xref: /netbsd-src/sys/compat/linux/arch/i386/linux_machdep.c (revision 81b108b45f75f89f1e3ffad9fb6f074e771c0935)
1 /*	$NetBSD: linux_machdep.c,v 1.29 1996/05/03 19:42:11 christos 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 #include <compat/linux/linux_ioctl.h>
61 
62 #include <machine/cpu.h>
63 #include <machine/cpufunc.h>
64 #include <machine/psl.h>
65 #include <machine/reg.h>
66 #include <machine/segments.h>
67 #include <machine/specialreg.h>
68 #include <machine/sysarch.h>
69 #include <machine/vm86.h>
70 #include <machine/linux_machdep.h>
71 
72 /*
73  * To see whether pcvt is configured (for virtual console ioctl calls).
74  */
75 #include "vt.h"
76 #if NVT > 0
77 #include <arch/i386/isa/pcvt/pcvt_ioctl.h>
78 #endif
79 
80 #ifdef USER_LDT
81 #include <machine/cpu.h>
82 int linux_read_ldt __P((struct proc *, struct linux_sys_modify_ldt_args *,
83     register_t *));
84 int linux_write_ldt __P((struct proc *, struct linux_sys_modify_ldt_args *,
85     register_t *));
86 #endif
87 
88 /*
89  * Deal with some i386-specific things in the Linux emulation code.
90  * This means just signals for now, will include stuff like
91  * I/O map permissions and V86 mode sometime.
92  */
93 
94 /*
95  * Send an interrupt to process.
96  *
97  * Stack is set up to allow sigcode stored
98  * in u. to call routine, followed by kcall
99  * to sigreturn routine below.  After sigreturn
100  * resets the signal mask, the stack, and the
101  * frame pointer, it returns to the user
102  * specified pc, psl.
103  */
104 
105 void
106 linux_sendsig(catcher, sig, mask, code)
107 	sig_t catcher;
108 	int sig, mask;
109 	u_long code;
110 {
111 	register struct proc *p = curproc;
112 	register struct trapframe *tf;
113 	struct linux_sigframe *fp, frame;
114 	struct sigacts *psp = p->p_sigacts;
115 	int oonstack;
116 	extern char linux_sigcode[], linux_esigcode[];
117 
118 	tf = p->p_md.md_regs;
119 	oonstack = psp->ps_sigstk.ss_flags & SS_ONSTACK;
120 
121 	/*
122 	 * Allocate space for the signal handler context.
123 	 */
124 	if ((psp->ps_flags & SAS_ALTSTACK) && !oonstack &&
125 	    (psp->ps_sigonstack & sigmask(sig))) {
126 		fp = (struct linux_sigframe *)(psp->ps_sigstk.ss_sp +
127 		    psp->ps_sigstk.ss_size - sizeof(struct linux_sigframe));
128 		psp->ps_sigstk.ss_flags |= SS_ONSTACK;
129 	} else {
130 		fp = (struct linux_sigframe *)tf->tf_esp - 1;
131 	}
132 
133 	frame.sf_handler = catcher;
134 	frame.sf_sig = bsd_to_linux_sig[sig];
135 
136 	/*
137 	 * Build the signal context to be used by sigreturn.
138 	 */
139 	frame.sf_sc.sc_mask   = mask;
140 #ifdef VM86
141 	if (tf->tf_eflags & PSL_VM) {
142 		frame.sf_sc.sc_gs = tf->tf_vm86_gs;
143 		frame.sf_sc.sc_fs = tf->tf_vm86_fs;
144 		frame.sf_sc.sc_es = tf->tf_vm86_es;
145 		frame.sf_sc.sc_ds = tf->tf_vm86_ds;
146 		frame.sf_sc.sc_eflags = get_vflags(p);
147 	} else
148 #endif
149 	{
150 		__asm("movl %%gs,%w0" : "=r" (frame.sf_sc.sc_gs));
151 		__asm("movl %%fs,%w0" : "=r" (frame.sf_sc.sc_fs));
152 		frame.sf_sc.sc_es = tf->tf_es;
153 		frame.sf_sc.sc_ds = tf->tf_ds;
154 		frame.sf_sc.sc_eflags = tf->tf_eflags;
155 	}
156 	frame.sf_sc.sc_edi = tf->tf_edi;
157 	frame.sf_sc.sc_esi = tf->tf_esi;
158 	frame.sf_sc.sc_ebp = tf->tf_ebp;
159 	frame.sf_sc.sc_ebx = tf->tf_ebx;
160 	frame.sf_sc.sc_edx = tf->tf_edx;
161 	frame.sf_sc.sc_ecx = tf->tf_ecx;
162 	frame.sf_sc.sc_eax = tf->tf_eax;
163 	frame.sf_sc.sc_eip = tf->tf_eip;
164 	frame.sf_sc.sc_cs = tf->tf_cs;
165 	frame.sf_sc.sc_esp_at_signal = tf->tf_esp;
166 	frame.sf_sc.sc_ss = tf->tf_ss;
167 	frame.sf_sc.sc_err = tf->tf_err;
168 	frame.sf_sc.sc_trapno = tf->tf_trapno;
169 
170 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
171 		/*
172 		 * Process has trashed its stack; give it an illegal
173 		 * instruction to halt it in its tracks.
174 		 */
175 		sigexit(p, SIGILL);
176 		/* NOTREACHED */
177 	}
178 
179 	/*
180 	 * Build context to run handler in.
181 	 */
182 	tf->tf_es = GSEL(GUDATA_SEL, SEL_UPL);
183 	tf->tf_ds = GSEL(GUDATA_SEL, SEL_UPL);
184 	tf->tf_eip = (int)(((char *)PS_STRINGS) -
185 	     (linux_esigcode - linux_sigcode));
186 	tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL);
187 	tf->tf_eflags &= ~(PSL_T|PSL_VM|PSL_AC);
188 	tf->tf_esp = (int)fp;
189 	tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL);
190 }
191 
192 /*
193  * System call to cleanup state after a signal
194  * has been taken.  Reset signal mask and
195  * stack state from context left by sendsig (above).
196  * Return to previous pc and psl as specified by
197  * context left by sendsig. Check carefully to
198  * make sure that the user has not modified the
199  * psl to gain improper privileges or to cause
200  * a machine fault.
201  */
202 int
203 linux_sys_sigreturn(p, v, retval)
204 	struct proc *p;
205 	void *v;
206 	register_t *retval;
207 {
208 	struct linux_sys_sigreturn_args /* {
209 		syscallarg(struct linux_sigcontext *) scp;
210 	} */ *uap = v;
211 	struct linux_sigcontext *scp, context;
212 	register struct trapframe *tf;
213 
214 	tf = p->p_md.md_regs;
215 
216 	/*
217 	 * The trampoline code hands us the context.
218 	 * It is unsafe to keep track of it ourselves, in the event that a
219 	 * program jumps out of a signal handler.
220 	 */
221 	scp = SCARG(uap, scp);
222 	if (copyin((caddr_t)scp, &context, sizeof(*scp)) != 0)
223 		return (EFAULT);
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 		set_vflags(p, context.sc_eflags);
235 	} else
236 #endif
237 	{
238 		/*
239 		 * Check for security violations.  If we're returning to
240 		 * protected mode, the CPU will validate the segment registers
241 		 * automatically and generate a trap on violations.  We handle
242 		 * the trap, rather than doing all of the checking here.
243 		 */
244 		if (((context.sc_eflags ^ tf->tf_eflags) & PSL_USERSTATIC) != 0 ||
245 		    !USERMODE(context.sc_cs, context.sc_eflags))
246 			return (EINVAL);
247 
248 		/* %fs and %gs were restored by the trampoline. */
249 		tf->tf_es = context.sc_es;
250 		tf->tf_ds = context.sc_ds;
251 		tf->tf_eflags = context.sc_eflags;
252 	}
253 	tf->tf_edi = context.sc_edi;
254 	tf->tf_esi = context.sc_esi;
255 	tf->tf_ebp = context.sc_ebp;
256 	tf->tf_ebx = context.sc_ebx;
257 	tf->tf_edx = context.sc_edx;
258 	tf->tf_ecx = context.sc_ecx;
259 	tf->tf_eax = context.sc_eax;
260 	tf->tf_eip = context.sc_eip;
261 	tf->tf_cs = context.sc_cs;
262 	tf->tf_esp = context.sc_esp_at_signal;
263 	tf->tf_ss = context.sc_ss;
264 
265 	p->p_sigacts->ps_sigstk.ss_flags &= ~SS_ONSTACK;
266 	p->p_sigmask = context.sc_mask & ~sigcantmask;
267 
268 	return (EJUSTRETURN);
269 }
270 
271 #ifdef USER_LDT
272 
273 int
274 linux_read_ldt(p, uap, retval)
275 	struct proc *p;
276 	struct linux_sys_modify_ldt_args /* {
277 		syscallarg(int) func;
278 		syscallarg(void *) ptr;
279 		syscallarg(size_t) bytecount;
280 	} */ *uap;
281 	register_t *retval;
282 {
283 	struct i386_get_ldt_args gl;
284 	int error;
285 	caddr_t sg;
286 	char *parms;
287 
288 	sg = stackgap_init(p->p_emul);
289 
290 	gl.start = 0;
291 	gl.desc = SCARG(uap, ptr);
292 	gl.num = SCARG(uap, bytecount) / sizeof(union descriptor);
293 
294 	parms = stackgap_alloc(&sg, sizeof(gl));
295 
296 	if ((error = copyout(&gl, parms, sizeof(gl))) != 0)
297 		return (error);
298 
299 	if ((error = i386_get_ldt(p, parms, retval)) != 0)
300 		return (error);
301 
302 	*retval *= sizeof(union descriptor);
303 	return (0);
304 }
305 
306 struct linux_ldt_info {
307 	u_int entry_number;
308 	u_long base_addr;
309 	u_int limit;
310 	u_int seg_32bit:1;
311 	u_int contents:2;
312 	u_int read_exec_only:1;
313 	u_int limit_in_pages:1;
314 	u_int seg_not_present:1;
315 };
316 
317 int
318 linux_write_ldt(p, uap, retval)
319 	struct proc *p;
320 	struct linux_sys_modify_ldt_args /* {
321 		syscallarg(int) func;
322 		syscallarg(void *) ptr;
323 		syscallarg(size_t) bytecount;
324 	} */ *uap;
325 	register_t *retval;
326 {
327 	struct linux_ldt_info ldt_info;
328 	struct segment_descriptor sd;
329 	struct i386_set_ldt_args sl;
330 	int error;
331 	caddr_t sg;
332 	char *parms;
333 
334 	if (SCARG(uap, bytecount) != sizeof(ldt_info))
335 		return (EINVAL);
336 	if ((error = copyin(SCARG(uap, ptr), &ldt_info, sizeof(ldt_info))) != 0)
337 		return error;
338 	if (ldt_info.contents == 3)
339 		return (EINVAL);
340 
341 	sg = stackgap_init(p->p_emul);
342 
343 	sd.sd_lobase = ldt_info.base_addr & 0xffffff;
344 	sd.sd_hibase = (ldt_info.base_addr >> 24) & 0xff;
345 	sd.sd_lolimit = ldt_info.limit & 0xffff;
346 	sd.sd_hilimit = (ldt_info.limit >> 16) & 0xf;
347 	sd.sd_type =
348 	    16 | (ldt_info.contents << 2) | (!ldt_info.read_exec_only << 1);
349 	sd.sd_dpl = SEL_UPL;
350 	sd.sd_p = !ldt_info.seg_not_present;
351 	sd.sd_def32 = ldt_info.seg_32bit;
352 	sd.sd_gran = ldt_info.limit_in_pages;
353 
354 	sl.start = ldt_info.entry_number;
355 	sl.desc = stackgap_alloc(&sg, sizeof(sd));
356 	sl.num = 1;
357 
358 #if 0
359 	printf("linux_write_ldt: idx=%d, base=%x, limit=%x\n",
360 	    ldt_info.entry_number, ldt_info.base_addr, ldt_info.limit);
361 #endif
362 
363 	parms = stackgap_alloc(&sg, sizeof(sl));
364 
365 	if ((error = copyout(&sd, sl.desc, sizeof(sd))) != 0)
366 		return (error);
367 	if ((error = copyout(&sl, parms, sizeof(sl))) != 0)
368 		return (error);
369 
370 	if ((error = i386_set_ldt(p, parms, retval)) != 0)
371 		return (error);
372 
373 	*retval = 0;
374 	return (0);
375 }
376 
377 #endif /* USER_LDT */
378 
379 int
380 linux_sys_modify_ldt(p, v, retval)
381 	struct proc *p;
382 	void *v;
383 	register_t *retval;
384 {
385 	struct linux_sys_modify_ldt_args /* {
386 		syscallarg(int) func;
387 		syscallarg(void *) ptr;
388 		syscallarg(size_t) bytecount;
389 	} */ *uap = v;
390 
391 	switch (SCARG(uap, func)) {
392 #ifdef USER_LDT
393 	case 0:
394 		return (linux_read_ldt(p, uap, retval));
395 
396 	case 1:
397 		return (linux_write_ldt(p, uap, retval));
398 #endif /* USER_LDT */
399 
400 	default:
401 		return (ENOSYS);
402 	}
403 }
404 
405 /*
406  * XXX Pathetic hack to make svgalib work. This will fake the major
407  * device number of an opened VT so that svgalib likes it. grmbl.
408  * Should probably do it 'wrong the right way' and use a mapping
409  * array for all major device numbers, and map linux_mknod too.
410  */
411 dev_t
412 linux_fakedev(dev)
413 	dev_t dev;
414 {
415 
416 	if (major(dev) == NETBSD_CONS_MAJOR)
417 		return makedev(LINUX_CONS_MAJOR, (minor(dev) + 1));
418 	return dev;
419 }
420 
421 /*
422  * We come here in a last attempt to satisfy a Linux ioctl() call
423  */
424 int
425 linux_machdepioctl(p, v, retval)
426 	struct proc *p;
427 	void *v;
428 	register_t *retval;
429 {
430 	struct linux_sys_ioctl_args /* {
431 		syscallarg(int) fd;
432 		syscallarg(u_long) com;
433 		syscallarg(caddr_t) data;
434 	} */ *uap = v;
435 	struct sys_ioctl_args bia;
436 	u_long com;
437 #if NVT > 0
438 	int error;
439 	struct vt_mode lvt;
440 	caddr_t bvtp, sg;
441 #endif
442 
443 	SCARG(&bia, fd) = SCARG(uap, fd);
444 	SCARG(&bia, data) = SCARG(uap, data);
445 	com = SCARG(uap, com);
446 
447 	switch (com) {
448 #if NVT > 0
449 	case LINUX_KDGKBMODE:
450 		com = KDGKBMODE;
451 		break;
452 	case LINUX_KDSKBMODE:
453 		com = KDSKBMODE;
454 		if ((unsigned)SCARG(uap, data) == LINUX_K_MEDIUMRAW)
455 			SCARG(&bia, data) = (caddr_t)K_RAW;
456 		break;
457 	case LINUX_KDMKTONE:
458 		com = KDMKTONE;
459 		break;
460 	case LINUX_KDSETMODE:
461 		com = KDSETMODE;
462 		break;
463 	case LINUX_KDENABIO:
464 		com = KDENABIO;
465 		break;
466 	case LINUX_KDDISABIO:
467 		com = KDDISABIO;
468 		break;
469 	case LINUX_KDGETLED:
470 		com = KDGETLED;
471 		break;
472 	case LINUX_KDSETLED:
473 		com = KDSETLED;
474 		break;
475 	case LINUX_VT_OPENQRY:
476 		com = VT_OPENQRY;
477 		break;
478 	case LINUX_VT_GETMODE:
479 		SCARG(&bia, com) = VT_GETMODE;
480 		if ((error = sys_ioctl(p, &bia, retval)))
481 			return error;
482 		if ((error = copyin(SCARG(uap, data), (caddr_t)&lvt,
483 		    sizeof (struct vt_mode))))
484 			return error;
485 		lvt.relsig = bsd_to_linux_sig[lvt.relsig];
486 		lvt.acqsig = bsd_to_linux_sig[lvt.acqsig];
487 		lvt.frsig = bsd_to_linux_sig[lvt.frsig];
488 		return copyout((caddr_t)&lvt, SCARG(uap, data),
489 		    sizeof (struct vt_mode));
490 	case LINUX_VT_SETMODE:
491 		com = VT_SETMODE;
492 		if ((error = copyin(SCARG(uap, data), (caddr_t)&lvt,
493 		    sizeof (struct vt_mode))))
494 			return error;
495 		lvt.relsig = linux_to_bsd_sig[lvt.relsig];
496 		lvt.acqsig = linux_to_bsd_sig[lvt.acqsig];
497 		lvt.frsig = linux_to_bsd_sig[lvt.frsig];
498 		sg = stackgap_init(p->p_emul);
499 		bvtp = stackgap_alloc(&sg, sizeof (struct vt_mode));
500 		if ((error = copyout(&lvt, bvtp, sizeof (struct vt_mode))))
501 			return error;
502 		SCARG(&bia, data) = bvtp;
503 		break;
504 	case LINUX_VT_RELDISP:
505 		com = VT_RELDISP;
506 		break;
507 	case LINUX_VT_ACTIVATE:
508 		com = VT_ACTIVATE;
509 		break;
510 	case LINUX_VT_WAITACTIVE:
511 		com = VT_WAITACTIVE;
512 		break;
513 #endif
514 	default:
515 		printf("linux_machdepioctl: invalid ioctl %08lx\n", com);
516 		return EINVAL;
517 	}
518 	SCARG(&bia, com) = com;
519 	return sys_ioctl(p, &bia, retval);
520 }
521 
522 /*
523  * Set I/O permissions for a process. Just set the maximum level
524  * right away (ignoring the argument), otherwise we would have
525  * to rely on I/O permission maps, which are not implemented.
526  */
527 int
528 linux_sys_iopl(p, v, retval)
529 	struct proc *p;
530 	void *v;
531 	register_t *retval;
532 {
533 #if 0
534 	struct linux_sys_iopl_args /* {
535 		syscallarg(int) level;
536 	} */ *uap = v;
537 #endif
538 	struct trapframe *fp = p->p_md.md_regs;
539 
540 	if (suser(p->p_ucred, &p->p_acflag) != 0)
541 		return EPERM;
542 	fp->tf_eflags |= PSL_IOPL;
543 	*retval = 0;
544 	return 0;
545 }
546 
547 /*
548  * See above. If a root process tries to set access to an I/O port,
549  * just let it have the whole range.
550  */
551 int
552 linux_sys_ioperm(p, v, retval)
553 	struct proc *p;
554 	void *v;
555 	register_t *retval;
556 {
557 	struct linux_sys_ioperm_args /* {
558 		syscallarg(unsigned int) lo;
559 		syscallarg(unsigned int) hi;
560 		syscallarg(int) val;
561 	} */ *uap = v;
562 	struct trapframe *fp = p->p_md.md_regs;
563 
564 	if (suser(p->p_ucred, &p->p_acflag) != 0)
565 		return EPERM;
566 	if (SCARG(uap, val))
567 		fp->tf_eflags |= PSL_IOPL;
568 	*retval = 0;
569 	return 0;
570 }
571