Home | History | Annotate | Line # | Download | only in i386
linux_machdep.c revision 1.18
      1 /*	$NetBSD: linux_machdep.c,v 1.18 1995/09/08 07:57:15 fvdl Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1995 Frank van der Linden
      5  * All rights reserved.
      6  *
      7  * Redistribution and use in source and binary forms, with or without
      8  * modification, are permitted provided that the following conditions
      9  * are met:
     10  * 1. Redistributions of source code must retain the above copyright
     11  *    notice, this list of conditions and the following disclaimer.
     12  * 2. Redistributions in binary form must reproduce the above copyright
     13  *    notice, this list of conditions and the following disclaimer in the
     14  *    documentation and/or other materials provided with the distribution.
     15  * 3. All advertising materials mentioning features or use of this software
     16  *    must display the following acknowledgement:
     17  *      This product includes software developed for the NetBSD Project
     18  *      by Frank van der Linden
     19  * 4. The name of the author may not be used to endorse or promote products
     20  *    derived from this software without specific prior written permission
     21  *
     22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
     23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     25  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
     26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
     27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
     28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
     29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
     30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
     31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     32  */
     33 
     34 #include <sys/param.h>
     35 #include <sys/systm.h>
     36 #include <sys/signalvar.h>
     37 #include <sys/kernel.h>
     38 #include <sys/map.h>
     39 #include <sys/proc.h>
     40 #include <sys/user.h>
     41 #include <sys/buf.h>
     42 #include <sys/reboot.h>
     43 #include <sys/conf.h>
     44 #include <sys/file.h>
     45 #include <sys/callout.h>
     46 #include <sys/malloc.h>
     47 #include <sys/mbuf.h>
     48 #include <sys/msgbuf.h>
     49 #include <sys/mount.h>
     50 #include <sys/vnode.h>
     51 #include <sys/device.h>
     52 #include <sys/sysctl.h>
     53 #include <sys/syscallargs.h>
     54 #include <sys/filedesc.h>
     55 
     56 #include <compat/linux/linux_types.h>
     57 #include <compat/linux/linux_signal.h>
     58 #include <compat/linux/linux_syscallargs.h>
     59 #include <compat/linux/linux_util.h>
     60 
     61 #include <machine/cpu.h>
     62 #include <machine/cpufunc.h>
     63 #include <machine/psl.h>
     64 #include <machine/reg.h>
     65 #include <machine/segments.h>
     66 #include <machine/specialreg.h>
     67 #include <machine/sysarch.h>
     68 #include <machine/linux_machdep.h>
     69 
     70 /*
     71  * To see whether pcvt is configured (for virtual console ioctl calls).
     72  */
     73 #include "vt.h"
     74 #if NVT > 0
     75 #include <arch/i386/isa/pcvt/pcvt_ioctl.h>
     76 #endif
     77 
     78 /*
     79  * Deal with some i386-specific things in the Linux emulation code.
     80  * This means just signals for now, will include stuff like
     81  * I/O map permissions and V86 mode sometime.
     82  */
     83 
     84 /*
     85  * Send an interrupt to process.
     86  *
     87  * Stack is set up to allow sigcode stored
     88  * in u. to call routine, followed by kcall
     89  * to sigreturn routine below.  After sigreturn
     90  * resets the signal mask, the stack, and the
     91  * frame pointer, it returns to the user
     92  * specified pc, psl.
     93  */
     94 
     95 void
     96 linux_sendsig(catcher, sig, mask, code)
     97 	sig_t catcher;
     98 	int sig, mask;
     99 	u_long code;
    100 {
    101 	register struct proc *p = curproc;
    102 	register struct trapframe *tf;
    103 	struct linux_sigframe *fp, frame;
    104 	struct sigacts *psp = p->p_sigacts;
    105 	int oonstack;
    106 	extern char linux_sigcode[], linux_esigcode[];
    107 
    108 	tf = p->p_md.md_regs;
    109 	oonstack = psp->ps_sigstk.ss_flags & SS_ONSTACK;
    110 
    111 	/*
    112 	 * Allocate space for the signal handler context.
    113 	 */
    114 	if ((psp->ps_flags & SAS_ALTSTACK) && !oonstack &&
    115 	    (psp->ps_sigonstack & sigmask(sig))) {
    116 		fp = (struct linux_sigframe *)(psp->ps_sigstk.ss_base +
    117 		    psp->ps_sigstk.ss_size - sizeof(struct linux_sigframe));
    118 		psp->ps_sigstk.ss_flags |= SS_ONSTACK;
    119 	} else {
    120 		fp = (struct linux_sigframe *)tf->tf_esp - 1;
    121 	}
    122 
    123 	frame.sf_handler = catcher;
    124 	frame.sf_sig = bsd_to_linux_sig[sig];
    125 
    126 	/*
    127 	 * Build the signal context to be used by sigreturn.
    128 	 */
    129 	frame.sf_sc.sc_mask   = mask;
    130 #ifdef VM86
    131 	if (tf->tf_eflags & PSL_VM) {
    132 		frame.sf_sc.sc_gs = tf->tf_vm86_gs;
    133 		frame.sf_sc.sc_fs = tf->tf_vm86_fs;
    134 		frame.sf_sc.sc_es = tf->tf_vm86_es;
    135 		frame.sf_sc.sc_ds = tf->tf_vm86_ds;
    136 	} else
    137 #endif
    138 	{
    139 		__asm("movl %%gs,%w0" : "=r" (frame.sf_sc.sc_gs));
    140 		__asm("movl %%fs,%w0" : "=r" (frame.sf_sc.sc_fs));
    141 		frame.sf_sc.sc_es = tf->tf_es;
    142 		frame.sf_sc.sc_ds = tf->tf_ds;
    143 	}
    144 	frame.sf_sc.sc_edi    = tf->tf_edi;
    145 	frame.sf_sc.sc_esi    = tf->tf_esi;
    146 	frame.sf_sc.sc_ebp    = tf->tf_ebp;
    147 	frame.sf_sc.sc_ebx    = tf->tf_ebx;
    148 	frame.sf_sc.sc_edx    = tf->tf_edx;
    149 	frame.sf_sc.sc_ecx    = tf->tf_ecx;
    150 	frame.sf_sc.sc_eax    = tf->tf_eax;
    151 	frame.sf_sc.sc_eip    = tf->tf_eip;
    152 	frame.sf_sc.sc_cs     = tf->tf_cs;
    153 	frame.sf_sc.sc_eflags = tf->tf_eflags;
    154 	frame.sf_sc.sc_esp_at_signal = tf->tf_esp;
    155 	frame.sf_sc.sc_ss     = tf->tf_ss;
    156 	frame.sf_sc.sc_err    = tf->tf_err;
    157 	frame.sf_sc.sc_trapno = tf->tf_trapno;
    158 
    159 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
    160 		/*
    161 		 * Process has trashed its stack; give it an illegal
    162 		 * instruction to halt it in its tracks.
    163 		 */
    164 		sigexit(p, SIGILL);
    165 		/* NOTREACHED */
    166 	}
    167 
    168 	/*
    169 	 * Build context to run handler in.
    170 	 */
    171 	tf->tf_esp = (int)fp;
    172 	tf->tf_eip = (int)(((char *)PS_STRINGS) -
    173 	     (linux_esigcode - linux_sigcode));
    174 #ifdef VM86
    175 	tf->tf_eflags &= ~PSL_VM;
    176 #endif
    177 	tf->tf_cs = LSEL(LUCODE_SEL, SEL_UPL);
    178 	tf->tf_ds = LSEL(LUDATA_SEL, SEL_UPL);
    179 	tf->tf_es = LSEL(LUDATA_SEL, SEL_UPL);
    180 	tf->tf_ss = LSEL(LUDATA_SEL, SEL_UPL);
    181 }
    182 
    183 /*
    184  * System call to cleanup state after a signal
    185  * has been taken.  Reset signal mask and
    186  * stack state from context left by sendsig (above).
    187  * Return to previous pc and psl as specified by
    188  * context left by sendsig. Check carefully to
    189  * make sure that the user has not modified the
    190  * psl to gain improper privileges or to cause
    191  * a machine fault.
    192  */
    193 int
    194 linux_sigreturn(p, uap, retval)
    195 	struct proc *p;
    196 	struct linux_sigreturn_args /* {
    197 		syscallarg(struct linux_sigcontext *) scp;
    198 	} */ *uap;
    199 	register_t *retval;
    200 {
    201 	struct linux_sigcontext *scp, context;
    202 	register struct trapframe *tf;
    203 
    204 	tf = p->p_md.md_regs;
    205 
    206 	/*
    207 	 * The trampoline code hands us the context.
    208 	 * It is unsafe to keep track of it ourselves, in the event that a
    209 	 * program jumps out of a signal handler.
    210 	 */
    211 	scp = SCARG(uap, scp);
    212 	if (copyin((caddr_t)scp, &context, sizeof(*scp)) != 0)
    213 		return (EFAULT);
    214 
    215 	/*
    216 	 * Check for security violations.
    217 	 */
    218 	if (((context.sc_eflags ^ tf->tf_eflags) & PSL_USERSTATIC) != 0 ||
    219 	    ISPL(context.sc_cs) != SEL_UPL)
    220 		return (EINVAL);
    221 
    222 	p->p_sigacts->ps_sigstk.ss_flags &= ~SS_ONSTACK;
    223 	p->p_sigmask = context.sc_mask & ~sigcantmask;
    224 
    225 	/*
    226 	 * Restore signal context.
    227 	 */
    228 #ifdef VM86
    229 	if (context.sc_eflags & PSL_VM) {
    230 		tf->tf_vm86_gs = context.sc_gs;
    231 		tf->tf_vm86_fs = context.sc_fs;
    232 		tf->tf_vm86_es = context.sc_es;
    233 		tf->tf_vm86_ds = context.sc_ds;
    234 	} else
    235 #endif
    236 	{
    237 		/* %fs and %gs were restored by the trampoline. */
    238 		tf->tf_es = context.sc_es;
    239 		tf->tf_ds = context.sc_ds;
    240 	}
    241 	tf->tf_edi    = context.sc_edi;
    242 	tf->tf_esi    = context.sc_esi;
    243 	tf->tf_ebp    = context.sc_ebp;
    244 	tf->tf_ebx    = context.sc_ebx;
    245 	tf->tf_edx    = context.sc_edx;
    246 	tf->tf_ecx    = context.sc_ecx;
    247 	tf->tf_eax    = context.sc_eax;
    248 	tf->tf_eip    = context.sc_eip;
    249 	tf->tf_cs     = context.sc_cs;
    250 	tf->tf_eflags = context.sc_eflags;
    251 	tf->tf_esp    = context.sc_esp_at_signal;
    252 	tf->tf_ss     = context.sc_ss;
    253 
    254 	return (EJUSTRETURN);
    255 }
    256 
    257 #ifdef USER_LDT
    258 
    259 int
    260 linux_read_ldt(p, uap, retval)
    261 	struct proc *p;
    262 	struct linux_modify_ldt_args /* {
    263 		syscallarg(int) func;
    264 		syscallarg(void *) ptr;
    265 		syscallarg(size_t) bytecount;
    266 	} */ *uap;
    267 	register_t *retval;
    268 {
    269 	struct i386_get_ldt_args gl;
    270 	int error;
    271 	caddr_t sg;
    272 	char *parms;
    273 
    274 	sg = stackgap_init(p->p_emul);
    275 
    276 	gl.start = 0;
    277 	gl.desc = SCARG(uap, ptr);
    278 	gl.num = SCARG(uap, bytecount) / sizeof(union descriptor);
    279 
    280 	parms = stackgap_alloc(&sg, sizeof(gl));
    281 
    282 	if (error = copyout(&gl, parms, sizeof(gl)))
    283 		return (error);
    284 
    285 	if (error = i386_get_ldt(p, parms, retval))
    286 		return (error);
    287 
    288 	*retval *= sizeof(union descriptor);
    289 	return (0);
    290 }
    291 
    292 struct linux_ldt_info {
    293 	u_int entry_number;
    294 	u_long base_addr;
    295 	u_int limit;
    296 	u_int seg_32bit:1;
    297 	u_int contents:2;
    298 	u_int read_exec_only:1;
    299 	u_int limit_in_pages:1;
    300 	u_int seg_not_present:1;
    301 };
    302 
    303 int
    304 linux_write_ldt(p, uap, retval)
    305 	struct proc *p;
    306 	struct linux_modify_ldt_args /* {
    307 		syscallarg(int) func;
    308 		syscallarg(void *) ptr;
    309 		syscallarg(size_t) bytecount;
    310 	} */ *uap;
    311 	register_t *retval;
    312 {
    313 	struct linux_ldt_info ldt_info;
    314 	struct segment_descriptor sd;
    315 	struct i386_set_ldt_args sl;
    316 	int error;
    317 	caddr_t sg;
    318 	char *parms;
    319 
    320 	if (SCARG(uap, bytecount) != sizeof(ldt_info))
    321 		return (EINVAL);
    322 	if (error = copyin(SCARG(uap, ptr), &ldt_info, sizeof(ldt_info)))
    323 		return error;
    324 	if (ldt_info.contents == 3)
    325 		return (EINVAL);
    326 
    327 	sg = stackgap_init(p->p_emul);
    328 
    329 	sd.sd_lobase = ldt_info.base_addr & 0xffffff;
    330 	sd.sd_hibase = (ldt_info.base_addr >> 24) & 0xff;
    331 	sd.sd_lolimit = ldt_info.limit & 0xffff;
    332 	sd.sd_hilimit = (ldt_info.limit >> 16) & 0xf;
    333 	sd.sd_type =
    334 	    16 | (ldt_info.contents << 2) | (!ldt_info.read_exec_only << 1);
    335 	sd.sd_dpl = SEL_UPL;
    336 	sd.sd_p = !ldt_info.seg_not_present;
    337 	sd.sd_def32 = ldt_info.seg_32bit;
    338 	sd.sd_gran = ldt_info.limit_in_pages;
    339 
    340 	sl.start = ldt_info.entry_number;
    341 	sl.desc = stackgap_alloc(&sg, sizeof(sd));
    342 	sl.num = 1;
    343 
    344 #if 0
    345 	printf("linux_write_ldt: idx=%d, base=%x, limit=%x\n",
    346 	    ldt_info.entry_number, ldt_info.base_addr, ldt_info.limit);
    347 #endif
    348 
    349 	parms = stackgap_alloc(&sg, sizeof(sl));
    350 
    351 	if (error = copyout(&sd, sl.desc, sizeof(sd)))
    352 		return (error);
    353 	if (error = copyout(&sl, parms, sizeof(sl)))
    354 		return (error);
    355 
    356 	if (error = i386_set_ldt(p, parms, retval))
    357 		return (error);
    358 
    359 	*retval = 0;
    360 	return (0);
    361 }
    362 
    363 #endif /* USER_LDT */
    364 
    365 int
    366 linux_modify_ldt(p, uap, retval)
    367 	struct proc *p;
    368 	struct linux_modify_ldt_args /* {
    369 		syscallarg(int) func;
    370 		syscallarg(void *) ptr;
    371 		syscallarg(size_t) bytecount;
    372 	} */ *uap;
    373 	register_t *retval;
    374 {
    375 
    376 	switch (SCARG(uap, func)) {
    377 #ifdef USER_LDT
    378 	case 0:
    379 		return (linux_read_ldt(p, uap, retval));
    380 
    381 	case 1:
    382 		return (linux_write_ldt(p, uap, retval));
    383 #endif /* USER_LDT */
    384 
    385 	default:
    386 		return (ENOSYS);
    387 	}
    388 }
    389 
    390 /*
    391  * XXX Pathetic hack to make svgalib work. This will fake the major
    392  * device number of an opened VT so that svgalib likes it. grmbl.
    393  * Should probably do it 'wrong the right way' and use a mapping
    394  * array for all major device numbers, and map linux_mknod too.
    395  */
    396 dev_t
    397 linux_fakedev(dev)
    398 	dev_t dev;
    399 {
    400 	if (major(dev) == NETBSD_CONS_MAJOR)
    401 		return makedev(LINUX_CONS_MAJOR, (minor(dev) + 1));
    402 	return dev;
    403 }
    404 
    405 /*
    406  * We come here in a last attempt to satisfy a Linux ioctl() call
    407  */
    408 int
    409 linux_machdepioctl(p, uap, retval)
    410 	struct proc *p;
    411 	struct linux_ioctl_args /* {
    412 		syscallarg(int) fd;
    413 		syscallarg(u_long) com;
    414 		syscallarg(caddr_t) data;
    415 	} */ *uap;
    416 	register_t *retval;
    417 {
    418 	struct ioctl_args bia, tmparg;
    419 	u_long com;
    420 #if NVT > 0
    421 	int error, mode;
    422 	struct vt_mode lvt;
    423 	caddr_t bvtp, sg;
    424 	u_int fd;
    425 	struct file *fp;
    426 	struct filedesc *fdp;
    427 #endif
    428 
    429 	SCARG(&bia, fd) = SCARG(uap, fd);
    430 	SCARG(&bia, data) = SCARG(uap, data);
    431 	com = SCARG(uap, com);
    432 
    433 	switch (com) {
    434 #if NVT > 0
    435 	case LINUX_KDGKBMODE:
    436 		/*
    437 		 * Could be implemented but somehow KDGKBMODE is the
    438 		 * same as KBDGLEDS.
    439 		 */
    440 		mode = K_XLATE;
    441 		return copyout(&mode, SCARG(uap, data), sizeof mode);
    442 	case LINUX_KDSKBMODE:
    443 		com = KDSKBMODE;
    444 		if ((unsigned)SCARG(uap, data) == LINUX_K_MEDIUMRAW)
    445 			SCARG(&bia, data) = (caddr_t)K_RAW;
    446 		break;
    447 	case LINUX_KDMKTONE:
    448 		com = KDMKTONE;
    449 		break;
    450 	case LINUX_KDSETMODE:
    451 		com = KDSETMODE;
    452 		break;
    453 	case LINUX_KDENABIO:
    454 		com = KDENABIO;
    455 		break;
    456 	case LINUX_KDDISABIO:
    457 		com = KDDISABIO;
    458 		break;
    459 	case LINUX_KDGETLED:
    460 		com = KDGETLED;
    461 		break;
    462 	case LINUX_KDSETLED:
    463 		com = KDSETLED;
    464 		break;
    465 	case LINUX_VT_OPENQRY:
    466 		com = VT_OPENQRY;
    467 		break;
    468 	case LINUX_VT_GETMODE:
    469 		SCARG(&bia, com) = VT_GETMODE;
    470 		if ((error = ioctl(p, &bia, retval)))
    471 			return error;
    472 		if ((error = copyin(SCARG(uap, data), (caddr_t)&lvt,
    473 		    sizeof (struct vt_mode))))
    474 			return error;
    475 		lvt.relsig = bsd_to_linux_sig[lvt.relsig];
    476 		lvt.acqsig = bsd_to_linux_sig[lvt.acqsig];
    477 		lvt.frsig = bsd_to_linux_sig[lvt.frsig];
    478 		return copyout((caddr_t)&lvt, SCARG(uap, data),
    479 		    sizeof (struct vt_mode));
    480 	case LINUX_VT_SETMODE:
    481 		com = VT_SETMODE;
    482 		if ((error = copyin(SCARG(uap, data), (caddr_t)&lvt,
    483 		    sizeof (struct vt_mode))))
    484 			return error;
    485 		lvt.relsig = linux_to_bsd_sig[lvt.relsig];
    486 		lvt.acqsig = linux_to_bsd_sig[lvt.acqsig];
    487 		lvt.frsig = linux_to_bsd_sig[lvt.frsig];
    488 		sg = stackgap_init(p->p_emul);
    489 		bvtp = stackgap_alloc(&sg, sizeof (struct vt_mode));
    490 		if ((error = copyout(&lvt, bvtp, sizeof (struct vt_mode))))
    491 			return error;
    492 		SCARG(&bia, data) = bvtp;
    493 		break;
    494 	case LINUX_VT_RELDISP:
    495 		com = VT_RELDISP;
    496 		break;
    497 	case LINUX_VT_ACTIVATE:
    498 		com = VT_ACTIVATE;
    499 		break;
    500 	case LINUX_VT_WAITACTIVE:
    501 		com = VT_WAITACTIVE;
    502 		break;
    503 #endif
    504 	default:
    505 		return EINVAL;
    506 	}
    507 	SCARG(&bia, com) = com;
    508 	return ioctl(p, &bia, retval);
    509 }
    510 
    511 /*
    512  * Set I/O permissions for a process. Just set the maximum level
    513  * right away (ignoring the argument), otherwise we would have
    514  * to rely on I/O permission maps, which are not implemented.
    515  */
    516 int
    517 linux_iopl(p, uap, retval)
    518 	struct proc *p;
    519 	struct linux_iopl_args /* {
    520 		syscallarg(int) level;
    521 	} */ *uap;
    522 	register_t *retval;
    523 {
    524 	struct trapframe *fp = p->p_md.md_regs;
    525 
    526 	if (suser(p->p_ucred, &p->p_acflag) != 0)
    527 		return EPERM;
    528 	fp->tf_eflags |= PSL_IOPL;
    529 	*retval = 0;
    530 	return 0;
    531 }
    532 
    533 /*
    534  * See above. If a root process tries to set access to an I/O port,
    535  * just let it have the whole range.
    536  */
    537 int
    538 linux_ioperm(p, uap, retval)
    539 	struct proc *p;
    540 	struct linux_ioperm_args /* {
    541 		syscallarg(unsigned int) lo;
    542 		syscallarg(unsigned int) hi;
    543 		syscallarg(int) val;
    544 	} */ *uap;
    545 	register_t *retval;
    546 {
    547 	struct trapframe *fp = p->p_md.md_regs;
    548 
    549 	if (suser(p->p_ucred, &p->p_acflag) != 0)
    550 		return EPERM;
    551 	if (SCARG(uap, val))
    552 		fp->tf_eflags |= PSL_IOPL;
    553 	*retval = 0;
    554 	return 0;
    555 }
    556