1 /* $NetBSD: linux_machdep.c,v 1.56 2018/01/01 08:03:43 maxv 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 1.56 2018/01/01 08:03:43 maxv 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/mcontext.h> 51 #include <machine/specialreg.h> 52 #include <machine/vmparam.h> 53 #include <machine/cpufunc.h> 54 #include <x86/include/sysarch.h> 55 56 /* 57 * To see whether wscons is configured (for virtual console ioctl calls). 58 */ 59 #if defined(_KERNEL_OPT) 60 #include "opt_user_ldt.h" 61 #include "wsdisplay.h" 62 #endif 63 64 #if (NWSDISPLAY > 0) 65 #include <dev/wscons/wsconsio.h> 66 #include <dev/wscons/wsdisplay_usl_io.h> 67 #endif 68 69 70 #include <compat/linux/common/linux_signal.h> 71 #include <compat/linux/common/linux_errno.h> 72 #include <compat/linux/common/linux_exec.h> 73 #include <compat/linux/common/linux_ioctl.h> 74 #include <compat/linux/common/linux_prctl.h> 75 #include <compat/linux/common/linux_machdep.h> 76 #include <compat/linux/common/linux_ipc.h> 77 #include <compat/linux/common/linux_sem.h> 78 #include <compat/linux/linux_syscall.h> 79 #include <compat/linux/linux_syscallargs.h> 80 81 static void linux_buildcontext(struct lwp *, void *, void *); 82 83 void 84 linux_setregs(struct lwp *l, struct exec_package *epp, vaddr_t stack) 85 { 86 struct pcb *pcb = lwp_getpcb(l); 87 struct trapframe *tf; 88 89 #ifdef USER_LDT 90 pmap_ldt_cleanup(l); 91 #endif 92 93 fpu_save_area_clear(l, __NetBSD_NPXCW__); 94 pcb->pcb_flags = 0; 95 96 l->l_proc->p_flag &= ~PK_32; 97 98 l->l_md.md_flags = MDL_IRET; 99 100 tf = l->l_md.md_regs; 101 tf->tf_rax = 0; 102 tf->tf_rbx = 0; 103 tf->tf_rcx = epp->ep_entry; 104 tf->tf_rdx = 0; 105 tf->tf_rsi = 0; 106 tf->tf_rdi = 0; 107 tf->tf_rbp = 0; 108 tf->tf_rsp = stack; 109 tf->tf_r8 = 0; 110 tf->tf_r9 = 0; 111 tf->tf_r10 = 0; 112 tf->tf_r11 = 0; 113 tf->tf_r12 = 0; 114 tf->tf_r13 = 0; 115 tf->tf_r14 = 0; 116 tf->tf_r15 = 0; 117 tf->tf_rip = epp->ep_entry; 118 tf->tf_rflags = PSL_USERSET; 119 tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL); 120 tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL); 121 tf->tf_ds = GSEL(GUDATA_SEL, SEL_UPL); 122 tf->tf_es = 0; 123 cpu_segregs64_zero(l); 124 125 return; 126 } 127 128 void 129 linux_sendsig(const ksiginfo_t *ksi, const sigset_t *mask) 130 { 131 struct lwp *l = curlwp; 132 struct proc *p = l->l_proc; 133 struct pcb *pcb = lwp_getpcb(l); 134 struct sigacts *ps = p->p_sigacts; 135 int onstack, error; 136 int sig = ksi->ksi_signo; 137 struct linux_rt_sigframe *sfp, sigframe; 138 struct linux__fpstate *fpsp; 139 struct fpreg fpregs; 140 struct trapframe *tf = l->l_md.md_regs; 141 sig_t catcher = SIGACTION(p, sig).sa_handler; 142 linux_sigset_t lmask; 143 char *sp; 144 145 /* Do we need to jump onto the signal stack? */ 146 onstack = 147 (l->l_sigstk.ss_flags & (SS_DISABLE | SS_ONSTACK)) == 0 && 148 (SIGACTION(p, sig).sa_flags & SA_ONSTACK) != 0; 149 150 /* Allocate space for the signal handler context. */ 151 if (onstack) 152 sp = ((char *)l->l_sigstk.ss_sp + 153 l->l_sigstk.ss_size); 154 else 155 sp = (char *)tf->tf_rsp - 128; 156 157 /* Save FPU state */ 158 sp = (char *) (((long)sp - sizeof (*fpsp)) & ~0xfUL); 159 fpsp = (struct linux__fpstate *)sp; 160 161 /* 162 * Populate the rt_sigframe 163 */ 164 sp = (char *) 165 ((((long)sp - sizeof(struct linux_rt_sigframe)) & ~0xfUL) - 8); 166 sfp = (struct linux_rt_sigframe *)sp; 167 168 memset(&sigframe, 0, sizeof(sigframe)); 169 if (ps->sa_sigdesc[sig].sd_vers != 0) 170 sigframe.pretcode = 171 (char *)(u_long)ps->sa_sigdesc[sig].sd_tramp; 172 else 173 sigframe.pretcode = NULL; 174 175 /* 176 * The user context 177 */ 178 sigframe.uc.luc_flags = 0; 179 sigframe.uc.luc_link = NULL; 180 181 /* This is used regardless of SA_ONSTACK in Linux */ 182 sigframe.uc.luc_stack.ss_sp = l->l_sigstk.ss_sp; 183 sigframe.uc.luc_stack.ss_size = l->l_sigstk.ss_size; 184 sigframe.uc.luc_stack.ss_flags = 0; 185 if (l->l_sigstk.ss_flags & SS_ONSTACK) 186 sigframe.uc.luc_stack.ss_flags |= LINUX_SS_ONSTACK; 187 if (l->l_sigstk.ss_flags & SS_DISABLE) 188 sigframe.uc.luc_stack.ss_flags |= LINUX_SS_DISABLE; 189 190 sigframe.uc.luc_mcontext.r8 = tf->tf_r8; 191 sigframe.uc.luc_mcontext.r9 = tf->tf_r9; 192 sigframe.uc.luc_mcontext.r10 = tf->tf_r10; 193 sigframe.uc.luc_mcontext.r11 = tf->tf_r11; 194 sigframe.uc.luc_mcontext.r12 = tf->tf_r12; 195 sigframe.uc.luc_mcontext.r13 = tf->tf_r13; 196 sigframe.uc.luc_mcontext.r14 = tf->tf_r14; 197 sigframe.uc.luc_mcontext.r15 = tf->tf_r15; 198 sigframe.uc.luc_mcontext.rdi = tf->tf_rdi; 199 sigframe.uc.luc_mcontext.rsi = tf->tf_rsi; 200 sigframe.uc.luc_mcontext.rbp = tf->tf_rbp; 201 sigframe.uc.luc_mcontext.rbx = tf->tf_rbx; 202 sigframe.uc.luc_mcontext.rdx = tf->tf_rdx; 203 sigframe.uc.luc_mcontext.rax = tf->tf_rax; 204 sigframe.uc.luc_mcontext.rcx = tf->tf_rcx; 205 sigframe.uc.luc_mcontext.rsp = tf->tf_rsp; 206 sigframe.uc.luc_mcontext.rip = tf->tf_rip; 207 sigframe.uc.luc_mcontext.eflags = tf->tf_rflags; 208 sigframe.uc.luc_mcontext.cs = GSEL(GUCODE_SEL, SEL_UPL); 209 sigframe.uc.luc_mcontext.gs = tf->tf_gs & 0xFFFF; 210 sigframe.uc.luc_mcontext.fs = tf->tf_fs & 0xFFFF; 211 sigframe.uc.luc_mcontext.err = tf->tf_err; 212 sigframe.uc.luc_mcontext.trapno = tf->tf_trapno; 213 native_to_linux_sigset(&lmask, mask); 214 sigframe.uc.luc_mcontext.oldmask = lmask.sig[0]; 215 sigframe.uc.luc_mcontext.cr2 = (long)pcb->pcb_onfault; 216 sigframe.uc.luc_mcontext.fpstate = fpsp; 217 native_to_linux_sigset(&sigframe.uc.luc_sigmask, mask); 218 native_to_linux_siginfo(&sigframe.info, &ksi->ksi_info); 219 sendsig_reset(l, sig); 220 mutex_exit(p->p_lock); 221 error = 0; 222 223 /* 224 * Save FPU state, if any 225 */ 226 if (fpsp != NULL) { 227 size_t fp_size = sizeof fpregs; 228 /* The netbsd and linux structures both match the fxsave data */ 229 (void)process_read_fpregs(l, &fpregs, &fp_size); 230 error = copyout(&fpregs, fpsp, sizeof(*fpsp)); 231 } 232 233 if (error == 0) 234 error = copyout(&sigframe, sp, sizeof(sigframe)); 235 236 mutex_enter(p->p_lock); 237 238 if (error != 0) { 239 sigexit(l, SIGILL); 240 return; 241 } 242 243 if ((vaddr_t)catcher >= VM_MAXUSER_ADDRESS) { 244 sigexit(l, SIGILL); 245 return; 246 } 247 248 linux_buildcontext(l, catcher, sp); 249 tf->tf_rdi = sigframe.info.lsi_signo; 250 tf->tf_rax = 0; 251 tf->tf_rsi = (long)&sfp->info; 252 tf->tf_rdx = (long)&sfp->uc; 253 254 /* 255 * Remember we use signal stack 256 */ 257 if (onstack) 258 l->l_sigstk.ss_flags |= SS_ONSTACK; 259 return; 260 } 261 262 int 263 linux_sys_modify_ldt(struct lwp *l, const struct linux_sys_modify_ldt_args *v, register_t *retval) 264 { 265 printf("linux_sys_modify_ldt\n"); 266 return 0; 267 } 268 269 int 270 linux_sys_iopl(struct lwp *l, const struct linux_sys_iopl_args *v, register_t *retval) 271 { 272 return 0; 273 } 274 275 int 276 linux_sys_ioperm(struct lwp *l, const struct linux_sys_ioperm_args *v, register_t *retval) 277 { 278 return 0; 279 } 280 281 dev_t 282 linux_fakedev(dev_t dev, int raw) 283 { 284 285 extern const struct cdevsw ptc_cdevsw, pts_cdevsw; 286 const struct cdevsw *cd = cdevsw_lookup(dev); 287 288 if (raw) { 289 #if (NWSDISPLAY > 0) 290 extern const struct cdevsw wsdisplay_cdevsw; 291 if (cd == &wsdisplay_cdevsw) 292 return makedev(LINUX_CONS_MAJOR, (minor(dev) + 1)); 293 #endif 294 } 295 296 if (cd == &ptc_cdevsw) 297 return makedev(LINUX_PTC_MAJOR, minor(dev)); 298 if (cd == &pts_cdevsw) 299 return makedev(LINUX_PTS_MAJOR, minor(dev)); 300 301 return ((minor(dev) & 0xff) | ((major(dev) & 0xfff) << 8) 302 | (((unsigned long long int) (minor(dev) & ~0xff)) << 12) 303 | (((unsigned long long int) (major(dev) & ~0xfff)) << 32)); 304 } 305 306 int 307 linux_machdepioctl(struct lwp *l, const struct linux_sys_ioctl_args *v, register_t *retval) 308 { 309 return 0; 310 } 311 312 int 313 linux_sys_rt_sigreturn(struct lwp *l, const void *v, register_t *retval) 314 { 315 struct linux_ucontext *luctx; 316 struct trapframe *tf = l->l_md.md_regs; 317 struct linux_sigcontext *lsigctx; 318 struct linux_rt_sigframe frame, *fp; 319 ucontext_t uctx; 320 mcontext_t *mctx; 321 struct fxsave *fxarea; 322 int error; 323 324 fp = (struct linux_rt_sigframe *)(tf->tf_rsp - 8); 325 if ((error = copyin(fp, &frame, sizeof(frame))) != 0) { 326 mutex_enter(l->l_proc->p_lock); 327 sigexit(l, SIGILL); 328 return error; 329 } 330 luctx = &frame.uc; 331 lsigctx = &luctx->luc_mcontext; 332 333 memset(&uctx, 0, sizeof(uctx)); 334 mctx = (mcontext_t *)&uctx.uc_mcontext; 335 fxarea = (struct fxsave *)&mctx->__fpregs; 336 337 /* 338 * Set the flags. Linux always have CPU, stack and signal state, 339 * FPU is optional. uc_flags is not used to tell what we have. 340 */ 341 uctx.uc_flags = (_UC_SIGMASK|_UC_CPU|_UC_STACK|_UC_CLRSTACK); 342 if (lsigctx->fpstate != NULL) 343 uctx.uc_flags |= _UC_FPU; 344 uctx.uc_link = NULL; 345 346 /* 347 * Signal set 348 */ 349 linux_to_native_sigset(&uctx.uc_sigmask, &luctx->luc_sigmask); 350 351 /* 352 * CPU state 353 */ 354 mctx->__gregs[_REG_R8] = lsigctx->r8; 355 mctx->__gregs[_REG_R9] = lsigctx->r9; 356 mctx->__gregs[_REG_R10] = lsigctx->r10; 357 mctx->__gregs[_REG_R11] = lsigctx->r11; 358 mctx->__gregs[_REG_R12] = lsigctx->r12; 359 mctx->__gregs[_REG_R13] = lsigctx->r13; 360 mctx->__gregs[_REG_R14] = lsigctx->r14; 361 mctx->__gregs[_REG_R15] = lsigctx->r15; 362 mctx->__gregs[_REG_RDI] = lsigctx->rdi; 363 mctx->__gregs[_REG_RSI] = lsigctx->rsi; 364 mctx->__gregs[_REG_RBP] = lsigctx->rbp; 365 mctx->__gregs[_REG_RBX] = lsigctx->rbx; 366 mctx->__gregs[_REG_RAX] = lsigctx->rax; 367 mctx->__gregs[_REG_RDX] = lsigctx->rdx; 368 mctx->__gregs[_REG_RCX] = lsigctx->rcx; 369 mctx->__gregs[_REG_RIP] = lsigctx->rip; 370 mctx->__gregs[_REG_RFLAGS] = lsigctx->eflags; 371 mctx->__gregs[_REG_CS] = lsigctx->cs & 0xFFFF; 372 mctx->__gregs[_REG_GS] = lsigctx->gs & 0xFFFF; 373 mctx->__gregs[_REG_FS] = lsigctx->fs & 0xFFFF; 374 mctx->__gregs[_REG_ERR] = lsigctx->err; 375 mctx->__gregs[_REG_TRAPNO] = lsigctx->trapno; 376 mctx->__gregs[_REG_ES] = tf->tf_es & 0xFFFF; 377 mctx->__gregs[_REG_DS] = tf->tf_ds & 0xFFFF; 378 mctx->__gregs[_REG_RSP] = lsigctx->rsp; /* XXX */ 379 mctx->__gregs[_REG_SS] = tf->tf_ss & 0xFFFF; 380 381 /* 382 * FPU state 383 */ 384 if (lsigctx->fpstate != NULL) { 385 /* Both structures match the fxstate data */ 386 error = copyin(lsigctx->fpstate, fxarea, sizeof(*fxarea)); 387 if (error != 0) { 388 mutex_enter(l->l_proc->p_lock); 389 sigexit(l, SIGILL); 390 return error; 391 } 392 } 393 394 /* 395 * And the stack 396 */ 397 uctx.uc_stack.ss_flags = 0; 398 if (luctx->luc_stack.ss_flags & LINUX_SS_ONSTACK) 399 uctx.uc_stack.ss_flags |= SS_ONSTACK; 400 401 if (luctx->luc_stack.ss_flags & LINUX_SS_DISABLE) 402 uctx.uc_stack.ss_flags |= SS_DISABLE; 403 404 uctx.uc_stack.ss_sp = luctx->luc_stack.ss_sp; 405 uctx.uc_stack.ss_size = luctx->luc_stack.ss_size; 406 407 /* 408 * And let setucontext deal with that. 409 */ 410 mutex_enter(l->l_proc->p_lock); 411 error = setucontext(l, &uctx); 412 mutex_exit(l->l_proc->p_lock); 413 if (error) 414 return error; 415 416 return EJUSTRETURN; 417 } 418 419 int 420 linux_sys_arch_prctl(struct lwp *l, 421 const struct linux_sys_arch_prctl_args *uap, register_t *retval) 422 { 423 /* { 424 syscallarg(int) code; 425 syscallarg(unsigned long) addr; 426 } */ 427 void *addr = (void *)SCARG(uap, addr); 428 429 switch(SCARG(uap, code)) { 430 case LINUX_ARCH_SET_GS: 431 return x86_set_sdbase(addr, 'g', l, true); 432 433 case LINUX_ARCH_GET_GS: 434 return x86_get_sdbase(addr, 'g'); 435 436 case LINUX_ARCH_SET_FS: 437 return x86_set_sdbase(addr, 'f', l, true); 438 439 case LINUX_ARCH_GET_FS: 440 return x86_get_sdbase(addr, 'f'); 441 442 default: 443 #ifdef DEBUG_LINUX 444 printf("linux_sys_arch_prctl: unexpected code %d\n", 445 SCARG(uap, code)); 446 #endif 447 return EINVAL; 448 } 449 /* NOTREACHED */ 450 } 451 452 const int linux_vsyscall_to_syscall[] = { 453 LINUX_SYS_gettimeofday, 454 LINUX_SYS_time, 455 LINUX_SYS_nosys, /* nosys */ 456 LINUX_SYS_nosys, /* nosys */ 457 }; 458 459 int 460 linux_usertrap(struct lwp *l, vaddr_t trapaddr, void *arg) 461 { 462 struct trapframe *tf = arg; 463 uint64_t retaddr; 464 size_t vsyscallnr; 465 466 /* 467 * Check for a vsyscall. %rip must be the fault address, 468 * and the address must be in the Linux vsyscall area. 469 * Also, vsyscalls are only done at 1024-byte boundaries. 470 */ 471 472 if (__predict_true(trapaddr < LINUX_VSYSCALL_START)) 473 return 0; 474 475 if (trapaddr != tf->tf_rip) 476 return 0; 477 478 if ((tf->tf_rip & (LINUX_VSYSCALL_SIZE - 1)) != 0) 479 return 0; 480 481 vsyscallnr = (tf->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SIZE; 482 483 if (vsyscallnr > LINUX_VSYSCALL_MAXNR) 484 return 0; 485 486 /* 487 * Get the return address from the top of the stack, 488 * and fix up the return address. 489 * This assumes the faulting instruction was callq *reg, 490 * which is the only way that vsyscalls are ever entered. 491 */ 492 if (copyin((void *)tf->tf_rsp, &retaddr, sizeof retaddr) != 0) 493 return 0; 494 if ((vaddr_t)retaddr >= VM_MAXUSER_ADDRESS) 495 return 0; 496 tf->tf_rip = retaddr; 497 tf->tf_rax = linux_vsyscall_to_syscall[vsyscallnr]; 498 tf->tf_rsp += 8; /* "pop" the return address */ 499 500 #if 0 501 printf("usertrap: rip %p rsp %p retaddr %p vsys %d sys %d\n", 502 (void *)tf->tf_rip, (void *)tf->tf_rsp, (void *)retaddr, 503 vsyscallnr, (int)tf->tf_rax); 504 #endif 505 506 (*l->l_proc->p_md.md_syscall)(tf); 507 508 return 1; 509 } 510 511 static void 512 linux_buildcontext(struct lwp *l, void *catcher, void *f) 513 { 514 struct trapframe *tf = l->l_md.md_regs; 515 516 tf->tf_ds = GSEL(GUDATA_SEL, SEL_UPL); 517 tf->tf_rip = (u_int64_t)catcher; 518 tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL); 519 tf->tf_rflags &= ~PSL_CLEARSIG; 520 tf->tf_rsp = (u_int64_t)f; 521 tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL); 522 } 523