1 /* $NetBSD: linux_machdep.c,v 1.40 2012/07/08 20:14:12 dsl Exp $ */ 2 3 /*- 4 * Copyright (c) 2005 Emmanuel Dreyfus, all rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 3. All advertising materials mentioning features or use of this software 15 * must display the following acknowledgement: 16 * This product includes software developed by Emmanuel Dreyfus 17 * 4. The name of the author may not be used to endorse or promote 18 * products derived from this software without specific prior written 19 * permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE THE AUTHOR AND CONTRIBUTORS ``AS IS'' 22 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, 23 * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 24 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS 25 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 28 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 29 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 30 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 31 * POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include <sys/cdefs.h> 35 36 __KERNEL_RCSID(0, "$NetBSD: linux_machdep.c,v 1.40 2012/07/08 20:14:12 dsl Exp $"); 37 38 #include <sys/param.h> 39 #include <sys/types.h> 40 #include <sys/systm.h> 41 #include <sys/signal.h> 42 #include <sys/exec.h> 43 #include <sys/proc.h> 44 #include <sys/ptrace.h> /* for process_read_fpregs() */ 45 #include <sys/ucontext.h> 46 #include <sys/conf.h> 47 48 #include <machine/reg.h> 49 #include <machine/pcb.h> 50 #include <machine/fpu.h> 51 #include <machine/mcontext.h> 52 #include <machine/specialreg.h> 53 #include <machine/vmparam.h> 54 #include <machine/cpufunc.h> 55 56 /* 57 * To see whether wscons is configured (for virtual console ioctl calls). 58 */ 59 #if defined(_KERNEL_OPT) 60 #include "wsdisplay.h" 61 #endif 62 #if (NWSDISPLAY > 0) 63 #include <dev/wscons/wsconsio.h> 64 #include <dev/wscons/wsdisplay_usl_io.h> 65 #endif 66 67 68 #include <compat/linux/common/linux_signal.h> 69 #include <compat/linux/common/linux_errno.h> 70 #include <compat/linux/common/linux_exec.h> 71 #include <compat/linux/common/linux_ioctl.h> 72 #include <compat/linux/common/linux_prctl.h> 73 #include <compat/linux/common/linux_machdep.h> 74 #include <compat/linux/common/linux_ipc.h> 75 #include <compat/linux/common/linux_sem.h> 76 #include <compat/linux/linux_syscall.h> 77 #include <compat/linux/linux_syscallargs.h> 78 79 static void linux_buildcontext(struct lwp *, void *, void *); 80 81 void 82 linux_setregs(struct lwp *l, struct exec_package *epp, vaddr_t stack) 83 { 84 struct pcb *pcb = lwp_getpcb(l); 85 struct trapframe *tf; 86 87 /* If we were using the FPU, forget about it. */ 88 if (pcb->pcb_fpcpu != NULL) 89 fpusave_lwp(l, 0); 90 91 l->l_md.md_flags &= ~MDL_USEDFPU; 92 pcb->pcb_flags = 0; 93 pcb->pcb_savefpu.fp_fxsave.fx_fcw = __NetBSD_NPXCW__; 94 pcb->pcb_savefpu.fp_fxsave.fx_mxcsr = __INITIAL_MXCSR__; 95 pcb->pcb_savefpu.fp_fxsave.fx_mxcsr_mask = __INITIAL_MXCSR_MASK__; 96 97 l->l_proc->p_flag &= ~PK_32; 98 99 tf = l->l_md.md_regs; 100 tf->tf_rax = 0; 101 tf->tf_rbx = 0; 102 tf->tf_rcx = epp->ep_entry; 103 tf->tf_rdx = 0; 104 tf->tf_rsi = 0; 105 tf->tf_rdi = 0; 106 tf->tf_rbp = 0; 107 tf->tf_rsp = stack; 108 tf->tf_r8 = 0; 109 tf->tf_r9 = 0; 110 tf->tf_r10 = 0; 111 tf->tf_r11 = 0; 112 tf->tf_r12 = 0; 113 tf->tf_r13 = 0; 114 tf->tf_r14 = 0; 115 tf->tf_r15 = 0; 116 tf->tf_rip = epp->ep_entry; 117 tf->tf_rflags = PSL_USERSET; 118 tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL); 119 tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL); 120 tf->tf_ds = 0; 121 tf->tf_es = 0; 122 cpu_fsgs_zero(l); 123 124 return; 125 } 126 127 void 128 linux_sendsig(const ksiginfo_t *ksi, const sigset_t *mask) 129 { 130 struct lwp *l = curlwp; 131 struct proc *p = l->l_proc; 132 struct pcb *pcb = lwp_getpcb(l); 133 struct sigacts *ps = p->p_sigacts; 134 int onstack, error; 135 int sig = ksi->ksi_signo; 136 struct linux_rt_sigframe *sfp, sigframe; 137 struct linux__fpstate *fpsp, fpstate; 138 struct fpreg fpregs; 139 struct trapframe *tf = l->l_md.md_regs; 140 sig_t catcher = SIGACTION(p, sig).sa_handler; 141 linux_sigset_t lmask; 142 char *sp; 143 144 /* Do we need to jump onto the signal stack? */ 145 onstack = 146 (l->l_sigstk.ss_flags & (SS_DISABLE | SS_ONSTACK)) == 0 && 147 (SIGACTION(p, sig).sa_flags & SA_ONSTACK) != 0; 148 149 /* Allocate space for the signal handler context. */ 150 if (onstack) 151 sp = ((char *)l->l_sigstk.ss_sp + 152 l->l_sigstk.ss_size); 153 else 154 sp = (char *)tf->tf_rsp - 128; 155 156 /* 157 * Save FPU state, if any 158 */ 159 if (l->l_md.md_flags & MDL_USEDFPU) { 160 sp = (char *) 161 (((long)sp - sizeof(struct linux__fpstate)) & ~0xfUL); 162 fpsp = (struct linux__fpstate *)sp; 163 } else 164 fpsp = NULL; 165 166 /* 167 * Populate the rt_sigframe 168 */ 169 sp = (char *) 170 ((((long)sp - sizeof(struct linux_rt_sigframe)) & ~0xfUL) - 8); 171 sfp = (struct linux_rt_sigframe *)sp; 172 173 memset(&sigframe, 0, sizeof(sigframe)); 174 if (ps->sa_sigdesc[sig].sd_vers != 0) 175 sigframe.pretcode = 176 (char *)(u_long)ps->sa_sigdesc[sig].sd_tramp; 177 else 178 sigframe.pretcode = NULL; 179 180 /* 181 * The user context 182 */ 183 sigframe.uc.luc_flags = 0; 184 sigframe.uc.luc_link = NULL; 185 186 /* This is used regardless of SA_ONSTACK in Linux */ 187 sigframe.uc.luc_stack.ss_sp = l->l_sigstk.ss_sp; 188 sigframe.uc.luc_stack.ss_size = l->l_sigstk.ss_size; 189 sigframe.uc.luc_stack.ss_flags = 0; 190 if (l->l_sigstk.ss_flags & SS_ONSTACK) 191 sigframe.uc.luc_stack.ss_flags |= LINUX_SS_ONSTACK; 192 if (l->l_sigstk.ss_flags & SS_DISABLE) 193 sigframe.uc.luc_stack.ss_flags |= LINUX_SS_DISABLE; 194 195 sigframe.uc.luc_mcontext.r8 = tf->tf_r8; 196 sigframe.uc.luc_mcontext.r9 = tf->tf_r9; 197 sigframe.uc.luc_mcontext.r10 = tf->tf_r10; 198 sigframe.uc.luc_mcontext.r11 = tf->tf_r11; 199 sigframe.uc.luc_mcontext.r12 = tf->tf_r12; 200 sigframe.uc.luc_mcontext.r13 = tf->tf_r13; 201 sigframe.uc.luc_mcontext.r14 = tf->tf_r14; 202 sigframe.uc.luc_mcontext.r15 = tf->tf_r15; 203 sigframe.uc.luc_mcontext.rdi = tf->tf_rdi; 204 sigframe.uc.luc_mcontext.rsi = tf->tf_rsi; 205 sigframe.uc.luc_mcontext.rbp = tf->tf_rbp; 206 sigframe.uc.luc_mcontext.rbx = tf->tf_rbx; 207 sigframe.uc.luc_mcontext.rdx = tf->tf_rdx; 208 sigframe.uc.luc_mcontext.rax = tf->tf_rax; 209 sigframe.uc.luc_mcontext.rcx = tf->tf_rcx; 210 sigframe.uc.luc_mcontext.rsp = tf->tf_rsp; 211 sigframe.uc.luc_mcontext.rip = tf->tf_rip; 212 sigframe.uc.luc_mcontext.eflags = tf->tf_rflags; 213 sigframe.uc.luc_mcontext.cs = tf->tf_cs; 214 sigframe.uc.luc_mcontext.gs = tf->tf_gs; 215 sigframe.uc.luc_mcontext.fs = tf->tf_fs; 216 sigframe.uc.luc_mcontext.err = tf->tf_err; 217 sigframe.uc.luc_mcontext.trapno = tf->tf_trapno; 218 native_to_linux_sigset(&lmask, mask); 219 sigframe.uc.luc_mcontext.oldmask = lmask.sig[0]; 220 sigframe.uc.luc_mcontext.cr2 = (long)pcb->pcb_onfault; 221 sigframe.uc.luc_mcontext.fpstate = fpsp; 222 native_to_linux_sigset(&sigframe.uc.luc_sigmask, mask); 223 native_to_linux_siginfo(&sigframe.info, &ksi->ksi_info); 224 sendsig_reset(l, sig); 225 mutex_exit(p->p_lock); 226 error = 0; 227 228 /* 229 * Save FPU state, if any 230 */ 231 if (fpsp != NULL) { 232 (void)process_read_fpregs(l, &fpregs); 233 memset(&fpstate, 0, sizeof(fpstate)); 234 fpstate.cwd = fpregs.fp_fcw; 235 fpstate.swd = fpregs.fp_fsw; 236 fpstate.twd = fpregs.fp_ftw; 237 fpstate.fop = fpregs.fp_fop; 238 fpstate.rip = fpregs.fp_rip; 239 fpstate.rdp = fpregs.fp_rdp; 240 fpstate.mxcsr = fpregs.fp_mxcsr; 241 fpstate.mxcsr_mask = fpregs.fp_mxcsr_mask; 242 memcpy(&fpstate.st_space, &fpregs.fp_st, 243 sizeof(fpstate.st_space)); 244 memcpy(&fpstate.xmm_space, &fpregs.fp_xmm, 245 sizeof(fpstate.xmm_space)); 246 error = copyout(&fpstate, fpsp, sizeof(fpstate)); 247 } 248 249 if (error == 0) 250 error = copyout(&sigframe, sp, sizeof(sigframe)); 251 252 mutex_enter(p->p_lock); 253 254 if (error != 0) { 255 sigexit(l, SIGILL); 256 return; 257 } 258 259 linux_buildcontext(l, catcher, sp); 260 tf->tf_rdi = sigframe.info.lsi_signo; 261 tf->tf_rax = 0; 262 tf->tf_rsi = (long)&sfp->info; 263 tf->tf_rdx = (long)&sfp->uc; 264 265 /* 266 * Remember we use signal stack 267 */ 268 if (onstack) 269 l->l_sigstk.ss_flags |= SS_ONSTACK; 270 return; 271 } 272 273 int 274 linux_sys_modify_ldt(struct lwp *l, const struct linux_sys_modify_ldt_args *v, register_t *retval) 275 { 276 printf("linux_sys_modify_ldt\n"); 277 return 0; 278 } 279 280 int 281 linux_sys_iopl(struct lwp *l, const struct linux_sys_iopl_args *v, register_t *retval) 282 { 283 return 0; 284 } 285 286 int 287 linux_sys_ioperm(struct lwp *l, const struct linux_sys_ioperm_args *v, register_t *retval) 288 { 289 return 0; 290 } 291 292 dev_t 293 linux_fakedev(dev_t dev, int raw) 294 { 295 296 extern const struct cdevsw ptc_cdevsw, pts_cdevsw; 297 const struct cdevsw *cd = cdevsw_lookup(dev); 298 299 if (raw) { 300 #if (NWSDISPLAY > 0) 301 extern const struct cdevsw wsdisplay_cdevsw; 302 if (cd == &wsdisplay_cdevsw) 303 return makedev(LINUX_CONS_MAJOR, (minor(dev) + 1)); 304 #endif 305 } 306 307 if (cd == &ptc_cdevsw) 308 return makedev(LINUX_PTC_MAJOR, minor(dev)); 309 if (cd == &pts_cdevsw) 310 return makedev(LINUX_PTS_MAJOR, minor(dev)); 311 312 return ((minor(dev) & 0xff) | ((major(dev) & 0xfff) << 8) 313 | (((unsigned long long int) (minor(dev) & ~0xff)) << 12) 314 | (((unsigned long long int) (major(dev) & ~0xfff)) << 32)); 315 } 316 317 int 318 linux_machdepioctl(struct lwp *l, const struct linux_sys_ioctl_args *v, register_t *retval) 319 { 320 return 0; 321 } 322 323 int 324 linux_sys_rt_sigreturn(struct lwp *l, const void *v, register_t *retval) 325 { 326 struct linux_ucontext *luctx; 327 struct trapframe *tf = l->l_md.md_regs; 328 struct linux_sigcontext *lsigctx; 329 struct linux__fpstate fpstate; 330 struct linux_rt_sigframe frame, *fp; 331 ucontext_t uctx; 332 mcontext_t *mctx; 333 struct fxsave64 *fxarea; 334 int error; 335 336 fp = (struct linux_rt_sigframe *)(tf->tf_rsp - 8); 337 if ((error = copyin(fp, &frame, sizeof(frame))) != 0) { 338 mutex_enter(l->l_proc->p_lock); 339 sigexit(l, SIGILL); 340 return error; 341 } 342 luctx = &frame.uc; 343 lsigctx = &luctx->luc_mcontext; 344 345 memset(&uctx, 0, sizeof(uctx)); 346 mctx = (mcontext_t *)&uctx.uc_mcontext; 347 fxarea = (struct fxsave64 *)&mctx->__fpregs; 348 349 /* 350 * Set the flags. Linux always have CPU, stack and signal state, 351 * FPU is optional. uc_flags is not used to tell what we have. 352 */ 353 uctx.uc_flags = (_UC_SIGMASK|_UC_CPU|_UC_STACK|_UC_CLRSTACK); 354 if (lsigctx->fpstate != NULL) 355 uctx.uc_flags |= _UC_FPU; 356 uctx.uc_link = NULL; 357 358 /* 359 * Signal set 360 */ 361 linux_to_native_sigset(&uctx.uc_sigmask, &luctx->luc_sigmask); 362 363 /* 364 * CPU state 365 */ 366 mctx->__gregs[_REG_R8] = lsigctx->r8; 367 mctx->__gregs[_REG_R9] = lsigctx->r9; 368 mctx->__gregs[_REG_R10] = lsigctx->r10; 369 mctx->__gregs[_REG_R11] = lsigctx->r11; 370 mctx->__gregs[_REG_R12] = lsigctx->r12; 371 mctx->__gregs[_REG_R13] = lsigctx->r13; 372 mctx->__gregs[_REG_R14] = lsigctx->r14; 373 mctx->__gregs[_REG_R15] = lsigctx->r15; 374 mctx->__gregs[_REG_RDI] = lsigctx->rdi; 375 mctx->__gregs[_REG_RSI] = lsigctx->rsi; 376 mctx->__gregs[_REG_RBP] = lsigctx->rbp; 377 mctx->__gregs[_REG_RBX] = lsigctx->rbx; 378 mctx->__gregs[_REG_RAX] = lsigctx->rax; 379 mctx->__gregs[_REG_RDX] = lsigctx->rdx; 380 mctx->__gregs[_REG_RCX] = lsigctx->rcx; 381 mctx->__gregs[_REG_RIP] = lsigctx->rip; 382 mctx->__gregs[_REG_RFLAGS] = lsigctx->eflags; 383 mctx->__gregs[_REG_CS] = lsigctx->cs; 384 mctx->__gregs[_REG_GS] = lsigctx->gs; 385 mctx->__gregs[_REG_FS] = lsigctx->fs; 386 mctx->__gregs[_REG_ERR] = lsigctx->err; 387 mctx->__gregs[_REG_TRAPNO] = lsigctx->trapno; 388 mctx->__gregs[_REG_ES] = tf->tf_es; 389 mctx->__gregs[_REG_DS] = tf->tf_ds; 390 mctx->__gregs[_REG_RSP] = lsigctx->rsp; /* XXX */ 391 mctx->__gregs[_REG_SS] = tf->tf_ss; 392 393 /* 394 * FPU state 395 */ 396 if (lsigctx->fpstate != NULL) { 397 error = copyin(lsigctx->fpstate, &fpstate, sizeof(fpstate)); 398 if (error != 0) { 399 mutex_enter(l->l_proc->p_lock); 400 sigexit(l, SIGILL); 401 return error; 402 } 403 404 fxarea->fx_fcw = fpstate.cwd; 405 fxarea->fx_fsw = fpstate.swd; 406 fxarea->fx_ftw = fpstate.twd; 407 fxarea->fx_fop = fpstate.fop; 408 fxarea->fx_rip = fpstate.rip; 409 fxarea->fx_rdp = fpstate.rdp; 410 fxarea->fx_mxcsr = fpstate.mxcsr; 411 fxarea->fx_mxcsr_mask = fpstate.mxcsr_mask; 412 memcpy(&fxarea->fx_st, &fpstate.st_space, 413 sizeof(fxarea->fx_st)); 414 memcpy(&fxarea->fx_xmm, &fpstate.xmm_space, 415 sizeof(fxarea->fx_xmm)); 416 } 417 418 /* 419 * And the stack 420 */ 421 uctx.uc_stack.ss_flags = 0; 422 if (luctx->luc_stack.ss_flags & LINUX_SS_ONSTACK) 423 uctx.uc_stack.ss_flags |= SS_ONSTACK; 424 425 if (luctx->luc_stack.ss_flags & LINUX_SS_DISABLE) 426 uctx.uc_stack.ss_flags |= SS_DISABLE; 427 428 uctx.uc_stack.ss_sp = luctx->luc_stack.ss_sp; 429 uctx.uc_stack.ss_size = luctx->luc_stack.ss_size; 430 431 /* 432 * And let setucontext deal with that. 433 */ 434 mutex_enter(l->l_proc->p_lock); 435 error = setucontext(l, &uctx); 436 mutex_exit(l->l_proc->p_lock); 437 if (error) 438 return error; 439 440 return EJUSTRETURN; 441 } 442 443 int 444 linux_sys_arch_prctl(struct lwp *l, 445 const struct linux_sys_arch_prctl_args *uap, register_t *retval) 446 { 447 /* { 448 syscallarg(int) code; 449 syscallarg(unsigned long) addr; 450 } */ 451 void *addr = (void *)SCARG(uap, addr); 452 453 switch(SCARG(uap, code)) { 454 case LINUX_ARCH_SET_GS: 455 return x86_set_sdbase(addr, 'g', l, true); 456 457 case LINUX_ARCH_GET_GS: 458 return x86_get_sdbase(addr, 'g'); 459 460 case LINUX_ARCH_SET_FS: 461 return x86_set_sdbase(addr, 'f', l, true); 462 463 case LINUX_ARCH_GET_FS: 464 return x86_get_sdbase(addr, 'f'); 465 466 default: 467 #ifdef DEBUG_LINUX 468 printf("linux_sys_arch_prctl: unexpected code %d\n", 469 SCARG(uap, code)); 470 #endif 471 return EINVAL; 472 } 473 /* NOTREACHED */ 474 } 475 476 const int linux_vsyscall_to_syscall[] = { 477 LINUX_SYS_gettimeofday, 478 LINUX_SYS_time, 479 LINUX_SYS_nosys, /* nosys */ 480 LINUX_SYS_nosys, /* nosys */ 481 }; 482 483 int 484 linux_usertrap(struct lwp *l, vaddr_t trapaddr, void *arg) 485 { 486 struct trapframe *tf = arg; 487 uint64_t retaddr; 488 int vsyscallnr; 489 490 /* 491 * Check for a vsyscall. %rip must be the fault address, 492 * and the address must be in the Linux vsyscall area. 493 * Also, vsyscalls are only done at 1024-byte boundaries. 494 */ 495 496 if (__predict_true(trapaddr < LINUX_VSYSCALL_START)) 497 return 0; 498 499 if (trapaddr != tf->tf_rip) 500 return 0; 501 502 if ((tf->tf_rip & (LINUX_VSYSCALL_SIZE - 1)) != 0) 503 return 0; 504 505 vsyscallnr = (tf->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SIZE; 506 507 if (vsyscallnr > LINUX_VSYSCALL_MAXNR) 508 return 0; 509 510 /* 511 * Get the return address from the top of the stack, 512 * and fix up the return address. 513 * This assumes the faulting instruction was callq *reg, 514 * which is the only way that vsyscalls are ever entered. 515 */ 516 if (copyin((void *)tf->tf_rsp, &retaddr, sizeof retaddr) != 0) 517 return 0; 518 tf->tf_rip = retaddr; 519 tf->tf_rax = linux_vsyscall_to_syscall[vsyscallnr]; 520 tf->tf_rsp += 8; /* "pop" the return address */ 521 522 #if 0 523 printf("usertrap: rip %p rsp %p retaddr %p vsys %d sys %d\n", 524 (void *)tf->tf_rip, (void *)tf->tf_rsp, (void *)retaddr, 525 vsyscallnr, (int)tf->tf_rax); 526 #endif 527 528 (*l->l_proc->p_md.md_syscall)(tf); 529 530 return 1; 531 } 532 533 static void 534 linux_buildcontext(struct lwp *l, void *catcher, void *f) 535 { 536 struct trapframe *tf = l->l_md.md_regs; 537 538 tf->tf_ds = GSEL(GUDATA_SEL, SEL_UPL); 539 tf->tf_rip = (u_int64_t)catcher; 540 tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL); 541 tf->tf_rflags &= ~PSL_CLEARSIG; 542 tf->tf_rsp = (u_int64_t)f; 543 tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL); 544 } 545