABIGAP) != 0) { 226 /* 227 * Process has trashed its stack; give it an illegal 228 * instruction to halt it in its tracks. 229 */ 230 sigexit(l, SIGILL); 231 /* NOTREACHED */ 232 } 233 234 /* 235 * Add a sigcontext on the stack 236 */ 237 fp -= sizeof(struct linux_sigcontext); 238 if (copyout(&sc, (caddr_t)fp, sizeof (struct linux_sigcontext)) != 0) { 239 /* 240 * Process has trashed its stack; give it an illegal 241 * instruction to halt it in its tracks. 242 */ 243 sigexit(l, SIGILL); 244 /* NOTREACHED */ 245 } 246 247 /* 248 * Set the registers according to how the Linux process expects them. 249 * "Mind the gap" Linux expects a gap here. 250 */ 251 tf->fixreg[1] = fp - LINUX__SIGNAL_FRAMESIZE; 252 tf->lr = (int)catcher; 253 tf->fixreg[3] = (int)native_to_linux_signo[sig]; 254 tf->fixreg[4] = fp; 255 tf->srr0 = (int)p->p_sigctx.ps_sigcode; 256 257 #ifdef DEBUG_LINUX 258 printf("fp at end of linux_sendsig = %x\n", fp); 259 #endif 260 /* 261 * Remember that we're now on the signal stack. 262 */ 263 if (onstack) 264 p->p_sigctx.ps_sigstk.ss_flags |= SS_ONSTACK; 265 #ifdef DEBUG_LINUX 266 printf("linux_sendsig: exitting. fp=0x%lx\n",(long)fp); 267 #endif 268 } 269 270 /* 271 * System call to cleanup state after a signal 272 * has been taken. Reset signal mask and 273 * stack state from context left by sendsig (above). 274 * Return to previous pc and psl as specified by 275 * context left by sendsig. Check carefully to 276 * make sure that the user has not modified the 277 * psl to gain improper privileges or to cause 278 * a machine fault. 279 * 280 * XXX not tested 281 */ 282 int 283 linux_sys_rt_sigreturn(l, v, retval) 284 struct lwp *l; 285 void *v; 286 register_t *retval; 287 { 288 struct linux_sys_rt_sigreturn_args /* { 289 syscallarg(struct linux_rt_sigframe *) sfp; 290 } */ *uap = v; 291 struct proc *p = l->l_proc; 292 struct linux_rt_sigframe *scp, sigframe; 293 struct linux_sigregs sregs; 294 struct linux_pt_regs *lregs; 295 struct trapframe *tf; 296 sigset_t mask; 297 int i; 298 299 /* 300 * The trampoline code hands us the context. 301 * It is unsafe to keep track of it ourselves, in the event that a 302 * program jumps out of a signal handler. 303 */ 304 scp = SCARG(uap, sfp); 305 306 /* 307 * Get the context from user stack 308 */ 309 if (copyin((caddr_t)scp, &sigframe, sizeof(*scp))) 310 return (EFAULT); 311 312 /* 313 * Make sure, fpu is sync'ed 314 */ 315 save_fpu_lwp(curlwp); 316 317 /* 318 * Restore register context. 319 */ 320 if (copyin((caddr_t)sigframe.luc.luc_context.lregs, 321 &sregs, sizeof(sregs))) 322 return (EFAULT); 323 lregs = (struct linux_pt_regs *)&sregs.lgp_regs; 324 325 tf = trapframe(l); 326 #ifdef DEBUG_LINUX 327 printf("linux_sys_sigreturn: trapframe=0x%lx scp=0x%lx\n", 328 (unsigned long)tf, (unsigned long)scp); 329 #endif 330 331 if ((lregs->lmsr & PSL_USERSTATIC) != (tf->srr1 & PSL_USERSTATIC)) 332 return (EINVAL); 333 334 for (i = 0; i < 32; i++) 335 tf->fixreg[i] = lregs->lgpr[i]; 336 tf->lr = lregs->llink; 337 tf->cr = lregs->lccr; 338 tf->xer = lregs->lxer; 339 tf->ctr = lregs->lctr; 340 tf->srr0 = lregs->lnip; 341 tf->srr1 = lregs->lmsr; 342 343 memcpy(curpcb->pcb_fpu.fpr, (caddr_t)&sregs.lfp_regs, 344 sizeof(curpcb->pcb_fpu.fpr)); 345 346 /* 347 * Restore signal stack. 348 * 349 * XXX cannot find the onstack information in Linux sig context. 350 * Is signal stack really supported on Linux? 351 * 352 * It seems to be supported in libc6... 353 */ 354 /* if (sc.sc_onstack & SS_ONSTACK) 355 p->p_sigctx.ps_sigstk.ss_flags |= SS_ONSTACK; 356 else */ 357 p->p_sigctx.ps_sigstk.ss_flags &= ~SS_ONSTACK; 358 359 /* 360 * Grab the signal mask 361 */ 362 linux_to_native_sigset(&mask, &sigframe.luc.luc_sigmask); 363 (void) sigprocmask1(p, SIG_SETMASK, &mask, 0); 364 365 return (EJUSTRETURN); 366 } 367 368 369 /* 370 * The following needs code review for potential security issues 371 */ 372 int 373 linux_sys_sigreturn(l, v, retval) 374 struct lwp *l; 375 void *v; 376 register_t *retval; 377 { 378 struct linux_sys_sigreturn_args /* { 379 syscallarg(struct linux_sigcontext *) scp; 380 } */ *uap = v; 381 struct proc *p = l->l_proc; 382 struct linux_sigcontext *scp, context; 383 struct linux_sigregs sregs; 384 struct linux_pt_regs *lregs; 385 struct trapframe *tf; 386 sigset_t mask; 387 int i; 388 389 /* 390 * The trampoline code hands us the context. 391 * It is unsafe to keep track of it ourselves, in the event that a 392 * program jumps out of a signal handler. 393 */ 394 scp = SCARG(uap, scp); 395 396 /* 397 * Get the context from user stack 398 */ 399 if (copyin(scp, &context, sizeof(*scp))) 400 return (EFAULT); 401 402 /* 403 * Make sure, fpu is in sync 404 */ 405 save_fpu_lwp(curlwp); 406 407 /* 408 * Restore register context. 409 */ 410 if (copyin((caddr_t)context.lregs, &sregs, sizeof(sregs))) 411 return (EFAULT); 412 lregs = (struct linux_pt_regs *)&sregs.lgp_regs; 413 414 tf = trapframe(l); 415 #ifdef DEBUG_LINUX 416 printf("linux_sys_sigreturn: trapframe=0x%lx scp=0x%lx\n", 417 (unsigned long)tf, (unsigned long)scp); 418 #endif 419 420 if ((lregs->lmsr & PSL_USERSTATIC) != (tf->srr1 & PSL_USERSTATIC)) 421 return (EINVAL); 422 423 for (i = 0; i < 32; i++) 424 tf->fixreg[i] = lregs->lgpr[i]; 425 tf->lr = lregs->llink; 426 tf->cr = lregs->lccr; 427 tf->xer = lregs->lxer; 428 tf->ctr = lregs->lctr; 429 tf->srr0 = lregs->lnip; 430 tf->srr1 = lregs->lmsr; 431 432 memcpy(curpcb->pcb_fpu.fpr, (caddr_t)&sregs.lfp_regs, 433 sizeof(curpcb->pcb_fpu.fpr)); 434 435 /* 436 * Restore signal stack. 437 * 438 * XXX cannot find the onstack information in Linux sig context. 439 * Is signal stack really supported on Linux? 440 */ 441 #if 0 442 if (sc.sc_onstack & SS_ONSTACK) 443 p->p_sigctx.ps_sigstk.ss_flags |= SS_ONSTACK; 444 else 445 #endif 446 p->p_sigctx.ps_sigstk.ss_flags &= ~SS_ONSTACK; 447 448 /* Restore signal mask. */ 449 linux_old_extra_to_native_sigset(&mask, &context.lmask, 450 &context._unused[3]); 451 (void) sigprocmask1(p, SIG_SETMASK, &mask, 0); 452 453 return (EJUSTRETURN); 454 } 455 456 457 #if 0 458 int 459 linux_sys_modify_ldt(p, v, retval) 460 struct proc *p; 461 void *v; 462 register_t *retval; 463 { 464 /* 465 * This syscall is not implemented in Linux/PowerPC: we should not 466 * be here 467 */ 468 #ifdef DEBUG_LINUX 469 printf("linux_sys_modify_ldt: should not be here.\n"); 470 #endif 471 return 0; 472 } 473 #endif 474 475 /* 476 * major device numbers remapping 477 */ 478 dev_t 479 linux_fakedev(dev, raw) 480 dev_t dev; 481 int raw; 482 { 483 /* XXX write me */ 484 return dev; 485 } 486 487 /* 488 * We come here in a last attempt to satisfy a Linux ioctl() call 489 */ 490 int 491 linux_machdepioctl(p, v, retval) 492 struct proc *p; 493 void *v; 494 register_t *retval; 495 { 496 struct linux_sys_ioctl_args /* { 497 syscallarg(int) fd; 498 syscallarg(u_long) com; 499 syscallarg(caddr_t) data; 500 } */ *uap = v; 501 struct sys_ioctl_args bia; 502 u_long com; 503 504 SCARG(&bia, fd) = SCARG(uap, fd); 505 SCARG(&bia, data) = SCARG(uap, data); 506 com = SCARG(uap, com); 507 508 switch (com) { 509 default: 510 printf("linux_machdepioctl: invalid ioctl %08lx\n", com); 511 return EINVAL; 512 } 513 SCARG(&bia, com) = com; 514 /* XXX NJWLWP */ 515 return sys_ioctl(curlwp, &bia, retval); 516 } 517 #if 0 518 /* 519 * Set I/O permissions for a process. Just set the maximum level 520 * right away (ignoring the argument), otherwise we would have 521 * to rely on I/O permission maps, which are not implemented. 522 */ 523 int 524 linux_sys_iopl(l, v, retval) 525 struct lwp *l; 526 void *v; 527 register_t *retval; 528 { 529 /* 530 * This syscall is not implemented in Linux/PowerPC: we should not be here 531 */ 532 #ifdef DEBUG_LINUX 533 printf("linux_sys_iopl: should not be here.\n"); 534 #endif 535 return 0; 536 } 537 #endif 538 539 /* 540 * See above. If a root process tries to set access to an I/O port, 541 * just let it have the whole range. 542 */ 543 int 544 linux_sys_ioperm(l, v, retval) 545 struct lwp *l; 546 void *v; 547 register_t *retval; 548 { 549 /* 550 * This syscall is not implemented in Linux/PowerPC: we should not be here 551 */ 552 #ifdef DEBUG_LINUX 553 printf("linux_sys_ioperm: should not be here.\n"); 554 #endif 555 return 0; 556 } 557 558 /* 559 * wrapper linux_sys_new_uname() -> linux_sys_uname() 560 */ 561 int 562 linux_sys_new_uname(l, v, retval) 563 struct lwp *l; 564 void *v; 565 register_t *retval; 566 { 567 return linux_sys_uname(l, v, retval); 568 } 569 570 /* 571 * wrapper linux_sys_new_select() -> linux_sys_select() 572 */ 573 int 574 linux_sys_new_select(l, v, retval) 575 struct lwp *l; 576 void *v; 577 register_t *retval; 578 { 579 return linux_sys_select(l, v, retval); 580 } 581