1 /* $NetBSD: linux_machdep.c,v 1.19 2002/09/25 22:21:35 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 1.19 2002/09/25 22:21:35 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/syscallargs.h> 61 #include <sys/filedesc.h> 62 #include <sys/exec_elf.h> 63 #include <sys/disklabel.h> 64 #include <sys/ioctl.h> 65 #include <miscfs/specfs/specdev.h> 66 67 #include <compat/linux/common/linux_types.h> 68 #include <compat/linux/common/linux_signal.h> 69 #include <compat/linux/common/linux_util.h> 70 #include <compat/linux/common/linux_ioctl.h> 71 #include <compat/linux/common/linux_hdio.h> 72 #include <compat/linux/common/linux_exec.h> 73 #include <compat/linux/common/linux_machdep.h> 74 75 #include <compat/linux/linux_syscallargs.h> 76 77 #include <machine/cpu.h> 78 #include <machine/fpu.h> 79 #include <machine/psl.h> 80 #include <machine/reg.h> 81 #include <machine/vmparam.h> 82 83 /* 84 * To see whether wscons is configured (for virtual console ioctl calls). 85 */ 86 #if defined(_KERNEL_OPT) 87 #include "wsdisplay.h" 88 #endif 89 #if (NWSDISPLAY > 0) 90 #include <dev/wscons/wsconsio.h> 91 #include <dev/wscons/wsdisplay_usl_io.h> 92 #endif 93 94 /* 95 * Set set up registers on exec. 96 * XXX not used at the moment since in sys/kern/exec_conf, LINUX_COMPAT 97 * entry uses NetBSD's native setregs instead of linux_setregs 98 */ 99 void 100 linux_setregs(p, pack, stack) 101 struct proc *p; 102 struct exec_package *pack; 103 u_long stack; 104 { 105 setregs(p, pack, stack); 106 } 107 108 /* 109 * Send an interrupt to process. 110 * 111 * Adapted from arch/powerpc/powerpc/sig_machdep.c:sendsig and 112 * compat/linux/arch/i386/linux_machdep.c:linux_sendsig 113 * 114 * XXX Does not work well yet with RT signals 115 * 116 */ 117 118 void 119 linux_sendsig(sig, mask, code) /* XXX Check me */ 120 int sig; 121 sigset_t *mask; 122 u_long code; 123 { 124 struct proc *p = curproc; 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(p); 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; 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_proc(curproc); 199 memcpy(&frame.lfp_regs, curpcb->pcb_fpu.fpr, 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(p, 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(p, 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(p, v, retval) 274 struct proc *p; 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 linux_rt_sigframe *scp, sigframe; 282 struct linux_sigregs sregs; 283 struct linux_pt_regs *lregs; 284 struct trapframe *tf; 285 sigset_t mask; 286 int i; 287 288 /* 289 * The trampoline code hands us the context. 290 * It is unsafe to keep track of it ourselves, in the event that a 291 * program jumps out of a signal handler. 292 */ 293 scp = SCARG(uap, sfp); 294 295 /* 296 * Get the context from user stack 297 */ 298 if (copyin((caddr_t)scp, &sigframe, sizeof(*scp))) 299 return (EFAULT); 300 301 /* 302 * Make sure, fpu is sync'ed 303 */ 304 save_fpu_proc(curproc); 305 306 /* 307 * Restore register context. 308 */ 309 if (copyin((caddr_t)sigframe.luc.luc_context.lregs, 310 &sregs, sizeof(sregs))) 311 return (EFAULT); 312 lregs = (struct linux_pt_regs *)&sregs.lgp_regs; 313 314 tf = trapframe(p); 315 #ifdef DEBUG_LINUX 316 printf("linux_sys_sigreturn: trapframe=0x%lx scp=0x%lx\n", 317 (unsigned long)tf, (unsigned long)scp); 318 #endif 319 320 if ((lregs->lmsr & PSL_USERSTATIC) != (tf->srr1 & PSL_USERSTATIC)) 321 return (EINVAL); 322 323 for (i = 0; i < 32; i++) 324 tf->fixreg[i] = lregs->lgpr[i]; 325 tf->lr = lregs->llink; 326 tf->cr = lregs->lccr; 327 tf->xer = lregs->lxer; 328 tf->ctr = lregs->lctr; 329 tf->srr0 = lregs->lnip; 330 tf->srr1 = lregs->lmsr; 331 332 memcpy(curpcb->pcb_fpu.fpr, (caddr_t)&sregs.lfp_regs, 333 sizeof(curpcb->pcb_fpu.fpr)); 334 335 /* 336 * Restore signal stack. 337 * 338 * XXX cannot find the onstack information in Linux sig context. 339 * Is signal stack really supported on Linux? 340 * 341 * It seems to be supported in libc6... 342 */ 343 /* if (sc.sc_onstack & SS_ONSTACK) 344 p->p_sigctx.ps_sigstk.ss_flags |= SS_ONSTACK; 345 else */ 346 p->p_sigctx.ps_sigstk.ss_flags &= ~SS_ONSTACK; 347 348 /* 349 * Grab the signal mask 350 */ 351 linux_to_native_sigset(&mask, &sigframe.luc.luc_sigmask); 352 (void) sigprocmask1(p, SIG_SETMASK, &mask, 0); 353 354 return (EJUSTRETURN); 355 } 356 357 358 /* 359 * The following needs code review for potential security issues 360 */ 361 int 362 linux_sys_sigreturn(p, v, retval) 363 struct proc *p; 364 void *v; 365 register_t *retval; 366 { 367 struct linux_sys_sigreturn_args /* { 368 syscallarg(struct linux_sigcontext *) scp; 369 } */ *uap = v; 370 struct linux_sigcontext *scp, context; 371 struct linux_sigregs sregs; 372 struct linux_pt_regs *lregs; 373 struct trapframe *tf; 374 sigset_t mask; 375 int i; 376 377 /* 378 * The trampoline code hands us the context. 379 * It is unsafe to keep track of it ourselves, in the event that a 380 * program jumps out of a signal handler. 381 */ 382 scp = SCARG(uap, scp); 383 384 /* 385 * Get the context from user stack 386 */ 387 if (copyin(scp, &context, sizeof(*scp))) 388 return (EFAULT); 389 390 /* 391 * Make sure, fpu is in sync 392 */ 393 save_fpu_proc(curproc); 394 395 /* 396 * Restore register context. 397 */ 398 if (copyin((caddr_t)context.lregs, &sregs, sizeof(sregs))) 399 return (EFAULT); 400 lregs = (struct linux_pt_regs *)&sregs.lgp_regs; 401 402 tf = trapframe(p); 403 #ifdef DEBUG_LINUX 404 printf("linux_sys_sigreturn: trapframe=0x%lx scp=0x%lx\n", 405 (unsigned long)tf, (unsigned long)scp); 406 #endif 407 408 if ((lregs->lmsr & PSL_USERSTATIC) != (tf->srr1 & PSL_USERSTATIC)) 409 return (EINVAL); 410 411 for (i = 0; i < 32; i++) 412 tf->fixreg[i] = lregs->lgpr[i]; 413 tf->lr = lregs->llink; 414 tf->cr = lregs->lccr; 415 tf->xer = lregs->lxer; 416 tf->ctr = lregs->lctr; 417 tf->srr0 = lregs->lnip; 418 tf->srr1 = lregs->lmsr; 419 420 memcpy(curpcb->pcb_fpu.fpr, (caddr_t)&sregs.lfp_regs, 421 sizeof(curpcb->pcb_fpu.fpr)); 422 423 /* 424 * Restore signal stack. 425 * 426 * XXX cannot find the onstack information in Linux sig context. 427 * Is signal stack really supported on Linux? 428 */ 429 #if 0 430 if (sc.sc_onstack & SS_ONSTACK) 431 p->p_sigctx.ps_sigstk.ss_flags |= SS_ONSTACK; 432 else 433 #endif 434 p->p_sigctx.ps_sigstk.ss_flags &= ~SS_ONSTACK; 435 436 /* Restore signal mask. */ 437 linux_old_extra_to_native_sigset(&mask, &context.lmask, 438 &context._unused[3]); 439 (void) sigprocmask1(p, SIG_SETMASK, &mask, 0); 440 441 return (EJUSTRETURN); 442 } 443 444 445 #if 0 446 int 447 linux_sys_modify_ldt(p, v, retval) 448 struct proc *p; 449 void *v; 450 register_t *retval; 451 { 452 /* 453 * This syscall is not implemented in Linux/PowerPC: we should not 454 * be here 455 */ 456 #ifdef DEBUG_LINUX 457 printf("linux_sys_modify_ldt: should not be here.\n"); 458 #endif 459 return 0; 460 } 461 #endif 462 463 /* 464 * major device numbers remapping 465 */ 466 dev_t 467 linux_fakedev(dev, raw) 468 dev_t dev; 469 int raw; 470 { 471 /* XXX write me */ 472 return dev; 473 } 474 475 /* 476 * We come here in a last attempt to satisfy a Linux ioctl() call 477 */ 478 int 479 linux_machdepioctl(p, v, retval) 480 struct proc *p; 481 void *v; 482 register_t *retval; 483 { 484 struct linux_sys_ioctl_args /* { 485 syscallarg(int) fd; 486 syscallarg(u_long) com; 487 syscallarg(caddr_t) data; 488 } */ *uap = v; 489 struct sys_ioctl_args bia; 490 u_long com; 491 492 SCARG(&bia, fd) = SCARG(uap, fd); 493 SCARG(&bia, data) = SCARG(uap, data); 494 com = SCARG(uap, com); 495 496 switch (com) { 497 default: 498 printf("linux_machdepioctl: invalid ioctl %08lx\n", com); 499 return EINVAL; 500 } 501 SCARG(&bia, com) = com; 502 return sys_ioctl(p, &bia, retval); 503 } 504 #if 0 505 /* 506 * Set I/O permissions for a process. Just set the maximum level 507 * right away (ignoring the argument), otherwise we would have 508 * to rely on I/O permission maps, which are not implemented. 509 */ 510 int 511 linux_sys_iopl(p, v, retval) 512 struct proc *p; 513 void *v; 514 register_t *retval; 515 { 516 /* 517 * This syscall is not implemented in Linux/PowerPC: we should not be here 518 */ 519 #ifdef DEBUG_LINUX 520 printf("linux_sys_iopl: should not be here.\n"); 521 #endif 522 return 0; 523 } 524 #endif 525 526 /* 527 * See above. If a root process tries to set access to an I/O port, 528 * just let it have the whole range. 529 */ 530 int 531 linux_sys_ioperm(p, v, retval) 532 struct proc *p; 533 void *v; 534 register_t *retval; 535 { 536 /* 537 * This syscall is not implemented in Linux/PowerPC: we should not be here 538 */ 539 #ifdef DEBUG_LINUX 540 printf("linux_sys_ioperm: should not be here.\n"); 541 #endif 542 return 0; 543 } 544 545 /* 546 * wrapper linux_sys_new_uname() -> linux_sys_uname() 547 */ 548 int 549 linux_sys_new_uname(p, v, retval) 550 struct proc *p; 551 void *v; 552 register_t *retval; 553 { 554 return linux_sys_uname(p, v, retval); 555 } 556 557 /* 558 * wrapper linux_sys_new_select() -> linux_sys_select() 559 */ 560 int 561 linux_sys_new_select(p, v, retval) 562 struct proc *p; 563 void *v; 564 register_t *retval; 565 { 566 return linux_sys_select(p, v, retval); 567 } 568