xref: /netbsd-src/sys/compat/linux/arch/i386/linux_machdep.c (revision 7c7c171d130af9949261bc7dce2150a03c3d239c)
1 /*	$NetBSD: linux_machdep.c,v 1.39 1998/01/24 13:19:48 mycroft 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 "opt_vm86.h"
35 #include "opt_user_ldt.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/signalvar.h>
40 #include <sys/kernel.h>
41 #include <sys/map.h>
42 #include <sys/proc.h>
43 #include <sys/user.h>
44 #include <sys/buf.h>
45 #include <sys/reboot.h>
46 #include <sys/conf.h>
47 #include <sys/exec.h>
48 #include <sys/file.h>
49 #include <sys/callout.h>
50 #include <sys/malloc.h>
51 #include <sys/mbuf.h>
52 #include <sys/msgbuf.h>
53 #include <sys/mount.h>
54 #include <sys/vnode.h>
55 #include <sys/device.h>
56 #include <sys/syscallargs.h>
57 #include <sys/filedesc.h>
58 #include <sys/exec_elf.h>
59 
60 #include <compat/linux/linux_types.h>
61 #include <compat/linux/linux_signal.h>
62 #include <compat/linux/linux_syscallargs.h>
63 #include <compat/linux/linux_util.h>
64 #include <compat/linux/linux_ioctl.h>
65 #include <compat/linux/linux_exec.h>
66 
67 #include <machine/cpu.h>
68 #include <machine/cpufunc.h>
69 #include <machine/psl.h>
70 #include <machine/reg.h>
71 #include <machine/segments.h>
72 #include <machine/specialreg.h>
73 #include <machine/sysarch.h>
74 #include <machine/vm86.h>
75 #include <machine/vmparam.h>
76 #include <machine/linux_machdep.h>
77 
78 /*
79  * To see whether pcvt is configured (for virtual console ioctl calls).
80  */
81 #ifndef NVT
82 #include "vt.h"
83 #endif
84 #if NVT > 0
85 #include <arch/i386/isa/pcvt/pcvt_ioctl.h>
86 #endif
87 
88 #ifdef USER_LDT
89 #include <machine/cpu.h>
90 int linux_read_ldt __P((struct proc *, struct linux_sys_modify_ldt_args *,
91     register_t *));
92 int linux_write_ldt __P((struct proc *, struct linux_sys_modify_ldt_args *,
93     register_t *));
94 #endif
95 
96 /*
97  * Deal with some i386-specific things in the Linux emulation code.
98  */
99 
100 void
101 linux_setregs(p, epp, stack)
102 	struct proc *p;
103 	struct exec_package *epp;
104 	u_long stack;
105 {
106 	register struct pcb *pcb = &p->p_addr->u_pcb;
107 
108 	pcb->pcb_savefpu.sv_env.en_cw = __Linux_NPXCW__;
109 	setregs(p, epp, stack);
110 }
111 
112 /*
113  * Send an interrupt to process.
114  *
115  * Stack is set up to allow sigcode stored
116  * in u. to call routine, followed by kcall
117  * to sigreturn routine below.  After sigreturn
118  * resets the signal mask, the stack, and the
119  * frame pointer, it returns to the user
120  * specified pc, psl.
121  */
122 
123 void
124 linux_sendsig(catcher, sig, mask, code)
125 	sig_t catcher;
126 	int sig, mask;
127 	u_long code;
128 {
129 	register struct proc *p = curproc;
130 	register struct trapframe *tf;
131 	struct linux_sigframe *fp, frame;
132 	struct sigacts *psp = p->p_sigacts;
133 	int oonstack;
134 	extern char linux_sigcode[], linux_esigcode[];
135 
136 	tf = p->p_md.md_regs;
137 	oonstack = psp->ps_sigstk.ss_flags & SS_ONSTACK;
138 
139 	/*
140 	 * Allocate space for the signal handler context.
141 	 */
142 	if ((psp->ps_flags & SAS_ALTSTACK) && !oonstack &&
143 	    (psp->ps_sigonstack & sigmask(sig))) {
144 		fp = (struct linux_sigframe *)(psp->ps_sigstk.ss_sp +
145 		    psp->ps_sigstk.ss_size - sizeof(struct linux_sigframe));
146 		psp->ps_sigstk.ss_flags |= SS_ONSTACK;
147 	} else {
148 		fp = (struct linux_sigframe *)tf->tf_esp - 1;
149 	}
150 
151 	frame.sf_handler = catcher;
152 	frame.sf_sig = bsd_to_linux_sig[sig];
153 
154 	/*
155 	 * Build the signal context to be used by sigreturn.
156 	 */
157 	frame.sf_sc.sc_mask   = mask;
158 #ifdef VM86
159 	if (tf->tf_eflags & PSL_VM) {
160 		frame.sf_sc.sc_gs = tf->tf_vm86_gs;
161 		frame.sf_sc.sc_fs = tf->tf_vm86_fs;
162 		frame.sf_sc.sc_es = tf->tf_vm86_es;
163 		frame.sf_sc.sc_ds = tf->tf_vm86_ds;
164 		frame.sf_sc.sc_eflags = get_vflags(p);
165 	} else
166 #endif
167 	{
168 		__asm("movl %%gs,%w0" : "=r" (frame.sf_sc.sc_gs));
169 		__asm("movl %%fs,%w0" : "=r" (frame.sf_sc.sc_fs));
170 		frame.sf_sc.sc_es = tf->tf_es;
171 		frame.sf_sc.sc_ds = tf->tf_ds;
172 		frame.sf_sc.sc_eflags = tf->tf_eflags;
173 	}
174 	frame.sf_sc.sc_edi = tf->tf_edi;
175 	frame.sf_sc.sc_esi = tf->tf_esi;
176 	frame.sf_sc.sc_ebp = tf->tf_ebp;
177 	frame.sf_sc.sc_ebx = tf->tf_ebx;
178 	frame.sf_sc.sc_edx = tf->tf_edx;
179 	frame.sf_sc.sc_ecx = tf->tf_ecx;
180 	frame.sf_sc.sc_eax = tf->tf_eax;
181 	frame.sf_sc.sc_eip = tf->tf_eip;
182 	frame.sf_sc.sc_cs = tf->tf_cs;
183 	frame.sf_sc.sc_esp_at_signal = tf->tf_esp;
184 	frame.sf_sc.sc_ss = tf->tf_ss;
185 	frame.sf_sc.sc_err = tf->tf_err;
186 	frame.sf_sc.sc_trapno = tf->tf_trapno;
187 
188 	if (copyout(&frame, fp, sizeof(frame)) != 0) {
189 		/*
190 		 * Process has trashed its stack; give it an illegal
191 		 * instruction to halt it in its tracks.
192 		 */
193 		sigexit(p, SIGILL);
194 		/* NOTREACHED */
195 	}
196 
197 	/*
198 	 * Build context to run handler in.
199 	 */
200 	tf->tf_es = GSEL(GUDATA_SEL, SEL_UPL);
201 	tf->tf_ds = GSEL(GUDATA_SEL, SEL_UPL);
202 	tf->tf_eip = (int)(((char *)PS_STRINGS) -
203 	     (linux_esigcode - linux_sigcode));
204 	tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL);
205 	tf->tf_eflags &= ~(PSL_T|PSL_VM|PSL_AC);
206 	tf->tf_esp = (int)fp;
207 	tf->tf_ss = GSEL(GUDATA_SEL, SEL_UPL);
208 }
209 
210 /*
211  * System call to cleanup state after a signal
212  * has been taken.  Reset signal mask and
213  * stack state from context left by sendsig (above).
214  * Return to previous pc and psl as specified by
215  * context left by sendsig. Check carefully to
216  * make sure that the user has not modified the
217  * psl to gain improper privileges or to cause
218  * a machine fault.
219  */
220 int
221 linux_sys_sigreturn(p, v, retval)
222 	struct proc *p;
223 	void *v;
224 	register_t *retval;
225 {
226 	struct linux_sys_sigreturn_args /* {
227 		syscallarg(struct linux_sigcontext *) scp;
228 	} */ *uap = v;
229 	struct linux_sigcontext *scp, context;
230 	register struct trapframe *tf;
231 
232 	tf = p->p_md.md_regs;
233 
234 	/*
235 	 * The trampoline code hands us the context.
236 	 * It is unsafe to keep track of it ourselves, in the event that a
237 	 * program jumps out of a signal handler.
238 	 */
239 	scp = SCARG(uap, scp);
240 	if (copyin((caddr_t)scp, &context, sizeof(*scp)) != 0)
241 		return (EFAULT);
242 
243 	/*
244 	 * Restore signal context.
245 	 */
246 #ifdef VM86
247 	if (context.sc_eflags & PSL_VM) {
248 		tf->tf_vm86_gs = context.sc_gs;
249 		tf->tf_vm86_fs = context.sc_fs;
250 		tf->tf_vm86_es = context.sc_es;
251 		tf->tf_vm86_ds = context.sc_ds;
252 		set_vflags(p, context.sc_eflags);
253 	} else
254 #endif
255 	{
256 		/*
257 		 * Check for security violations.  If we're returning to
258 		 * protected mode, the CPU will validate the segment registers
259 		 * automatically and generate a trap on violations.  We handle
260 		 * the trap, rather than doing all of the checking here.
261 		 */
262 		if (((context.sc_eflags ^ tf->tf_eflags) & PSL_USERSTATIC) != 0 ||
263 		    !USERMODE(context.sc_cs, context.sc_eflags))
264 			return (EINVAL);
265 
266 		/* %fs and %gs were restored by the trampoline. */
267 		tf->tf_es = context.sc_es;
268 		tf->tf_ds = context.sc_ds;
269 		tf->tf_eflags = context.sc_eflags;
270 	}
271 	tf->tf_edi = context.sc_edi;
272 	tf->tf_esi = context.sc_esi;
273 	tf->tf_ebp = context.sc_ebp;
274 	tf->tf_ebx = context.sc_ebx;
275 	tf->tf_edx = context.sc_edx;
276 	tf->tf_ecx = context.sc_ecx;
277 	tf->tf_eax = context.sc_eax;
278 	tf->tf_eip = context.sc_eip;
279 	tf->tf_cs = context.sc_cs;
280 	tf->tf_esp = context.sc_esp_at_signal;
281 	tf->tf_ss = context.sc_ss;
282 
283 	p->p_sigacts->ps_sigstk.ss_flags &= ~SS_ONSTACK;
284 	p->p_sigmask = context.sc_mask & ~sigcantmask;
285 
286 	return (EJUSTRETURN);
287 }
288 
289 #ifdef USER_LDT
290 
291 int
292 linux_read_ldt(p, uap, retval)
293 	struct proc *p;
294 	struct linux_sys_modify_ldt_args /* {
295 		syscallarg(int) func;
296 		syscallarg(void *) ptr;
297 		syscallarg(size_t) bytecount;
298 	} */ *uap;
299 	register_t *retval;
300 {
301 	struct i386_get_ldt_args gl;
302 	int error;
303 	caddr_t sg;
304 	char *parms;
305 
306 	sg = stackgap_init(p->p_emul);
307 
308 	gl.start = 0;
309 	gl.desc = SCARG(uap, ptr);
310 	gl.num = SCARG(uap, bytecount) / sizeof(union descriptor);
311 
312 	parms = stackgap_alloc(&sg, sizeof(gl));
313 
314 	if ((error = copyout(&gl, parms, sizeof(gl))) != 0)
315 		return (error);
316 
317 	if ((error = i386_get_ldt(p, parms, retval)) != 0)
318 		return (error);
319 
320 	*retval *= sizeof(union descriptor);
321 	return (0);
322 }
323 
324 struct linux_ldt_info {
325 	u_int entry_number;
326 	u_long base_addr;
327 	u_int limit;
328 	u_int seg_32bit:1;
329 	u_int contents:2;
330 	u_int read_exec_only:1;
331 	u_int limit_in_pages:1;
332 	u_int seg_not_present:1;
333 };
334 
335 int
336 linux_write_ldt(p, uap, retval)
337 	struct proc *p;
338 	struct linux_sys_modify_ldt_args /* {
339 		syscallarg(int) func;
340 		syscallarg(void *) ptr;
341 		syscallarg(size_t) bytecount;
342 	} */ *uap;
343 	register_t *retval;
344 {
345 	struct linux_ldt_info ldt_info;
346 	struct segment_descriptor sd;
347 	struct i386_set_ldt_args sl;
348 	int error;
349 	caddr_t sg;
350 	char *parms;
351 
352 	if (SCARG(uap, bytecount) != sizeof(ldt_info))
353 		return (EINVAL);
354 	if ((error = copyin(SCARG(uap, ptr), &ldt_info, sizeof(ldt_info))) != 0)
355 		return error;
356 	if (ldt_info.contents == 3)
357 		return (EINVAL);
358 
359 	sg = stackgap_init(p->p_emul);
360 
361 	sd.sd_lobase = ldt_info.base_addr & 0xffffff;
362 	sd.sd_hibase = (ldt_info.base_addr >> 24) & 0xff;
363 	sd.sd_lolimit = ldt_info.limit & 0xffff;
364 	sd.sd_hilimit = (ldt_info.limit >> 16) & 0xf;
365 	sd.sd_type =
366 	    16 | (ldt_info.contents << 2) | (!ldt_info.read_exec_only << 1);
367 	sd.sd_dpl = SEL_UPL;
368 	sd.sd_p = !ldt_info.seg_not_present;
369 	sd.sd_def32 = ldt_info.seg_32bit;
370 	sd.sd_gran = ldt_info.limit_in_pages;
371 
372 	sl.start = ldt_info.entry_number;
373 	sl.desc = stackgap_alloc(&sg, sizeof(sd));
374 	sl.num = 1;
375 
376 #if 0
377 	printf("linux_write_ldt: idx=%d, base=%x, limit=%x\n",
378 	    ldt_info.entry_number, ldt_info.base_addr, ldt_info.limit);
379 #endif
380 
381 	parms = stackgap_alloc(&sg, sizeof(sl));
382 
383 	if ((error = copyout(&sd, sl.desc, sizeof(sd))) != 0)
384 		return (error);
385 	if ((error = copyout(&sl, parms, sizeof(sl))) != 0)
386 		return (error);
387 
388 	if ((error = i386_set_ldt(p, parms, retval)) != 0)
389 		return (error);
390 
391 	*retval = 0;
392 	return (0);
393 }
394 
395 #endif /* USER_LDT */
396 
397 int
398 linux_sys_modify_ldt(p, v, retval)
399 	struct proc *p;
400 	void *v;
401 	register_t *retval;
402 {
403 	struct linux_sys_modify_ldt_args /* {
404 		syscallarg(int) func;
405 		syscallarg(void *) ptr;
406 		syscallarg(size_t) bytecount;
407 	} */ *uap = v;
408 
409 	switch (SCARG(uap, func)) {
410 #ifdef USER_LDT
411 	case 0:
412 		return (linux_read_ldt(p, uap, retval));
413 
414 	case 1:
415 		return (linux_write_ldt(p, uap, retval));
416 #endif /* USER_LDT */
417 
418 	default:
419 		return (ENOSYS);
420 	}
421 }
422 
423 /*
424  * XXX Pathetic hack to make svgalib work. This will fake the major
425  * device number of an opened VT so that svgalib likes it. grmbl.
426  * Should probably do it 'wrong the right way' and use a mapping
427  * array for all major device numbers, and map linux_mknod too.
428  */
429 dev_t
430 linux_fakedev(dev)
431 	dev_t dev;
432 {
433 
434 	if (major(dev) == NETBSD_CONS_MAJOR)
435 		return makedev(LINUX_CONS_MAJOR, (minor(dev) + 1));
436 	return dev;
437 }
438 
439 /*
440  * We come here in a last attempt to satisfy a Linux ioctl() call
441  */
442 int
443 linux_machdepioctl(p, v, retval)
444 	struct proc *p;
445 	void *v;
446 	register_t *retval;
447 {
448 	struct linux_sys_ioctl_args /* {
449 		syscallarg(int) fd;
450 		syscallarg(u_long) com;
451 		syscallarg(caddr_t) data;
452 	} */ *uap = v;
453 	struct sys_ioctl_args bia;
454 	u_long com;
455 #if NVT > 0
456 	int error;
457 	struct vt_mode lvt;
458 	caddr_t bvtp, sg;
459 #endif
460 
461 	SCARG(&bia, fd) = SCARG(uap, fd);
462 	SCARG(&bia, data) = SCARG(uap, data);
463 	com = SCARG(uap, com);
464 
465 	switch (com) {
466 #if NVT > 0
467 	case LINUX_KDGKBMODE:
468 		com = KDGKBMODE;
469 		break;
470 	case LINUX_KDSKBMODE:
471 		com = KDSKBMODE;
472 		if ((unsigned)SCARG(uap, data) == LINUX_K_MEDIUMRAW)
473 			SCARG(&bia, data) = (caddr_t)K_RAW;
474 		break;
475 	case LINUX_KDMKTONE:
476 		com = KDMKTONE;
477 		break;
478 	case LINUX_KDSETMODE:
479 		com = KDSETMODE;
480 		break;
481 	case LINUX_KDENABIO:
482 		com = KDENABIO;
483 		break;
484 	case LINUX_KDDISABIO:
485 		com = KDDISABIO;
486 		break;
487 	case LINUX_KDGETLED:
488 		com = KDGETLED;
489 		break;
490 	case LINUX_KDSETLED:
491 		com = KDSETLED;
492 		break;
493 	case LINUX_VT_OPENQRY:
494 		com = VT_OPENQRY;
495 		break;
496 	case LINUX_VT_GETMODE:
497 		SCARG(&bia, com) = VT_GETMODE;
498 		if ((error = sys_ioctl(p, &bia, retval)))
499 			return error;
500 		if ((error = copyin(SCARG(uap, data), (caddr_t)&lvt,
501 		    sizeof (struct vt_mode))))
502 			return error;
503 		lvt.relsig = bsd_to_linux_sig[lvt.relsig];
504 		lvt.acqsig = bsd_to_linux_sig[lvt.acqsig];
505 		lvt.frsig = bsd_to_linux_sig[lvt.frsig];
506 		return copyout((caddr_t)&lvt, SCARG(uap, data),
507 		    sizeof (struct vt_mode));
508 	case LINUX_VT_SETMODE:
509 		com = VT_SETMODE;
510 		if ((error = copyin(SCARG(uap, data), (caddr_t)&lvt,
511 		    sizeof (struct vt_mode))))
512 			return error;
513 		lvt.relsig = linux_to_bsd_sig[lvt.relsig];
514 		lvt.acqsig = linux_to_bsd_sig[lvt.acqsig];
515 		lvt.frsig = linux_to_bsd_sig[lvt.frsig];
516 		sg = stackgap_init(p->p_emul);
517 		bvtp = stackgap_alloc(&sg, sizeof (struct vt_mode));
518 		if ((error = copyout(&lvt, bvtp, sizeof (struct vt_mode))))
519 			return error;
520 		SCARG(&bia, data) = bvtp;
521 		break;
522 	case LINUX_VT_RELDISP:
523 		com = VT_RELDISP;
524 		break;
525 	case LINUX_VT_ACTIVATE:
526 		com = VT_ACTIVATE;
527 		break;
528 	case LINUX_VT_WAITACTIVE:
529 		com = VT_WAITACTIVE;
530 		break;
531 #endif
532 	default:
533 		printf("linux_machdepioctl: invalid ioctl %08lx\n", com);
534 		return EINVAL;
535 	}
536 	SCARG(&bia, com) = com;
537 	return sys_ioctl(p, &bia, retval);
538 }
539 
540 /*
541  * Set I/O permissions for a process. Just set the maximum level
542  * right away (ignoring the argument), otherwise we would have
543  * to rely on I/O permission maps, which are not implemented.
544  */
545 int
546 linux_sys_iopl(p, v, retval)
547 	struct proc *p;
548 	void *v;
549 	register_t *retval;
550 {
551 #if 0
552 	struct linux_sys_iopl_args /* {
553 		syscallarg(int) level;
554 	} */ *uap = v;
555 #endif
556 	struct trapframe *fp = p->p_md.md_regs;
557 
558 	if (suser(p->p_ucred, &p->p_acflag) != 0)
559 		return EPERM;
560 	fp->tf_eflags |= PSL_IOPL;
561 	*retval = 0;
562 	return 0;
563 }
564 
565 /*
566  * See above. If a root process tries to set access to an I/O port,
567  * just let it have the whole range.
568  */
569 int
570 linux_sys_ioperm(p, v, retval)
571 	struct proc *p;
572 	void *v;
573 	register_t *retval;
574 {
575 	struct linux_sys_ioperm_args /* {
576 		syscallarg(unsigned int) lo;
577 		syscallarg(unsigned int) hi;
578 		syscallarg(int) val;
579 	} */ *uap = v;
580 	struct trapframe *fp = p->p_md.md_regs;
581 
582 	if (suser(p->p_ucred, &p->p_acflag) != 0)
583 		return EPERM;
584 	if (SCARG(uap, val))
585 		fp->tf_eflags |= PSL_IOPL;
586 	*retval = 0;
587 	return 0;
588 }
589