1/* $NetBSD: linux_machdep.c,v 1.48 2011/06/09 13:41:40 matt 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 1.48 2011/06/09 13:41:40 matt 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 93void 94linux_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 110void 111setup_linux_rt_sigframe(struct trapframe *tf, const ksiginfo_t *ksi, 112 const sigset_t *mask) 113{ 114 struct lwp *l = curlwp; 115 struct proc *p = l->l_proc; 116 struct linux_rt_sigframe *sfp, sigframe; 117 int onstack, error; 118 int fsize, rndfsize; 119 int sig = ksi->ksi_signo; 120 extern char linux_rt_sigcode[], linux_rt_esigcode[]; 121 122 /* Do we need to jump onto the signal stack? */ 123 onstack = (l->l_sigstk.ss_flags & (SS_DISABLE | SS_ONSTACK)) == 0 && 124 (SIGACTION(p, sig).sa_flags & SA_ONSTACK) != 0; 125 126 /* Allocate space for the signal handler context. */ 127 fsize = sizeof(struct linux_rt_sigframe); 128 rndfsize = ((fsize + 15) / 16) * 16; 129 130 if (onstack) 131 sfp = (struct linux_rt_sigframe *) 132 ((char *)l->l_sigstk.ss_sp + l->l_sigstk.ss_size); 133 else 134 sfp = (struct linux_rt_sigframe *)(alpha_pal_rdusp()); 135 sfp = (struct linux_rt_sigframe *)((char *)sfp - rndfsize); 136 137#ifdef DEBUG 138 if ((sigdebug & SDB_KSTACK) && (p->p_pid == sigpid)) 139 printf("linux_sendsig(%d): sig %d ssp %p usp %p\n", p->p_pid, 140 sig, &onstack, sfp); 141#endif /* DEBUG */ 142 143 /* 144 * Build the signal context to be used by sigreturn. 145 */ 146 memset(&sigframe.uc, 0, sizeof(struct linux_ucontext)); 147 sigframe.uc.uc_mcontext.sc_onstack = onstack; 148 149 /* Setup potentially partial signal mask in sc_mask. */ 150 /* But get all of it in uc_sigmask */ 151 native_to_linux_old_sigset(&sigframe.uc.uc_mcontext.sc_mask, mask); 152 native_to_linux_sigset(&sigframe.uc.uc_sigmask, mask); 153 154 sigframe.uc.uc_mcontext.sc_pc = tf->tf_regs[FRAME_PC]; 155 sigframe.uc.uc_mcontext.sc_ps = ALPHA_PSL_USERMODE; 156 frametoreg(tf, (struct reg *)sigframe.uc.uc_mcontext.sc_regs); 157 sigframe.uc.uc_mcontext.sc_regs[R_SP] = alpha_pal_rdusp(); 158 159 fpu_load(); 160 alpha_pal_wrfen(1); 161 sigframe.uc.uc_mcontext.sc_fpcr = alpha_read_fpcr(); 162 sigframe.uc.uc_mcontext.sc_fp_control = alpha_read_fp_c(l); 163 alpha_pal_wrfen(0); 164 165 sigframe.uc.uc_mcontext.sc_traparg_a0 = tf->tf_regs[FRAME_A0]; 166 sigframe.uc.uc_mcontext.sc_traparg_a1 = tf->tf_regs[FRAME_A1]; 167 sigframe.uc.uc_mcontext.sc_traparg_a2 = tf->tf_regs[FRAME_A2]; 168 native_to_linux_siginfo(&sigframe.info, &ksi->ksi_info); 169 sendsig_reset(l, sig); 170 mutex_exit(p->p_lock); 171 error = copyout((void *)&sigframe, (void *)sfp, fsize); 172 mutex_enter(p->p_lock); 173 174 if (error != 0) { 175#ifdef DEBUG 176 if ((sigdebug & SDB_KSTACK) && p->p_pid == sigpid) 177 printf("sendsig(%d): copyout failed on sig %d\n", 178 p->p_pid, sig); 179#endif 180 /* 181 * Process has trashed its stack; give it an illegal 182 * instruction to halt it in its tracks. 183 */ 184 sigexit(l, SIGILL); 185 /* NOTREACHED */ 186 } 187 188 /* Pass pointers to siginfo and ucontext in the regs */ 189 tf->tf_regs[FRAME_A1] = (unsigned long)&sfp->info; 190 tf->tf_regs[FRAME_A2] = (unsigned long)&sfp->uc; 191 192 /* Address of trampoline code. End up at this PC after mi_switch */ 193 tf->tf_regs[FRAME_PC] = 194 (u_int64_t)(p->p_psstrp - (linux_rt_esigcode - linux_rt_sigcode)); 195 196 /* Adjust the stack */ 197 alpha_pal_wrusp((unsigned long)sfp); 198 199 /* Remember that we're now on the signal stack. */ 200 if (onstack) 201 l->l_sigstk.ss_flags |= SS_ONSTACK; 202} 203 204void setup_linux_sigframe(struct trapframe *tf, const ksiginfo_t *ksi, 205 const sigset_t *mask) 206{ 207 struct lwp *l = curlwp; 208 struct proc *p = l->l_proc; 209 struct linux_sigframe *sfp, sigframe; 210 int onstack, error; 211 int fsize, rndfsize; 212 int sig = ksi->ksi_signo; 213 extern char linux_sigcode[], linux_esigcode[]; 214 215 /* Do we need to jump onto the signal stack? */ 216 onstack = (l->l_sigstk.ss_flags & (SS_DISABLE | SS_ONSTACK)) == 0 && 217 (SIGACTION(p, sig).sa_flags & SA_ONSTACK) != 0; 218 219 /* Allocate space for the signal handler context. */ 220 fsize = sizeof(struct linux_sigframe); 221 rndfsize = ((fsize + 15) / 16) * 16; 222 223 if (onstack) 224 sfp = (struct linux_sigframe *) 225 ((char *)l->l_sigstk.ss_sp + l->l_sigstk.ss_size); 226 else 227 sfp = (struct linux_sigframe *)(alpha_pal_rdusp()); 228 sfp = (struct linux_sigframe *)((char *)sfp - rndfsize); 229 230#ifdef DEBUG 231 if ((sigdebug & SDB_KSTACK) && (p->p_pid == sigpid)) 232 printf("linux_sendsig(%d): sig %d ssp %p usp %p\n", p->p_pid, 233 sig, &onstack, sfp); 234#endif /* DEBUG */ 235 236 /* 237 * Build the signal context to be used by sigreturn. 238 */ 239 memset(&sigframe.sf_sc, 0, sizeof(struct linux_sigcontext)); 240 sigframe.sf_sc.sc_onstack = onstack; 241 native_to_linux_old_sigset(&sigframe.sf_sc.sc_mask, mask); 242 sigframe.sf_sc.sc_pc = tf->tf_regs[FRAME_PC]; 243 sigframe.sf_sc.sc_ps = ALPHA_PSL_USERMODE; 244 frametoreg(tf, (struct reg *)sigframe.sf_sc.sc_regs); 245 sigframe.sf_sc.sc_regs[R_SP] = alpha_pal_rdusp(); 246 247 if (l == fpcurlwp) { 248 struct pcb *pcb = lwp_getpcb(l); 249 250 alpha_pal_wrfen(1); 251 savefpstate(&pcb->pcb_fp); 252 alpha_pal_wrfen(0); 253 sigframe.sf_sc.sc_fpcr = pcb->pcb_fp.fpr_cr; 254 fpcurlwp = NULL; 255 } 256 /* XXX ownedfp ? etc...? */ 257 258 sigframe.sf_sc.sc_traparg_a0 = tf->tf_regs[FRAME_A0]; 259 sigframe.sf_sc.sc_traparg_a1 = tf->tf_regs[FRAME_A1]; 260 sigframe.sf_sc.sc_traparg_a2 = tf->tf_regs[FRAME_A2]; 261 262 sendsig_reset(l, sig); 263 mutex_exit(p->p_lock); 264 error = copyout((void *)&sigframe, (void *)sfp, fsize); 265 mutex_enter(p->p_lock); 266 267 if (error != 0) { 268#ifdef DEBUG 269 if ((sigdebug & SDB_KSTACK) && p->p_pid == sigpid) 270 printf("sendsig(%d): copyout failed on sig %d\n", 271 p->p_pid, sig); 272#endif 273 /* 274 * Process has trashed its stack; give it an illegal 275 * instruction to halt it in its tracks. 276 */ 277 sigexit(l, SIGILL); 278 /* NOTREACHED */ 279 } 280 281 /* Pass pointers to sigcontext in the regs */ 282 tf->tf_regs[FRAME_A1] = 0; 283 tf->tf_regs[FRAME_A2] = (unsigned long)&sfp->sf_sc; 284 285 /* Address of trampoline code. End up at this PC after mi_switch */ 286 tf->tf_regs[FRAME_PC] = 287 (u_int64_t)(p->p_psstrp - (linux_esigcode - linux_sigcode)); 288 289 /* Adjust the stack */ 290 alpha_pal_wrusp((unsigned long)sfp); 291 292 /* Remember that we're now on the signal stack. */ 293 if (onstack) 294 l->l_sigstk.ss_flags |= SS_ONSTACK; 295} 296 297/* 298 * Send an interrupt to process. 299 * 300 * Stack is set up to allow sigcode stored 301 * in u. to call routine, followed by kcall 302 * to sigreturn routine below. After sigreturn 303 * resets the signal mask, the stack, and the 304 * frame pointer, it returns to the user 305 * specified pc, psl. 306 */ 307void 308linux_sendsig(const ksiginfo_t *ksi, const sigset_t *mask) 309{ 310 struct lwp *l = curlwp; 311 struct proc *p = l->l_proc; 312 struct trapframe *tf = l->l_md.md_tf; 313 const int sig = ksi->ksi_signo; 314 sig_t catcher = SIGACTION(p, sig).sa_handler; 315#ifdef notyet 316 struct linux_emuldata *edp; 317 318 /* Setup the signal frame (and part of the trapframe) */ 319 /*OLD: if (p->p_sigacts->ps_siginfo & sigmask(sig))*/ 320/* XXX XAX this is broken now. need someplace to store what 321 XXX XAX kind of signal handler a signal has.*/ 322#if 0 323 edp = (struct linux_emuldata *)p->p_emuldata; 324#else 325 edp = 0; 326#endif 327 if (edp && sigismember(&edp->ps_siginfo, sig)) 328 setup_linux_rt_sigframe(tf, ksi, mask); 329 else 330#endif /* notyet */ 331 setup_linux_sigframe(tf, ksi, mask); 332 333 /* Signal handler for trampoline code */ 334 tf->tf_regs[FRAME_T12] = (u_int64_t)catcher; 335 tf->tf_regs[FRAME_A0] = native_to_linux_signo[sig]; 336 337 /* 338 * Linux has a custom restorer option. To support it we would 339 * need to store an array of restorers and a sigcode block 340 * which knew to use it. Doesn't seem worth the trouble. 341 * -erh 342 */ 343 344#ifdef DEBUG 345 if (sigdebug & SDB_FOLLOW) 346 printf("sendsig(%d): pc %lx, catcher %lx\n", l->l_proc->p_pid, 347 tf->tf_regs[FRAME_PC], tf->tf_regs[FRAME_A3]); 348 if ((sigdebug & SDB_KSTACK) && l->l_proc->p_pid == sigpid) 349 printf("sendsig(%d): sig %d returns\n", l->l_proc->p_pid, sig); 350#endif 351} 352 353/* 354 * System call to cleanup state after a signal 355 * has been taken. Reset signal mask and 356 * stack state from context left by sendsig (above). 357 * Return to previous pc as specified by context 358 * left by sendsig. 359 * Linux real-time signals use a different sigframe, 360 * but the sigcontext is the same. 361 */ 362 363int 364linux_restore_sigcontext(struct lwp *l, struct linux_sigcontext context, 365 sigset_t *mask) 366{ 367 struct proc *p = l->l_proc; 368 struct pcb *pcb; 369 370 /* 371 * Linux doesn't (yet) have alternate signal stacks. 372 * However, the OSF/1 sigcontext which they use has 373 * an onstack member. This could be needed in the future. 374 */ 375 mutex_enter(p->p_lock); 376 if (context.sc_onstack & LINUX_SA_ONSTACK) 377 l->l_sigstk.ss_flags |= SS_ONSTACK; 378 else 379 l->l_sigstk.ss_flags &= ~SS_ONSTACK; 380 381 /* Reset the signal mask */ 382 (void) sigprocmask1(l, SIG_SETMASK, mask, 0); 383 mutex_exit(p->p_lock); 384 385 /* 386 * Check for security violations. 387 * Linux doesn't allow any changes to the PSL. 388 */ 389 if (context.sc_ps != ALPHA_PSL_USERMODE) 390 return(EINVAL); 391 392 l->l_md.md_tf->tf_regs[FRAME_PC] = context.sc_pc; 393 l->l_md.md_tf->tf_regs[FRAME_PS] = context.sc_ps; 394 395 regtoframe((struct reg *)context.sc_regs, l->l_md.md_tf); 396 alpha_pal_wrusp(context.sc_regs[R_SP]); 397 398 if (l == fpcurlwp) 399 fpcurlwp = NULL; 400 401 /* Restore fp regs and fpr_cr */ 402 pcb = lwp_getpcb(l); 403 memcpy(&pcb->pcb_fp, (struct fpreg *)context.sc_fpregs, 404 sizeof(struct fpreg)); 405 /* XXX sc_ownedfp ? */ 406 /* XXX sc_fp_control ? */ 407 408#ifdef DEBUG 409 if (sigdebug & SDB_FOLLOW) 410 printf("linux_rt_sigreturn(%d): returns\n", p->p_pid); 411#endif 412 return (EJUSTRETURN); 413} 414 415int 416linux_sys_rt_sigreturn(struct lwp *l, const struct linux_sys_rt_sigreturn_args *uap, register_t *retval) 417{ 418 /* { 419 syscallarg(struct linux_rt_sigframe *) sfp; 420 } */ 421 struct linux_rt_sigframe *sfp, sigframe; 422 sigset_t mask; 423 424 /* 425 * The trampoline code hands us the context. 426 * It is unsafe to keep track of it ourselves, in the event that a 427 * program jumps out of a signal handler. 428 */ 429 430 sfp = SCARG(uap, sfp); 431 432 if (ALIGN(sfp) != (u_int64_t)sfp) 433 return(EINVAL); 434 435 /* 436 * Fetch the frame structure. 437 */ 438 if (copyin((void *)sfp, &sigframe, 439 sizeof(struct linux_rt_sigframe)) != 0) 440 return (EFAULT); 441 442 /* Grab the signal mask */ 443 linux_to_native_sigset(&mask, &sigframe.uc.uc_sigmask); 444 445 return(linux_restore_sigcontext(l, sigframe.uc.uc_mcontext, &mask)); 446} 447 448 449int 450linux_sys_sigreturn(struct lwp *l, const struct linux_sys_sigreturn_args *uap, register_t *retval) 451{ 452 /* { 453 syscallarg(struct linux_sigframe *) sfp; 454 } */ 455 struct linux_sigframe *sfp, frame; 456 sigset_t mask; 457 458 /* 459 * The trampoline code hands us the context. 460 * It is unsafe to keep track of it ourselves, in the event that a 461 * program jumps out of a signal handler. 462 */ 463 464 sfp = SCARG(uap, sfp); 465 if (ALIGN(sfp) != (u_int64_t)sfp) 466 return(EINVAL); 467 468 /* 469 * Fetch the frame structure. 470 */ 471 if (copyin((void *)sfp, &frame, sizeof(struct linux_sigframe)) != 0) 472 return(EFAULT); 473 474 /* Grab the signal mask. */ 475 /* XXX use frame.extramask */ 476 linux_old_to_native_sigset(&mask, frame.sf_sc.sc_mask); 477 478 return(linux_restore_sigcontext(l, frame.sf_sc, &mask)); 479} 480 481/* 482 * We come here in a last attempt to satisfy a Linux ioctl() call 483 */ 484/* XXX XAX update this, add maps, etc... */ 485int 486linux_machdepioctl(struct lwp *l, const struct linux_sys_ioctl_args *uap, register_t *retval) 487{ 488 /* { 489 syscallarg(int) fd; 490 syscallarg(u_long) com; 491 syscallarg(void *) data; 492 } */ 493 struct sys_ioctl_args bia; 494 u_long com; 495 496 SCARG(&bia, fd) = SCARG(uap, fd); 497 SCARG(&bia, data) = SCARG(uap, data); 498 com = SCARG(uap, com); 499 500 switch (com) { 501 default: 502 printf("linux_machdepioctl: invalid ioctl %08lx\n", com); 503 return EINVAL; 504 } 505 SCARG(&bia, com) = com; 506 return sys_ioctl(l, &bia, retval); 507} 508 509/* XXX XAX fix this */ 510dev_t 511linux_fakedev(dev_t dev, int raw) 512{ 513 return dev; 514} 515 516int 517linux_usertrap(struct lwp *l, vaddr_t trapaddr, void *arg) 518{ 519 return 0; 520} 521