1 /* $NetBSD: linux_machdep.c,v 1.62 2024/10/01 16:35:42 riastradh 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 1.62 2024/10/01 16:35:42 riastradh 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_clear(l, __NetBSD_NPXCW__); 94 95 kpreempt_disable(); 96 pcb->pcb_flags = 0; 97 l->l_proc->p_flag &= ~PK_32; 98 l->l_md.md_flags = MDL_IRET; 99 cpu_segregs64_zero(l); 100 kpreempt_enable(); 101 102 tf = l->l_md.md_regs; 103 tf->tf_rax = 0; 104 tf->tf_rbx = 0; 105 tf->tf_rcx = epp->ep_entry; 106 tf->tf_rdx = 0; 107 tf->tf_rsi = 0; 108 tf->tf_rdi = 0; 109 tf->tf_rbp = 0; 110 tf->tf_rsp = stack; 111 tf->tf_r8 = 0; 112 tf->tf_r9 = 0; 113 tf->tf_r10 = 0; 114 tf->tf_r11 = 0; 115 tf->tf_r12 = 0; 116 tf->tf_r13 = 0; 117 tf->tf_r14 = 0; 118 tf->tf_r15 = 0; 119 tf->tf_rip = epp->ep_entry; 120 tf->tf_rflags = PSL_USERSET; 121 tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL); 122 tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL); 123 tf->tf_ds = GSEL(GUDATA_SEL, SEL_UPL); 124 tf->tf_es = 0; 125 126 return; 127 } 128 129 void 130 linux_sendsig(const ksiginfo_t *ksi, const sigset_t *mask) 131 { 132 struct lwp *l = curlwp; 133 struct proc *p = l->l_proc; 134 struct pcb *pcb = lwp_getpcb(l); 135 struct sigacts *ps = p->p_sigacts; 136 int onstack, error; 137 int sig = ksi->ksi_signo; 138 struct linux_rt_sigframe *sfp, sigframe; 139 struct linux__fpstate *fpsp; 140 struct fpreg fpregs; 141 struct trapframe *tf = l->l_md.md_regs; 142 sig_t catcher = SIGACTION(p, sig).sa_handler; 143 linux_sigset_t lmask; 144 char *sp; 145 146 /* Do we need to jump onto the signal stack? */ 147 onstack = 148 (l->l_sigstk.ss_flags & (SS_DISABLE | SS_ONSTACK)) == 0 && 149 (SIGACTION(p, sig).sa_flags & SA_ONSTACK) != 0; 150 151 /* Allocate space for the signal handler context. */ 152 if (onstack) 153 sp = ((char *)l->l_sigstk.ss_sp + 154 l->l_sigstk.ss_size); 155 else 156 sp = (char *)tf->tf_rsp - 128; 157 158 /* Save FPU state */ 159 sp = (char *) (((long)sp - sizeof (*fpsp)) & ~0xfUL); 160 fpsp = (struct linux__fpstate *)sp; 161 162 /* 163 * Populate the rt_sigframe 164 */ 165 sp = (char *) 166 ((((long)sp - sizeof(struct linux_rt_sigframe)) & ~0xfUL) - 8); 167 sfp = (struct linux_rt_sigframe *)sp; 168 169 memset(&sigframe, 0, sizeof(sigframe)); 170 if (ps->sa_sigdesc[sig].sd_vers != __SIGTRAMP_SIGCODE_VERSION) 171 sigframe.pretcode = 172 (char *)(u_long)ps->sa_sigdesc[sig].sd_tramp; 173 else 174 sigframe.pretcode = NULL; 175 176 /* 177 * The user context 178 */ 179 sigframe.uc.luc_flags = 0; 180 sigframe.uc.luc_link = NULL; 181 182 /* This is used regardless of SA_ONSTACK in Linux */ 183 sigframe.uc.luc_stack.ss_sp = l->l_sigstk.ss_sp; 184 sigframe.uc.luc_stack.ss_size = l->l_sigstk.ss_size; 185 sigframe.uc.luc_stack.ss_flags = 0; 186 if (l->l_sigstk.ss_flags & SS_ONSTACK) 187 sigframe.uc.luc_stack.ss_flags |= LINUX_SS_ONSTACK; 188 if (l->l_sigstk.ss_flags & SS_DISABLE) 189 sigframe.uc.luc_stack.ss_flags |= LINUX_SS_DISABLE; 190 191 sigframe.uc.luc_mcontext.r8 = tf->tf_r8; 192 sigframe.uc.luc_mcontext.r9 = tf->tf_r9; 193 sigframe.uc.luc_mcontext.r10 = tf->tf_r10; 194 sigframe.uc.luc_mcontext.r11 = tf->tf_r11; 195 sigframe.uc.luc_mcontext.r12 = tf->tf_r12; 196 sigframe.uc.luc_mcontext.r13 = tf->tf_r13; 197 sigframe.uc.luc_mcontext.r14 = tf->tf_r14; 198 sigframe.uc.luc_mcontext.r15 = tf->tf_r15; 199 sigframe.uc.luc_mcontext.rdi = tf->tf_rdi; 200 sigframe.uc.luc_mcontext.rsi = tf->tf_rsi; 201 sigframe.uc.luc_mcontext.rbp = tf->tf_rbp; 202 sigframe.uc.luc_mcontext.rbx = tf->tf_rbx; 203 sigframe.uc.luc_mcontext.rdx = tf->tf_rdx; 204 sigframe.uc.luc_mcontext.rax = tf->tf_rax; 205 sigframe.uc.luc_mcontext.rcx = tf->tf_rcx; 206 sigframe.uc.luc_mcontext.rsp = tf->tf_rsp; 207 sigframe.uc.luc_mcontext.rip = tf->tf_rip; 208 sigframe.uc.luc_mcontext.eflags = tf->tf_rflags; 209 sigframe.uc.luc_mcontext.cs = GSEL(GUCODE_SEL, SEL_UPL); 210 sigframe.uc.luc_mcontext.gs = tf->tf_gs & 0xFFFF; 211 sigframe.uc.luc_mcontext.fs = tf->tf_fs & 0xFFFF; 212 sigframe.uc.luc_mcontext.err = tf->tf_err; 213 sigframe.uc.luc_mcontext.trapno = tf->tf_trapno; 214 native_to_linux_sigset(&lmask, mask); 215 sigframe.uc.luc_mcontext.oldmask = lmask.sig[0]; 216 sigframe.uc.luc_mcontext.cr2 = (long)pcb->pcb_onfault; 217 sigframe.uc.luc_mcontext.fpstate = fpsp; 218 native_to_linux_sigset(&sigframe.uc.luc_sigmask, mask); 219 native_to_linux_siginfo(&sigframe.info, &ksi->ksi_info); 220 sendsig_reset(l, sig); 221 mutex_exit(p->p_lock); 222 error = 0; 223 224 /* 225 * Save FPU state, if any 226 */ 227 if (fpsp != NULL) { 228 size_t fp_size = sizeof fpregs; 229 /* The netbsd and linux structures both match the fxsave data */ 230 memset(&fpregs, 0, sizeof(fpregs)); 231 (void)process_read_fpregs(l, &fpregs, &fp_size); 232 error = copyout(&fpregs, fpsp, sizeof(*fpsp)); 233 } 234 235 if (error == 0) 236 error = copyout(&sigframe, sp, sizeof(sigframe)); 237 238 mutex_enter(p->p_lock); 239 240 if (error != 0) { 241 sigexit(l, SIGILL); 242 return; 243 } 244 245 if ((vaddr_t)catcher >= VM_MAXUSER_ADDRESS) { 246 sigexit(l, SIGILL); 247 return; 248 } 249 250 linux_buildcontext(l, catcher, sp); 251 tf->tf_rdi = sigframe.info.lsi_signo; 252 tf->tf_rax = 0; 253 tf->tf_rsi = (long)&sfp->info; 254 tf->tf_rdx = (long)&sfp->uc; 255 256 /* 257 * Remember we use signal stack 258 */ 259 if (onstack) 260 l->l_sigstk.ss_flags |= SS_ONSTACK; 261 return; 262 } 263 264 int 265 linux_sys_modify_ldt(struct lwp *l, const struct linux_sys_modify_ldt_args *v, register_t *retval) 266 { 267 return 0; 268 } 269 270 int 271 linux_sys_iopl(struct lwp *l, const struct linux_sys_iopl_args *v, register_t *retval) 272 { 273 return 0; 274 } 275 276 int 277 linux_sys_ioperm(struct lwp *l, const struct linux_sys_ioperm_args *v, register_t *retval) 278 { 279 return 0; 280 } 281 282 dev_t 283 linux_fakedev(dev_t dev, int raw) 284 { 285 286 extern const struct cdevsw ptc_cdevsw, pts_cdevsw; 287 const struct cdevsw *cd = cdevsw_lookup(dev); 288 289 if (raw) { 290 #if (NWSDISPLAY > 0) 291 extern const struct cdevsw wsdisplay_cdevsw; 292 if (cd == &wsdisplay_cdevsw) 293 return makedev(LINUX_CONS_MAJOR, (minor(dev) + 1)); 294 #endif 295 } 296 297 if (cd == &ptc_cdevsw) 298 return makedev(LINUX_PTC_MAJOR, minor(dev)); 299 if (cd == &pts_cdevsw) 300 return makedev(LINUX_PTS_MAJOR, minor(dev)); 301 302 return ((minor(dev) & 0xff) | ((major(dev) & 0xfff) << 8) 303 | (((unsigned long long int) (minor(dev) & ~0xff)) << 12) 304 | (((unsigned long long int) (major(dev) & ~0xfff)) << 32)); 305 } 306 307 int 308 linux_machdepioctl(struct lwp *l, const struct linux_sys_ioctl_args *v, register_t *retval) 309 { 310 return 0; 311 } 312 313 int 314 linux_sys_rt_sigreturn(struct lwp *l, const void *v, register_t *retval) 315 { 316 struct linux_ucontext *luctx; 317 struct trapframe *tf = l->l_md.md_regs; 318 struct linux_sigcontext *lsigctx; 319 struct linux_rt_sigframe frame, *fp; 320 ucontext_t uctx; 321 mcontext_t *mctx; 322 struct fxsave *fxarea; 323 int error; 324 325 fp = (struct linux_rt_sigframe *)(tf->tf_rsp - 8); 326 if ((error = copyin(fp, &frame, sizeof(frame))) != 0) { 327 mutex_enter(l->l_proc->p_lock); 328 sigexit(l, SIGILL); 329 return error; 330 } 331 luctx = &frame.uc; 332 lsigctx = &luctx->luc_mcontext; 333 334 memset(&uctx, 0, sizeof(uctx)); 335 mctx = (mcontext_t *)&uctx.uc_mcontext; 336 fxarea = (struct fxsave *)&mctx->__fpregs; 337 338 /* 339 * Set the flags. Linux always have CPU, stack and signal state, 340 * FPU is optional. uc_flags is not used to tell what we have. 341 */ 342 uctx.uc_flags = (_UC_SIGMASK|_UC_CPU|_UC_STACK|_UC_CLRSTACK); 343 if (lsigctx->fpstate != NULL) 344 uctx.uc_flags |= _UC_FPU; 345 uctx.uc_link = NULL; 346 347 /* 348 * Signal set 349 */ 350 linux_to_native_sigset(&uctx.uc_sigmask, &luctx->luc_sigmask); 351 352 /* 353 * CPU state 354 */ 355 mctx->__gregs[_REG_R8] = lsigctx->r8; 356 mctx->__gregs[_REG_R9] = lsigctx->r9; 357 mctx->__gregs[_REG_R10] = lsigctx->r10; 358 mctx->__gregs[_REG_R11] = lsigctx->r11; 359 mctx->__gregs[_REG_R12] = lsigctx->r12; 360 mctx->__gregs[_REG_R13] = lsigctx->r13; 361 mctx->__gregs[_REG_R14] = lsigctx->r14; 362 mctx->__gregs[_REG_R15] = lsigctx->r15; 363 mctx->__gregs[_REG_RDI] = lsigctx->rdi; 364 mctx->__gregs[_REG_RSI] = lsigctx->rsi; 365 mctx->__gregs[_REG_RBP] = lsigctx->rbp; 366 mctx->__gregs[_REG_RBX] = lsigctx->rbx; 367 mctx->__gregs[_REG_RAX] = lsigctx->rax; 368 mctx->__gregs[_REG_RDX] = lsigctx->rdx; 369 mctx->__gregs[_REG_RCX] = lsigctx->rcx; 370 mctx->__gregs[_REG_RIP] = lsigctx->rip; 371 mctx->__gregs[_REG_RFLAGS] = lsigctx->eflags; 372 mctx->__gregs[_REG_CS] = lsigctx->cs & 0xFFFF; 373 mctx->__gregs[_REG_GS] = lsigctx->gs & 0xFFFF; 374 mctx->__gregs[_REG_FS] = lsigctx->fs & 0xFFFF; 375 mctx->__gregs[_REG_ERR] = lsigctx->err; 376 mctx->__gregs[_REG_TRAPNO] = lsigctx->trapno; 377 mctx->__gregs[_REG_ES] = tf->tf_es & 0xFFFF; 378 mctx->__gregs[_REG_DS] = tf->tf_ds & 0xFFFF; 379 mctx->__gregs[_REG_RSP] = lsigctx->rsp; /* XXX */ 380 mctx->__gregs[_REG_SS] = tf->tf_ss & 0xFFFF; 381 382 /* 383 * FPU state 384 */ 385 if (lsigctx->fpstate != NULL) { 386 /* Both structures match the fxstate data */ 387 error = copyin(lsigctx->fpstate, fxarea, sizeof(*fxarea)); 388 if (error != 0) { 389 mutex_enter(l->l_proc->p_lock); 390 sigexit(l, SIGILL); 391 return error; 392 } 393 } 394 395 /* 396 * And the stack 397 */ 398 uctx.uc_stack.ss_flags = 0; 399 if (luctx->luc_stack.ss_flags & LINUX_SS_ONSTACK) 400 uctx.uc_stack.ss_flags |= SS_ONSTACK; 401 402 if (luctx->luc_stack.ss_flags & LINUX_SS_DISABLE) 403 uctx.uc_stack.ss_flags |= SS_DISABLE; 404 405 uctx.uc_stack.ss_sp = luctx->luc_stack.ss_sp; 406 uctx.uc_stack.ss_size = luctx->luc_stack.ss_size; 407 408 /* 409 * And let setucontext deal with that. 410 */ 411 mutex_enter(l->l_proc->p_lock); 412 error = setucontext(l, &uctx); 413 mutex_exit(l->l_proc->p_lock); 414 if (error) 415 return error; 416 417 return EJUSTRETURN; 418 } 419 420 int 421 linux_sys_arch_prctl(struct lwp *l, 422 const struct linux_sys_arch_prctl_args *uap, register_t *retval) 423 { 424 /* { 425 syscallarg(int) code; 426 syscallarg(unsigned long) addr; 427 } */ 428 void *addr = (void *)SCARG(uap, addr); 429 430 switch(SCARG(uap, code)) { 431 case LINUX_ARCH_SET_GS: 432 return x86_set_sdbase(addr, 'g', l, true); 433 434 case LINUX_ARCH_GET_GS: 435 return x86_get_sdbase(addr, 'g'); 436 437 case LINUX_ARCH_SET_FS: 438 return x86_set_sdbase(addr, 'f', l, true); 439 440 case LINUX_ARCH_GET_FS: 441 return x86_get_sdbase(addr, 'f'); 442 443 default: 444 #ifdef DEBUG_LINUX 445 printf("linux_sys_arch_prctl: unexpected code %d\n", 446 SCARG(uap, code)); 447 #endif 448 return EINVAL; 449 } 450 /* NOTREACHED */ 451 } 452 453 const int linux_vsyscall_to_syscall[] = { 454 LINUX_SYS_gettimeofday, 455 LINUX_SYS_time, 456 LINUX_SYS_nosys, /* nosys */ 457 LINUX_SYS_nosys, /* nosys */ 458 }; 459 460 int 461 linux_usertrap(struct lwp *l, vaddr_t trapaddr, void *arg) 462 { 463 struct trapframe *tf = arg; 464 uint64_t retaddr; 465 size_t vsyscallnr; 466 467 /* 468 * Check for a vsyscall. %rip must be the fault address, 469 * and the address must be in the Linux vsyscall area. 470 * Also, vsyscalls are only done at 1024-byte boundaries. 471 */ 472 473 if (__predict_true(trapaddr < LINUX_VSYSCALL_START)) 474 return 0; 475 476 if (trapaddr != tf->tf_rip) 477 return 0; 478 479 if ((tf->tf_rip & (LINUX_VSYSCALL_SIZE - 1)) != 0) 480 return 0; 481 482 vsyscallnr = (tf->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SIZE; 483 484 if (vsyscallnr > LINUX_VSYSCALL_MAXNR) 485 return 0; 486 487 /* 488 * Get the return address from the top of the stack, 489 * and fix up the return address. 490 * This assumes the faulting instruction was callq *reg, 491 * which is the only way that vsyscalls are ever entered. 492 */ 493 if (copyin((void *)tf->tf_rsp, &retaddr, sizeof retaddr) != 0) 494 return 0; 495 if ((vaddr_t)retaddr >= VM_MAXUSER_ADDRESS) 496 return 0; 497 tf->tf_rip = retaddr; 498 tf->tf_rax = linux_vsyscall_to_syscall[vsyscallnr]; 499 tf->tf_rsp += 8; /* "pop" the return address */ 500 501 #if 0 502 printf("usertrap: rip %p rsp %p retaddr %p vsys %d sys %d\n", 503 (void *)tf->tf_rip, (void *)tf->tf_rsp, (void *)retaddr, 504 vsyscallnr, (int)tf->tf_rax); 505 #endif 506 507 (*l->l_proc->p_md.md_syscall)(tf); 508 509 return 1; 510 } 511 512 static void 513 linux_buildcontext(struct lwp *l, void *catcher, void *f) 514 { 515 struct trapframe *tf = l->l_md.md_regs; 516 517 tf->tf_ds = GSEL(GUDATA_SEL, SEL_UPL); 518 tf->tf_rip = (u_int64_t)catcher; 519 tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL); 520 tf->tf_rflags &= ~PSL_CLEARSIG; 521 tf->tf_rsp = (u_int64_t)f; 522 tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL); 523 } 524