1 /* $NetBSD: linux_machdep.c,v 1.46 2010/07/07 01:30:33 chs Exp $ */ 2 3 /*- 4 * Copyright (c) 1998 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Eric Haszlakiewicz. 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 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 * 31 * Based on sys/arch/i386/i386/linux_machdep.c: 32 * linux_machdep.c,v 1.42 1998/09/11 12:50:06 mycroft Exp 33 * written by Frank van der Linden 34 * 35 */ 36 37 #include <sys/cdefs.h> 38 __KERNEL_RCSID(0, "$NetBSD: linux_machdep.c,v 1.46 2010/07/07 01:30:33 chs Exp $"); 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/signalvar.h> 43 #include <sys/kernel.h> 44 #include <sys/proc.h> 45 #include <sys/buf.h> 46 #include <sys/reboot.h> 47 #include <sys/conf.h> 48 #include <sys/exec.h> 49 #include <sys/file.h> 50 #include <sys/callout.h> 51 #include <sys/malloc.h> 52 #include <sys/mbuf.h> 53 #include <sys/msgbuf.h> 54 #include <sys/mount.h> 55 #include <sys/vnode.h> 56 #include <sys/device.h> 57 #include <sys/syscallargs.h> 58 #include <sys/filedesc.h> 59 #include <sys/exec_elf.h> 60 #include <sys/ioctl.h> 61 #include <sys/kauth.h> 62 63 #include <uvm/uvm_extern.h> 64 65 #include <compat/linux/common/linux_types.h> 66 #include <compat/linux/common/linux_signal.h> 67 #include <compat/linux/common/linux_siginfo.h> 68 #include <compat/linux/common/linux_util.h> 69 #include <compat/linux/common/linux_ioctl.h> 70 #include <compat/linux/common/linux_exec.h> 71 #include <compat/linux/common/linux_machdep.h> 72 #include <compat/linux/common/linux_emuldata.h> 73 74 #include <compat/linux/linux_syscallargs.h> 75 76 #include <machine/alpha.h> 77 #include <machine/reg.h> 78 79 #if defined(_KERNEL_OPT) 80 #include "wsdisplay.h" 81 #endif 82 #if (NWSDISPLAY >0) 83 #include <dev/wscons/wsdisplay_usl_io.h> 84 #endif 85 #ifdef DEBUG 86 #include <machine/sigdebug.h> 87 #endif 88 89 /* 90 * Deal with some alpha-specific things in the Linux emulation code. 91 */ 92 93 void 94 linux_setregs(struct lwp *l, struct exec_package *epp, vaddr_t stack) 95 { 96 #ifdef DEBUG 97 struct trapframe *tfp = l->l_md.md_tf; 98 #endif 99 100 setregs(l, epp, stack); 101 #ifdef DEBUG 102 /* 103 * Linux has registers set to zero on entry; for DEBUG kernels 104 * the alpha setregs() fills registers with 0xbabefacedeadbeef. 105 */ 106 memset(tfp->tf_regs, 0, FRAME_SIZE * sizeof tfp->tf_regs[0]); 107 #endif 108 } 109 110 void 111 setup_linux_rt_sigframe(struct trapframe *tf, int sig, const sigset_t *mask) 112 { 113 struct lwp *l = curlwp; 114 struct proc *p = l->l_proc; 115 struct linux_rt_sigframe *sfp, sigframe; 116 int onstack, error; 117 int fsize, rndfsize; 118 extern char linux_rt_sigcode[], linux_rt_esigcode[]; 119 120 /* Do we need to jump onto the signal stack? */ 121 onstack = (l->l_sigstk.ss_flags & (SS_DISABLE | SS_ONSTACK)) == 0 && 122 (SIGACTION(p, sig).sa_flags & SA_ONSTACK) != 0; 123 124 /* Allocate space for the signal handler context. */ 125 fsize = sizeof(struct linux_rt_sigframe); 126 rndfsize = ((fsize + 15) / 16) * 16; 127 128 if (onstack) 129 sfp = (struct linux_rt_sigframe *) 130 ((char *)l->l_sigstk.ss_sp + l->l_sigstk.ss_size); 131 else 132 sfp = (struct linux_rt_sigframe *)(alpha_pal_rdusp()); 133 sfp = (struct linux_rt_sigframe *)((char *)sfp - rndfsize); 134 135 #ifdef DEBUG 136 if ((sigdebug & SDB_KSTACK) && (p->p_pid == sigpid)) 137 printf("linux_sendsig(%d): sig %d ssp %p usp %p\n", p->p_pid, 138 sig, &onstack, sfp); 139 #endif /* DEBUG */ 140 141 /* 142 * Build the signal context to be used by sigreturn. 143 */ 144 memset(&sigframe.uc, 0, sizeof(struct linux_ucontext)); 145 sigframe.uc.uc_mcontext.sc_onstack = onstack; 146 147 /* Setup potentially partial signal mask in sc_mask. */ 148 /* But get all of it in uc_sigmask */ 149 native_to_linux_old_sigset(&sigframe.uc.uc_mcontext.sc_mask, mask); 150 native_to_linux_sigset(&sigframe.uc.uc_sigmask, mask); 151 152 sigframe.uc.uc_mcontext.sc_pc = tf->tf_regs[FRAME_PC]; 153 sigframe.uc.uc_mcontext.sc_ps = ALPHA_PSL_USERMODE; 154 frametoreg(tf, (struct reg *)sigframe.uc.uc_mcontext.sc_regs); 155 sigframe.uc.uc_mcontext.sc_regs[R_SP] = alpha_pal_rdusp(); 156 157 alpha_enable_fp(l, 1); 158 sigframe.uc.uc_mcontext.sc_fpcr = alpha_read_fpcr(); 159 sigframe.uc.uc_mcontext.sc_fp_control = alpha_read_fp_c(l); 160 alpha_pal_wrfen(0); 161 162 sigframe.uc.uc_mcontext.sc_traparg_a0 = tf->tf_regs[FRAME_A0]; 163 sigframe.uc.uc_mcontext.sc_traparg_a1 = tf->tf_regs[FRAME_A1]; 164 sigframe.uc.uc_mcontext.sc_traparg_a2 = tf->tf_regs[FRAME_A2]; 165 166 /* 167 * XXX XAX Create bogus siginfo data. This can't really 168 * XXX be fixed until NetBSD has realtime signals. 169 * XXX Or we do the emuldata thing. 170 * XXX -erh 171 */ 172 memset(&sigframe.info, 0, sizeof(struct linux_siginfo)); 173 sigframe.info.lsi_signo = sig; 174 sigframe.info.lsi_code = LINUX_SI_USER; 175 sigframe.info.lsi_pid = p->p_pid; 176 sigframe.info.lsi_uid = kauth_cred_geteuid(l->l_cred); /* Use real uid here? */ 177 178 sendsig_reset(l, sig); 179 mutex_exit(p->p_lock); 180 error = copyout((void *)&sigframe, (void *)sfp, fsize); 181 mutex_enter(p->p_lock); 182 183 if (error != 0) { 184 #ifdef DEBUG 185 if ((sigdebug & SDB_KSTACK) && p->p_pid == sigpid) 186 printf("sendsig(%d): copyout failed on sig %d\n", 187 p->p_pid, sig); 188 #endif 189 /* 190 * Process has trashed its stack; give it an illegal 191 * instruction to halt it in its tracks. 192 */ 193 sigexit(l, SIGILL); 194 /* NOTREACHED */ 195 } 196 197 /* Pass pointers to siginfo and ucontext in the regs */ 198 tf->tf_regs[FRAME_A1] = (unsigned long)&sfp->info; 199 tf->tf_regs[FRAME_A2] = (unsigned long)&sfp->uc; 200 201 /* Address of trampoline code. End up at this PC after mi_switch */ 202 tf->tf_regs[FRAME_PC] = 203 (u_int64_t)(p->p_psstr - (linux_rt_esigcode - linux_rt_sigcode)); 204 205 /* Adjust the stack */ 206 alpha_pal_wrusp((unsigned long)sfp); 207 208 /* Remember that we're now on the signal stack. */ 209 if (onstack) 210 l->l_sigstk.ss_flags |= SS_ONSTACK; 211 } 212 213 void setup_linux_sigframe(tf, sig, mask) 214 struct trapframe *tf; 215 int sig; 216 const sigset_t *mask; 217 { 218 struct lwp *l = curlwp; 219 struct proc *p = l->l_proc; 220 struct linux_sigframe *sfp, sigframe; 221 int onstack, error; 222 int fsize, rndfsize; 223 extern char linux_sigcode[], linux_esigcode[]; 224 225 /* Do we need to jump onto the signal stack? */ 226 onstack = (l->l_sigstk.ss_flags & (SS_DISABLE | SS_ONSTACK)) == 0 && 227 (SIGACTION(p, sig).sa_flags & SA_ONSTACK) != 0; 228 229 /* Allocate space for the signal handler context. */ 230 fsize = sizeof(struct linux_sigframe); 231 rndfsize = ((fsize + 15) / 16) * 16; 232 233 if (onstack) 234 sfp = (struct linux_sigframe *) 235 ((char *)l->l_sigstk.ss_sp + l->l_sigstk.ss_size); 236 else 237 sfp = (struct linux_sigframe *)(alpha_pal_rdusp()); 238 sfp = (struct linux_sigframe *)((char *)sfp - rndfsize); 239 240 #ifdef DEBUG 241 if ((sigdebug & SDB_KSTACK) && (p->p_pid == sigpid)) 242 printf("linux_sendsig(%d): sig %d ssp %p usp %p\n", p->p_pid, 243 sig, &onstack, sfp); 244 #endif /* DEBUG */ 245 246 /* 247 * Build the signal context to be used by sigreturn. 248 */ 249 memset(&sigframe.sf_sc, 0, sizeof(struct linux_sigcontext)); 250 sigframe.sf_sc.sc_onstack = onstack; 251 native_to_linux_old_sigset(&sigframe.sf_sc.sc_mask, mask); 252 sigframe.sf_sc.sc_pc = tf->tf_regs[FRAME_PC]; 253 sigframe.sf_sc.sc_ps = ALPHA_PSL_USERMODE; 254 frametoreg(tf, (struct reg *)sigframe.sf_sc.sc_regs); 255 sigframe.sf_sc.sc_regs[R_SP] = alpha_pal_rdusp(); 256 257 if (l == fpcurlwp) { 258 struct pcb *pcb = lwp_getpcb(l); 259 260 alpha_pal_wrfen(1); 261 savefpstate(&pcb->pcb_fp); 262 alpha_pal_wrfen(0); 263 sigframe.sf_sc.sc_fpcr = pcb->pcb_fp.fpr_cr; 264 fpcurlwp = NULL; 265 } 266 /* XXX ownedfp ? etc...? */ 267 268 sigframe.sf_sc.sc_traparg_a0 = tf->tf_regs[FRAME_A0]; 269 sigframe.sf_sc.sc_traparg_a1 = tf->tf_regs[FRAME_A1]; 270 sigframe.sf_sc.sc_traparg_a2 = tf->tf_regs[FRAME_A2]; 271 272 sendsig_reset(l, sig); 273 mutex_exit(p->p_lock); 274 error = copyout((void *)&sigframe, (void *)sfp, fsize); 275 mutex_enter(p->p_lock); 276 277 if (error != 0) { 278 #ifdef DEBUG 279 if ((sigdebug & SDB_KSTACK) && p->p_pid == sigpid) 280 printf("sendsig(%d): copyout failed on sig %d\n", 281 p->p_pid, sig); 282 #endif 283 /* 284 * Process has trashed its stack; give it an illegal 285 * instruction to halt it in its tracks. 286 */ 287 sigexit(l, SIGILL); 288 /* NOTREACHED */ 289 } 290 291 /* Pass pointers to sigcontext in the regs */ 292 tf->tf_regs[FRAME_A1] = 0; 293 tf->tf_regs[FRAME_A2] = (unsigned long)&sfp->sf_sc; 294 295 /* Address of trampoline code. End up at this PC after mi_switch */ 296 tf->tf_regs[FRAME_PC] = 297 (u_int64_t)(p->p_psstr - (linux_esigcode - linux_sigcode)); 298 299 /* Adjust the stack */ 300 alpha_pal_wrusp((unsigned long)sfp); 301 302 /* Remember that we're now on the signal stack. */ 303 if (onstack) 304 l->l_sigstk.ss_flags |= SS_ONSTACK; 305 } 306 307 /* 308 * Send an interrupt to process. 309 * 310 * Stack is set up to allow sigcode stored 311 * in u. to call routine, followed by kcall 312 * to sigreturn routine below. After sigreturn 313 * resets the signal mask, the stack, and the 314 * frame pointer, it returns to the user 315 * specified pc, psl. 316 */ 317 void 318 linux_sendsig(const ksiginfo_t *ksi, const sigset_t *mask) 319 { 320 struct lwp *l = curlwp; 321 struct proc *p = l->l_proc; 322 struct trapframe *tf = l->l_md.md_tf; 323 const int sig = ksi->ksi_signo; 324 sig_t catcher = SIGACTION(p, sig).sa_handler; 325 #ifdef notyet 326 struct linux_emuldata *edp; 327 328 /* Setup the signal frame (and part of the trapframe) */ 329 /*OLD: if (p->p_sigacts->ps_siginfo & sigmask(sig))*/ 330 /* XXX XAX this is broken now. need someplace to store what 331 XXX XAX kind of signal handler a signal has.*/ 332 #if 0 333 edp = (struct linux_emuldata *)p->p_emuldata; 334 #else 335 edp = 0; 336 #endif 337 if (edp && sigismember(&edp->ps_siginfo, sig)) 338 setup_linux_rt_sigframe(tf, sig, mask); 339 else 340 #endif /* notyet */ 341 setup_linux_sigframe(tf, sig, mask); 342 343 /* Signal handler for trampoline code */ 344 tf->tf_regs[FRAME_T12] = (u_int64_t)catcher; 345 tf->tf_regs[FRAME_A0] = native_to_linux_signo[sig]; 346 347 /* 348 * Linux has a custom restorer option. To support it we would 349 * need to store an array of restorers and a sigcode block 350 * which knew to use it. Doesn't seem worth the trouble. 351 * -erh 352 */ 353 354 #ifdef DEBUG 355 if (sigdebug & SDB_FOLLOW) 356 printf("sendsig(%d): pc %lx, catcher %lx\n", l->l_proc->p_pid, 357 tf->tf_regs[FRAME_PC], tf->tf_regs[FRAME_A3]); 358 if ((sigdebug & SDB_KSTACK) && l->l_proc->p_pid == sigpid) 359 printf("sendsig(%d): sig %d returns\n", l->l_proc->p_pid, sig); 360 #endif 361 } 362 363 /* 364 * System call to cleanup state after a signal 365 * has been taken. Reset signal mask and 366 * stack state from context left by sendsig (above). 367 * Return to previous pc as specified by context 368 * left by sendsig. 369 * Linux real-time signals use a different sigframe, 370 * but the sigcontext is the same. 371 */ 372 373 int 374 linux_restore_sigcontext(struct lwp *l, struct linux_sigcontext context, 375 sigset_t *mask) 376 { 377 struct proc *p = l->l_proc; 378 struct pcb *pcb; 379 380 /* 381 * Linux doesn't (yet) have alternate signal stacks. 382 * However, the OSF/1 sigcontext which they use has 383 * an onstack member. This could be needed in the future. 384 */ 385 mutex_enter(p->p_lock); 386 if (context.sc_onstack & LINUX_SA_ONSTACK) 387 l->l_sigstk.ss_flags |= SS_ONSTACK; 388 else 389 l->l_sigstk.ss_flags &= ~SS_ONSTACK; 390 391 /* Reset the signal mask */ 392 (void) sigprocmask1(l, SIG_SETMASK, mask, 0); 393 mutex_exit(p->p_lock); 394 395 /* 396 * Check for security violations. 397 * Linux doesn't allow any changes to the PSL. 398 */ 399 if (context.sc_ps != ALPHA_PSL_USERMODE) 400 return(EINVAL); 401 402 l->l_md.md_tf->tf_regs[FRAME_PC] = context.sc_pc; 403 l->l_md.md_tf->tf_regs[FRAME_PS] = context.sc_ps; 404 405 regtoframe((struct reg *)context.sc_regs, l->l_md.md_tf); 406 alpha_pal_wrusp(context.sc_regs[R_SP]); 407 408 if (l == fpcurlwp) 409 fpcurlwp = NULL; 410 411 /* Restore fp regs and fpr_cr */ 412 pcb = lwp_getpcb(l); 413 memcpy(&pcb->pcb_fp, (struct fpreg *)context.sc_fpregs, 414 sizeof(struct fpreg)); 415 /* XXX sc_ownedfp ? */ 416 /* XXX sc_fp_control ? */ 417 418 #ifdef DEBUG 419 if (sigdebug & SDB_FOLLOW) 420 printf("linux_rt_sigreturn(%d): returns\n", p->p_pid); 421 #endif 422 return (EJUSTRETURN); 423 } 424 425 int 426 linux_sys_rt_sigreturn(struct lwp *l, const struct linux_sys_rt_sigreturn_args *uap, register_t *retval) 427 { 428 /* { 429 syscallarg(struct linux_rt_sigframe *) sfp; 430 } */ 431 struct linux_rt_sigframe *sfp, sigframe; 432 sigset_t mask; 433 434 /* 435 * The trampoline code hands us the context. 436 * It is unsafe to keep track of it ourselves, in the event that a 437 * program jumps out of a signal handler. 438 */ 439 440 sfp = SCARG(uap, sfp); 441 442 if (ALIGN(sfp) != (u_int64_t)sfp) 443 return(EINVAL); 444 445 /* 446 * Fetch the frame structure. 447 */ 448 if (copyin((void *)sfp, &sigframe, 449 sizeof(struct linux_rt_sigframe)) != 0) 450 return (EFAULT); 451 452 /* Grab the signal mask */ 453 linux_to_native_sigset(&mask, &sigframe.uc.uc_sigmask); 454 455 return(linux_restore_sigcontext(l, sigframe.uc.uc_mcontext, &mask)); 456 } 457 458 459 int 460 linux_sys_sigreturn(struct lwp *l, const struct linux_sys_sigreturn_args *uap, register_t *retval) 461 { 462 /* { 463 syscallarg(struct linux_sigframe *) sfp; 464 } */ 465 struct linux_sigframe *sfp, frame; 466 sigset_t mask; 467 468 /* 469 * The trampoline code hands us the context. 470 * It is unsafe to keep track of it ourselves, in the event that a 471 * program jumps out of a signal handler. 472 */ 473 474 sfp = SCARG(uap, sfp); 475 if (ALIGN(sfp) != (u_int64_t)sfp) 476 return(EINVAL); 477 478 /* 479 * Fetch the frame structure. 480 */ 481 if (copyin((void *)sfp, &frame, sizeof(struct linux_sigframe)) != 0) 482 return(EFAULT); 483 484 /* Grab the signal mask. */ 485 /* XXX use frame.extramask */ 486 linux_old_to_native_sigset(&mask, frame.sf_sc.sc_mask); 487 488 return(linux_restore_sigcontext(l, frame.sf_sc, &mask)); 489 } 490 491 /* 492 * We come here in a last attempt to satisfy a Linux ioctl() call 493 */ 494 /* XXX XAX update this, add maps, etc... */ 495 int 496 linux_machdepioctl(struct lwp *l, const struct linux_sys_ioctl_args *uap, register_t *retval) 497 { 498 /* { 499 syscallarg(int) fd; 500 syscallarg(u_long) com; 501 syscallarg(void *) data; 502 } */ 503 struct sys_ioctl_args bia; 504 u_long com; 505 506 SCARG(&bia, fd) = SCARG(uap, fd); 507 SCARG(&bia, data) = SCARG(uap, data); 508 com = SCARG(uap, com); 509 510 switch (com) { 511 default: 512 printf("linux_machdepioctl: invalid ioctl %08lx\n", com); 513 return EINVAL; 514 } 515 SCARG(&bia, com) = com; 516 return sys_ioctl(l, &bia, retval); 517 } 518 519 /* XXX XAX fix this */ 520 dev_t 521 linux_fakedev(dev_t dev, int raw) 522 { 523 return dev; 524 } 525 526 int 527 linux_usertrap(struct lwp *l, vaddr_t trapaddr, void *arg) 528 { 529 return 0; 530 } 531