1 /* $NetBSD: linux_machdep.c,v 1.31 2005/12/11 12:20:16 christos 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 1.31 2005/12/11 12:20:16 christos 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 void 120 linux_sendsig(const ksiginfo_t *ksi, const sigset_t *mask) 121 { 122 const int sig = ksi->ksi_signo; 123 struct lwp *l = curlwp; 124 struct proc *p = l->l_proc; 125 struct trapframe *tf; 126 sig_t catcher = SIGACTION(p, sig).sa_handler; 127 struct linux_sigregs frame; 128 struct linux_pt_regs linux_regs; 129 struct linux_sigcontext sc; 130 register_t fp; 131 int onstack; 132 int i; 133 134 tf = trapframe(l); 135 136 /* 137 * Do we need to jump onto the signal stack? 138 */ 139 onstack = 140 (p->p_sigctx.ps_sigstk.ss_flags & (SS_DISABLE | SS_ONSTACK)) == 0 && 141 (SIGACTION(p, sig).sa_flags & SA_ONSTACK) != 0; 142 143 /* 144 * Signal stack is broken (see at the end of linux_sigreturn), so we do 145 * not use it yet. XXX fix this. 146 */ 147 onstack=0; 148 149 /* 150 * Allocate space for the signal handler context. 151 */ 152 if (onstack) { 153 fp = (register_t) 154 ((caddr_t)p->p_sigctx.ps_sigstk.ss_sp + 155 p->p_sigctx.ps_sigstk.ss_size); 156 } else { 157 fp = tf->fixreg[1]; 158 } 159 #ifdef DEBUG_LINUX 160 printf("fp at start of linux_sendsig = %x\n", fp); 161 #endif 162 fp -= sizeof(struct linux_sigregs); 163 fp &= ~0xf; 164 165 /* 166 * Prepare a sigcontext for later. 167 */ 168 memset(&sc, 0, sizeof sc); 169 sc.lsignal = (int)native_to_linux_signo[sig]; 170 sc.lhandler = (unsigned long)catcher; 171 native_to_linux_old_extra_sigset(&sc.lmask, &sc._unused[3], mask); 172 sc.lregs = (struct linux_pt_regs*)fp; 173 174 /* 175 * Setup the signal stack frame as Linux does it in 176 * arch/ppc/kernel/signal.c:setup_frame() 177 * 178 * Save register context. 179 */ 180 for (i = 0; i < 32; i++) 181 linux_regs.lgpr[i] = tf->fixreg[i]; 182 linux_regs.lnip = tf->srr0; 183 linux_regs.lmsr = tf->srr1 & PSL_USERSRR1; 184 linux_regs.lorig_gpr3 = tf->fixreg[3]; /* XXX Is that right? */ 185 linux_regs.lctr = tf->ctr; 186 linux_regs.llink = tf->lr; 187 linux_regs.lxer = tf->xer; 188 linux_regs.lccr = tf->cr; 189 linux_regs.lmq = 0; /* Unused, 601 only */ 190 linux_regs.ltrap = tf->exc; 191 linux_regs.ldar = tf->dar; 192 linux_regs.ldsisr = tf->dsisr; 193 linux_regs.lresult = 0; 194 195 memset(&frame, 0, sizeof(frame)); 196 memcpy(&frame.lgp_regs, &linux_regs, sizeof(linux_regs)); 197 198 save_fpu_lwp(curlwp, FPU_SAVE); 199 memcpy(&frame.lfp_regs, curpcb->pcb_fpu.fpreg, sizeof(frame.lfp_regs)); 200 201 /* 202 * Copy Linux's signal trampoline on the user stack It should not 203 * be used, but Linux binaries might expect it to be there. 204 */ 205 frame.ltramp[0] = 0x38997777; /* li r0, 0x7777 */ 206 frame.ltramp[1] = 0x44000002; /* sc */ 207 208 /* 209 * Move it to the user stack 210 * There is a little trick here, about the LINUX_ABIGAP: the 211 * linux_sigreg structure has a 56 int gap to support rs6000/xcoff 212 * binaries. But the Linux kernel seems to do without it, and it 213 * just skip it when building the stack frame. Hence the LINUX_ABIGAP. 214 */ 215 if (copyout(&frame, (caddr_t)fp, sizeof (frame) - LINUX_ABIGAP) != 0) { 216 /* 217 * Process has trashed its stack; give it an illegal 218 * instruction to halt it in its tracks. 219 */ 220 sigexit(l, SIGILL); 221 /* NOTREACHED */ 222 } 223 224 /* 225 * Add a sigcontext on the stack 226 */ 227 fp -= sizeof(struct linux_sigcontext); 228 if (copyout(&sc, (caddr_t)fp, sizeof (struct linux_sigcontext)) != 0) { 229 /* 230 * Process has trashed its stack; give it an illegal 231 * instruction to halt it in its tracks. 232 */ 233 sigexit(l, SIGILL); 234 /* NOTREACHED */ 235 } 236 237 /* 238 * Set the registers according to how the Linux process expects them. 239 * "Mind the gap" Linux expects a gap here. 240 */ 241 tf->fixreg[1] = fp - LINUX__SIGNAL_FRAMESIZE; 242 tf->lr = (int)catcher; 243 tf->fixreg[3] = (int)native_to_linux_signo[sig]; 244 tf->fixreg[4] = fp; 245 tf->srr0 = (int)p->p_sigctx.ps_sigcode; 246 247 #ifdef DEBUG_LINUX 248 printf("fp at end of linux_sendsig = %x\n", fp); 249 #endif 250 /* 251 * Remember that we're now on the signal stack. 252 */ 253 if (onstack) 254 p->p_sigctx.ps_sigstk.ss_flags |= SS_ONSTACK; 255 #ifdef DEBUG_LINUX 256 printf("linux_sendsig: exitting. fp=0x%lx\n",(long)fp); 257 #endif 258 } 259 260 /* 261 * System call to cleanup state after a signal 262 * has been taken. Reset signal mask and 263 * stack state from context left by sendsig (above). 264 * Return to previous pc and psl as specified by 265 * context left by sendsig. Check carefully to 266 * make sure that the user has not modified the 267 * psl to gain improper privileges or to cause 268 * a machine fault. 269 * 270 * XXX not tested 271 */ 272 int 273 linux_sys_rt_sigreturn(l, v, retval) 274 struct lwp *l; 275 void *v; 276 register_t *retval; 277 { 278 struct linux_sys_rt_sigreturn_args /* { 279 syscallarg(struct linux_rt_sigframe *) sfp; 280 } */ *uap = v; 281 struct proc *p = l->l_proc; 282 struct linux_rt_sigframe *scp, sigframe; 283 struct linux_sigregs sregs; 284 struct linux_pt_regs *lregs; 285 struct trapframe *tf; 286 sigset_t mask; 287 int i; 288 289 /* 290 * The trampoline code hands us the context. 291 * It is unsafe to keep track of it ourselves, in the event that a 292 * program jumps out of a signal handler. 293 */ 294 scp = SCARG(uap, sfp); 295 296 /* 297 * Get the context from user stack 298 */ 299 if (copyin((caddr_t)scp, &sigframe, sizeof(*scp))) 300 return (EFAULT); 301 302 /* 303 * Restore register context. 304 */ 305 if (copyin((caddr_t)sigframe.luc.luc_context.lregs, 306 &sregs, sizeof(sregs))) 307 return (EFAULT); 308 lregs = (struct linux_pt_regs *)&sregs.lgp_regs; 309 310 tf = trapframe(l); 311 #ifdef DEBUG_LINUX 312 printf("linux_sys_sigreturn: trapframe=0x%lx scp=0x%lx\n", 313 (unsigned long)tf, (unsigned long)scp); 314 #endif 315 316 if (!PSL_USEROK_P(lregs->lmsr)) 317 return (EINVAL); 318 319 for (i = 0; i < 32; i++) 320 tf->fixreg[i] = lregs->lgpr[i]; 321 tf->lr = lregs->llink; 322 tf->cr = lregs->lccr; 323 tf->xer = lregs->lxer; 324 tf->ctr = lregs->lctr; 325 tf->srr0 = lregs->lnip; 326 tf->srr1 = lregs->lmsr; 327 328 /* 329 * Make sure the fpu state is discarded 330 */ 331 save_fpu_lwp(curlwp, FPU_DISCARD); 332 333 memcpy(curpcb->pcb_fpu.fpreg, (caddr_t)&sregs.lfp_regs, 334 sizeof(curpcb->pcb_fpu.fpreg)); 335 336 /* 337 * Restore signal stack. 338 * 339 * XXX cannot find the onstack information in Linux sig context. 340 * Is signal stack really supported on Linux? 341 * 342 * It seems to be supported in libc6... 343 */ 344 /* if (sc.sc_onstack & SS_ONSTACK) 345 p->p_sigctx.ps_sigstk.ss_flags |= SS_ONSTACK; 346 else */ 347 p->p_sigctx.ps_sigstk.ss_flags &= ~SS_ONSTACK; 348 349 /* 350 * Grab the signal mask 351 */ 352 linux_to_native_sigset(&mask, &sigframe.luc.luc_sigmask); 353 (void) sigprocmask1(p, SIG_SETMASK, &mask, 0); 354 355 return (EJUSTRETURN); 356 } 357 358 359 /* 360 * The following needs code review for potential security issues 361 */ 362 int 363 linux_sys_sigreturn(l, v, retval) 364 struct lwp *l; 365 void *v; 366 register_t *retval; 367 { 368 struct linux_sys_sigreturn_args /* { 369 syscallarg(struct linux_sigcontext *) scp; 370 } */ *uap = v; 371 struct proc *p = l->l_proc; 372 struct linux_sigcontext *scp, context; 373 struct linux_sigregs sregs; 374 struct linux_pt_regs *lregs; 375 struct trapframe *tf; 376 sigset_t mask; 377 int i; 378 379 /* 380 * The trampoline code hands us the context. 381 * It is unsafe to keep track of it ourselves, in the event that a 382 * program jumps out of a signal handler. 383 */ 384 scp = SCARG(uap, scp); 385 386 /* 387 * Get the context from user stack 388 */ 389 if (copyin(scp, &context, sizeof(*scp))) 390 return (EFAULT); 391 392 /* 393 * Restore register context. 394 */ 395 if (copyin((caddr_t)context.lregs, &sregs, sizeof(sregs))) 396 return (EFAULT); 397 lregs = (struct linux_pt_regs *)&sregs.lgp_regs; 398 399 tf = trapframe(l); 400 #ifdef DEBUG_LINUX 401 printf("linux_sys_sigreturn: trapframe=0x%lx scp=0x%lx\n", 402 (unsigned long)tf, (unsigned long)scp); 403 #endif 404 405 if (!PSL_USEROK_P(lregs->lmsr)) 406 return (EINVAL); 407 408 for (i = 0; i < 32; i++) 409 tf->fixreg[i] = lregs->lgpr[i]; 410 tf->lr = lregs->llink; 411 tf->cr = lregs->lccr; 412 tf->xer = lregs->lxer; 413 tf->ctr = lregs->lctr; 414 tf->srr0 = lregs->lnip; 415 tf->srr1 = lregs->lmsr; 416 417 /* 418 * Make sure the fpu state is discarded 419 */ 420 save_fpu_lwp(curlwp, FPU_DISCARD); 421 422 memcpy(curpcb->pcb_fpu.fpreg, (caddr_t)&sregs.lfp_regs, 423 sizeof(curpcb->pcb_fpu.fpreg)); 424 425 /* 426 * Restore signal stack. 427 * 428 * XXX cannot find the onstack information in Linux sig context. 429 * Is signal stack really supported on Linux? 430 */ 431 #if 0 432 if (sc.sc_onstack & SS_ONSTACK) 433 p->p_sigctx.ps_sigstk.ss_flags |= SS_ONSTACK; 434 else 435 #endif 436 p->p_sigctx.ps_sigstk.ss_flags &= ~SS_ONSTACK; 437 438 /* Restore signal mask. */ 439 linux_old_extra_to_native_sigset(&mask, &context.lmask, 440 &context._unused[3]); 441 (void) sigprocmask1(p, SIG_SETMASK, &mask, 0); 442 443 return (EJUSTRETURN); 444 } 445 446 447 #if 0 448 int 449 linux_sys_modify_ldt(p, v, retval) 450 struct proc *p; 451 void *v; 452 register_t *retval; 453 { 454 /* 455 * This syscall is not implemented in Linux/PowerPC: we should not 456 * be here 457 */ 458 #ifdef DEBUG_LINUX 459 printf("linux_sys_modify_ldt: should not be here.\n"); 460 #endif 461 return 0; 462 } 463 #endif 464 465 /* 466 * major device numbers remapping 467 */ 468 dev_t 469 linux_fakedev(dev, raw) 470 dev_t dev; 471 int raw; 472 { 473 /* XXX write me */ 474 return dev; 475 } 476 477 /* 478 * We come here in a last attempt to satisfy a Linux ioctl() call 479 */ 480 int 481 linux_machdepioctl(l, v, retval) 482 struct lwp *l; 483 void *v; 484 register_t *retval; 485 { 486 struct linux_sys_ioctl_args /* { 487 syscallarg(int) fd; 488 syscallarg(u_long) com; 489 syscallarg(caddr_t) data; 490 } */ *uap = v; 491 struct sys_ioctl_args bia; 492 u_long com; 493 494 SCARG(&bia, fd) = SCARG(uap, fd); 495 SCARG(&bia, data) = SCARG(uap, data); 496 com = SCARG(uap, com); 497 498 switch (com) { 499 default: 500 printf("linux_machdepioctl: invalid ioctl %08lx\n", com); 501 return EINVAL; 502 } 503 SCARG(&bia, com) = com; 504 /* XXX NJWLWP */ 505 return sys_ioctl(curlwp, &bia, retval); 506 } 507 #if 0 508 /* 509 * Set I/O permissions for a process. Just set the maximum level 510 * right away (ignoring the argument), otherwise we would have 511 * to rely on I/O permission maps, which are not implemented. 512 */ 513 int 514 linux_sys_iopl(l, v, retval) 515 struct lwp *l; 516 void *v; 517 register_t *retval; 518 { 519 /* 520 * This syscall is not implemented in Linux/PowerPC: we should not be here 521 */ 522 #ifdef DEBUG_LINUX 523 printf("linux_sys_iopl: should not be here.\n"); 524 #endif 525 return 0; 526 } 527 #endif 528 529 /* 530 * See above. If a root process tries to set access to an I/O port, 531 * just let it have the whole range. 532 */ 533 int 534 linux_sys_ioperm(l, v, retval) 535 struct lwp *l; 536 void *v; 537 register_t *retval; 538 { 539 /* 540 * This syscall is not implemented in Linux/PowerPC: we should not be here 541 */ 542 #ifdef DEBUG_LINUX 543 printf("linux_sys_ioperm: should not be here.\n"); 544 #endif 545 return 0; 546 } 547 548 /* 549 * wrapper linux_sys_new_uname() -> linux_sys_uname() 550 */ 551 int 552 linux_sys_new_uname(l, v, retval) 553 struct lwp *l; 554 void *v; 555 register_t *retval; 556 { 557 return linux_sys_uname(l, v, retval); 558 } 559 560 /* 561 * wrapper linux_sys_new_select() -> linux_sys_select() 562 */ 563 int 564 linux_sys_new_select(l, v, retval) 565 struct lwp *l; 566 void *v; 567 register_t *retval; 568 { 569 return linux_sys_select(l, v, retval); 570 } 571 572 int 573 linux_usertrap(struct lwp *l, vaddr_t trapaddr, void *arg) 574 { 575 return 0; 576 } 577