xref: /netbsd-src/sys/compat/linux/arch/powerpc/linux_machdep.c (revision b5677b36047b601b9addaaa494a58ceae82c2a6c)
1 /*	$NetBSD: linux_machdep.c,v 1.39 2008/04/28 20:23:43 martin Exp $ */
2 
3 /*-
4  * Copyright (c) 1995, 2000, 2001 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Frank van der Linden and Emmanuel Dreyfus.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __KERNEL_RCSID(0, "$NetBSD: linux_machdep.c,v 1.39 2008/04/28 20:23:43 martin Exp $");
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/signalvar.h>
38 #include <sys/kernel.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/exec.h>
45 #include <sys/file.h>
46 #include <sys/callout.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/msgbuf.h>
50 #include <sys/mount.h>
51 #include <sys/vnode.h>
52 #include <sys/device.h>
53 #include <sys/syscallargs.h>
54 #include <sys/filedesc.h>
55 #include <sys/exec_elf.h>
56 #include <sys/disklabel.h>
57 #include <sys/ioctl.h>
58 #include <miscfs/specfs/specdev.h>
59 
60 #include <compat/linux/common/linux_types.h>
61 #include <compat/linux/common/linux_signal.h>
62 #include <compat/linux/common/linux_util.h>
63 #include <compat/linux/common/linux_ioctl.h>
64 #include <compat/linux/common/linux_hdio.h>
65 #include <compat/linux/common/linux_exec.h>
66 #include <compat/linux/common/linux_machdep.h>
67 
68 #include <compat/linux/linux_syscallargs.h>
69 
70 #include <sys/cpu.h>
71 #include <machine/fpu.h>
72 #include <machine/psl.h>
73 #include <machine/reg.h>
74 #include <machine/vmparam.h>
75 
76 /*
77  * To see whether wscons is configured (for virtual console ioctl calls).
78  */
79 #if defined(_KERNEL_OPT)
80 #include "wsdisplay.h"
81 #endif
82 #if (NWSDISPLAY > 0)
83 #include <dev/wscons/wsconsio.h>
84 #include <dev/wscons/wsdisplay_usl_io.h>
85 #endif
86 
87 /*
88  * Set set up registers on exec.
89  * XXX not used at the moment since in sys/kern/exec_conf, LINUX_COMPAT
90  * entry uses NetBSD's native setregs instead of linux_setregs
91  */
92 void
93 linux_setregs(struct lwp *l, struct exec_package *pack, u_long stack)
94 {
95 	setregs(l, pack, stack);
96 }
97 
98 /*
99  * Send an interrupt to process.
100  *
101  * Adapted from arch/powerpc/powerpc/sig_machdep.c:sendsig and
102  * compat/linux/arch/i386/linux_machdep.c:linux_sendsig
103  *
104  * XXX Does not work well yet with RT signals
105  *
106  */
107 
108 void
109 linux_sendsig(const ksiginfo_t *ksi, const sigset_t *mask)
110 {
111 	const int sig = ksi->ksi_signo;
112 	struct lwp *l = curlwp;
113 	struct proc *p = l->l_proc;
114 	struct trapframe *tf;
115 	sig_t catcher = SIGACTION(p, sig).sa_handler;
116 	struct linux_sigregs frame;
117 	struct linux_pt_regs linux_regs;
118 	struct linux_sigcontext sc;
119 	register_t fp;
120 	int onstack, error;
121 	int i;
122 
123 	tf = trapframe(l);
124 
125 	/*
126 	 * Do we need to jump onto the signal stack?
127 	 */
128 	onstack =
129 	    (l->l_sigstk.ss_flags & (SS_DISABLE | SS_ONSTACK)) == 0 &&
130 	    (SIGACTION(p, sig).sa_flags & SA_ONSTACK) != 0;
131 
132 	/*
133 	 * Signal stack is broken (see at the end of linux_sigreturn), so we do
134 	 * not use it yet. XXX fix this.
135 	 */
136 	onstack=0;
137 
138 	/*
139 	 * Allocate space for the signal handler context.
140 	 */
141 	if (onstack) {
142 		fp = (register_t)
143 		    ((char *)l->l_sigstk.ss_sp +
144 		    l->l_sigstk.ss_size);
145 	} else {
146 		fp = tf->fixreg[1];
147 	}
148 #ifdef DEBUG_LINUX
149 	printf("fp at start of linux_sendsig = %x\n", fp);
150 #endif
151 	fp -= sizeof(struct linux_sigregs);
152 	fp &= ~0xf;
153 
154 	/*
155 	 * Prepare a sigcontext for later.
156 	 */
157 	memset(&sc, 0, sizeof sc);
158 	sc.lsignal = (int)native_to_linux_signo[sig];
159 	sc.lhandler = (unsigned long)catcher;
160 	native_to_linux_old_extra_sigset(&sc.lmask, &sc._unused[3], mask);
161 	sc.lregs = (struct linux_pt_regs*)fp;
162 
163 	/*
164 	 * Setup the signal stack frame as Linux does it in
165 	 * arch/ppc/kernel/signal.c:setup_frame()
166 	 *
167 	 * Save register context.
168 	 */
169 	for (i = 0; i < 32; i++)
170 		linux_regs.lgpr[i] = tf->fixreg[i];
171 	linux_regs.lnip = tf->srr0;
172 	linux_regs.lmsr = tf->srr1 & PSL_USERSRR1;
173 	linux_regs.lorig_gpr3 = tf->fixreg[3]; /* XXX Is that right? */
174 	linux_regs.lctr = tf->ctr;
175 	linux_regs.llink = tf->lr;
176 	linux_regs.lxer = tf->xer;
177 	linux_regs.lccr = tf->cr;
178 	linux_regs.lmq = 0;  			/* Unused, 601 only */
179 	linux_regs.ltrap = tf->exc;
180 	linux_regs.ldar = tf->dar;
181 	linux_regs.ldsisr = tf->dsisr;
182 	linux_regs.lresult = 0;
183 
184 	memset(&frame, 0, sizeof(frame));
185 	memcpy(&frame.lgp_regs, &linux_regs, sizeof(linux_regs));
186 
187 	save_fpu_lwp(curlwp, FPU_SAVE);
188 	memcpy(&frame.lfp_regs, curpcb->pcb_fpu.fpreg, sizeof(frame.lfp_regs));
189 
190 	/*
191 	 * Copy Linux's signal trampoline on the user stack It should not
192 	 * be used, but Linux binaries might expect it to be there.
193 	 */
194 	frame.ltramp[0] = 0x38997777; /* li r0, 0x7777 */
195 	frame.ltramp[1] = 0x44000002; /* sc */
196 
197 	/*
198 	 * Move it to the user stack
199 	 * There is a little trick here, about the LINUX_ABIGAP: the
200 	 * linux_sigreg structure has a 56 int gap to support rs6000/xcoff
201 	 * binaries. But the Linux kernel seems to do without it, and it
202 	 * just skip it when building the stack frame. Hence the LINUX_ABIGAP.
203 	 */
204 	sendsig_reset(l, sig);
205 	mutex_exit(p->p_lock);
206 	error = copyout(&frame, (void *)fp, sizeof (frame) - LINUX_ABIGAP);
207 
208 	if (error != 0) {
209 		/*
210 		 * Process has trashed its stack; give it an illegal
211 		 * instruction to halt it in its tracks.
212 		 */
213 		mutex_enter(p->p_lock);
214 		sigexit(l, SIGILL);
215 		/* NOTREACHED */
216 	}
217 
218 	/*
219 	 * Add a sigcontext on the stack
220 	 */
221 	fp -= sizeof(struct linux_sigcontext);
222 	error = copyout(&sc, (void *)fp, sizeof (struct linux_sigcontext));
223 	mutex_enter(p->p_lock);
224 
225 	if (error != 0) {
226 		/*
227 		 * Process has trashed its stack; give it an illegal
228 		 * instruction to halt it in its tracks.
229 		 */
230 		sigexit(l, SIGILL);
231 		/* NOTREACHED */
232 	}
233 
234 	/*
235 	 * Set the registers according to how the Linux process expects them.
236 	 * "Mind the gap" Linux expects a gap here.
237 	 */
238 	tf->fixreg[1] = fp - LINUX__SIGNAL_FRAMESIZE;
239 	tf->lr = (int)catcher;
240 	tf->fixreg[3] = (int)native_to_linux_signo[sig];
241 	tf->fixreg[4] = fp;
242 	tf->srr0 = (int)p->p_sigctx.ps_sigcode;
243 
244 #ifdef DEBUG_LINUX
245 	printf("fp at end of linux_sendsig = %x\n", fp);
246 #endif
247 	/*
248 	 * Remember that we're now on the signal stack.
249 	 */
250 	if (onstack)
251 		l->l_sigstk.ss_flags |= SS_ONSTACK;
252 #ifdef DEBUG_LINUX
253 	printf("linux_sendsig: exitting. fp=0x%lx\n",(long)fp);
254 #endif
255 }
256 
257 /*
258  * System call to cleanup state after a signal
259  * has been taken.  Reset signal mask and
260  * stack state from context left by sendsig (above).
261  * Return to previous pc and psl as specified by
262  * context left by sendsig. Check carefully to
263  * make sure that the user has not modified the
264  * psl to gain improper privileges or to cause
265  * a machine fault.
266  *
267  * XXX not tested
268  */
269 int
270 linux_sys_rt_sigreturn(struct lwp *l, const struct linux_sys_rt_sigreturn_args *uap, register_t *retval)
271 {
272 	/* {
273 		syscallarg(struct linux_rt_sigframe *) sfp;
274 	} */
275 	struct proc *p = l->l_proc;
276 	struct linux_rt_sigframe *scp, sigframe;
277 	struct linux_sigregs sregs;
278 	struct linux_pt_regs *lregs;
279 	struct trapframe *tf;
280 	sigset_t mask;
281 	int i;
282 
283 	/*
284 	 * The trampoline code hands us the context.
285 	 * It is unsafe to keep track of it ourselves, in the event that a
286 	 * program jumps out of a signal handler.
287 	 */
288 	scp = SCARG(uap, sfp);
289 
290 	/*
291 	 * Get the context from user stack
292 	 */
293 	if (copyin((void *)scp, &sigframe, sizeof(*scp)))
294 		return (EFAULT);
295 
296 	/*
297 	 *  Restore register context.
298 	 */
299 	if (copyin((void *)sigframe.luc.luc_context.lregs,
300 		   &sregs, sizeof(sregs)))
301 		return (EFAULT);
302 	lregs = (struct linux_pt_regs *)&sregs.lgp_regs;
303 
304 	tf = trapframe(l);
305 #ifdef DEBUG_LINUX
306 	    (unsigned long)tf, (unsigned long)scp);
307 #endif
308 
309 	if (!PSL_USEROK_P(lregs->lmsr))
310 		return (EINVAL);
311 
312 	for (i = 0; i < 32; i++)
313 		tf->fixreg[i] = lregs->lgpr[i];
314 	tf->lr = lregs->llink;
315 	tf->cr = lregs->lccr;
316 	tf->xer = lregs->lxer;
317 	tf->ctr = lregs->lctr;
318 	tf->srr0 = lregs->lnip;
319 	tf->srr1 = lregs->lmsr;
320 
321 	/*
322 	 * Make sure the fpu state is discarded
323 	 */
324 	save_fpu_lwp(curlwp, FPU_DISCARD);
325 
326 	memcpy(curpcb->pcb_fpu.fpreg, (void *)&sregs.lfp_regs,
327 	       sizeof(curpcb->pcb_fpu.fpreg));
328 
329 	mutex_enter(p->p_lock);
330 
331 	/*
332 	 * Restore signal stack.
333 	 *
334 	 * XXX cannot find the onstack information in Linux sig context.
335 	 * Is signal stack really supported on Linux?
336 	 *
337 	 * It seems to be supported in libc6...
338 	 */
339 	/* if (sc.sc_onstack & SS_ONSTACK)
340 		l->l_sigstk.ss_flags |= SS_ONSTACK;
341 	else */
342 		l->l_sigstk.ss_flags &= ~SS_ONSTACK;
343 
344 	/*
345 	 * Grab the signal mask
346 	 */
347 	linux_to_native_sigset(&mask, &sigframe.luc.luc_sigmask);
348 	(void) sigprocmask1(l, SIG_SETMASK, &mask, 0);
349 
350 	mutex_exit(p->p_lock);
351 
352 	return (EJUSTRETURN);
353 }
354 
355 
356 /*
357  * The following needs code review for potential security issues
358  */
359 int
360 linux_sys_sigreturn(struct lwp *l, const struct linux_sys_sigreturn_args *uap, register_t *retval)
361 {
362 	/* {
363 		syscallarg(struct linux_sigcontext *) scp;
364 	} */
365 	struct proc *p = l->l_proc;
366 	struct linux_sigcontext *scp, context;
367 	struct linux_sigregs sregs;
368 	struct linux_pt_regs *lregs;
369 	struct trapframe *tf;
370 	sigset_t mask;
371 	int i;
372 
373 	/*
374 	 * The trampoline code hands us the context.
375 	 * It is unsafe to keep track of it ourselves, in the event that a
376 	 * program jumps out of a signal handler.
377 	 */
378 	scp = SCARG(uap, scp);
379 
380 	/*
381 	 * Get the context from user stack
382 	 */
383 	if (copyin(scp, &context, sizeof(*scp)))
384 		return (EFAULT);
385 
386 	/*
387 	 *  Restore register context.
388 	 */
389 	if (copyin((void *)context.lregs, &sregs, sizeof(sregs)))
390 		return (EFAULT);
391 	lregs = (struct linux_pt_regs *)&sregs.lgp_regs;
392 
393 	tf = trapframe(l);
394 #ifdef DEBUG_LINUX
395 	printf("linux_sys_sigreturn: trapframe=0x%lx scp=0x%lx\n",
396 	    (unsigned long)tf, (unsigned long)scp);
397 #endif
398 
399 	if (!PSL_USEROK_P(lregs->lmsr))
400 		return (EINVAL);
401 
402 	for (i = 0; i < 32; i++)
403 		tf->fixreg[i] = lregs->lgpr[i];
404 	tf->lr = lregs->llink;
405 	tf->cr = lregs->lccr;
406 	tf->xer = lregs->lxer;
407 	tf->ctr = lregs->lctr;
408 	tf->srr0 = lregs->lnip;
409 	tf->srr1 = lregs->lmsr;
410 
411 	/*
412 	 * Make sure the fpu state is discarded
413 	 */
414 	save_fpu_lwp(curlwp, FPU_DISCARD);
415 
416 	memcpy(curpcb->pcb_fpu.fpreg, (void *)&sregs.lfp_regs,
417 	       sizeof(curpcb->pcb_fpu.fpreg));
418 
419 	mutex_enter(p->p_lock);
420 
421 	/*
422 	 * Restore signal stack.
423 	 *
424 	 * XXX cannot find the onstack information in Linux sig context.
425 	 * Is signal stack really supported on Linux?
426 	 */
427 #if 0
428 	if (sc.sc_onstack & SS_ONSTACK)
429 		l->l_sigstk.ss_flags |= SS_ONSTACK;
430 	else
431 #endif
432 		l->l_sigstk.ss_flags &= ~SS_ONSTACK;
433 
434 	/* Restore signal mask. */
435 	linux_old_extra_to_native_sigset(&mask, &context.lmask,
436 	    &context._unused[3]);
437 	(void) sigprocmask1(l, SIG_SETMASK, &mask, 0);
438 
439 	mutex_exit(p->p_lock);
440 
441 	return (EJUSTRETURN);
442 }
443 
444 
445 #if 0
446 int
447 linux_sys_modify_ldt(struct proc *p, void *v, register_t *retval)
448 {
449 	/*
450 	 * This syscall is not implemented in Linux/PowerPC: we should not
451 	 * be here
452 	 */
453 #ifdef DEBUG_LINUX
454 	printf("linux_sys_modify_ldt: should not be here.\n");
455 #endif
456   return 0;
457 }
458 #endif
459 
460 /*
461  * major device numbers remapping
462  */
463 dev_t
464 linux_fakedev(dev_t dev, int raw)
465 {
466 	/* XXX write me */
467 	return dev;
468 }
469 
470 /*
471  * We come here in a last attempt to satisfy a Linux ioctl() call
472  */
473 int
474 linux_machdepioctl(struct lwp *l, const struct linux_sys_ioctl_args *uap, register_t *retval)
475 {
476 	/* {
477 		syscallarg(int) fd;
478 		syscallarg(u_long) com;
479 		syscallarg(void *) data;
480 	} */
481 	struct sys_ioctl_args bia;
482 	u_long com;
483 
484 	SCARG(&bia, fd) = SCARG(uap, fd);
485 	SCARG(&bia, data) = SCARG(uap, data);
486 	com = SCARG(uap, com);
487 
488 	switch (com) {
489 	default:
490 		printf("linux_machdepioctl: invalid ioctl %08lx\n", com);
491 		return EINVAL;
492 	}
493 	SCARG(&bia, com) = com;
494 	/* XXX NJWLWP */
495 	return sys_ioctl(curlwp, &bia, retval);
496 }
497 #if 0
498 /*
499  * Set I/O permissions for a process. Just set the maximum level
500  * right away (ignoring the argument), otherwise we would have
501  * to rely on I/O permission maps, which are not implemented.
502  */
503 int
504 linux_sys_iopl(struct lwp *l, const void *v, register_t *retval)
505 {
506 	/*
507 	 * This syscall is not implemented in Linux/PowerPC: we should not be here
508 	 */
509 #ifdef DEBUG_LINUX
510 	printf("linux_sys_iopl: should not be here.\n");
511 #endif
512 	return 0;
513 }
514 #endif
515 
516 /*
517  * See above. If a root process tries to set access to an I/O port,
518  * just let it have the whole range.
519  */
520 int
521 linux_sys_ioperm(struct lwp *l, const struct linux_sys_ioperm_args *uap, register_t *retval)
522 {
523 	/*
524 	 * This syscall is not implemented in Linux/PowerPC: we should not be here
525 	 */
526 #ifdef DEBUG_LINUX
527 	printf("linux_sys_ioperm: should not be here.\n");
528 #endif
529 	return 0;
530 }
531 
532 /*
533  * wrapper linux_sys_new_uname() -> linux_sys_uname()
534  */
535 int
536 linux_sys_new_uname(struct lwp *l, const struct linux_sys_new_uname_args *uap, register_t *retval)
537 {
538 	return linux_sys_uname(l, (const void *)uap, retval);
539 }
540 
541 /*
542  * wrapper linux_sys_new_select() -> linux_sys_select()
543  */
544 int
545 linux_sys_new_select(struct lwp *l, const struct linux_sys_new_select_args *uap, register_t *retval)
546 {
547 	return linux_sys_select(l, (const void *)uap, retval);
548 }
549 
550 int
551 linux_usertrap(struct lwp *l, vaddr_t trapaddr, void *arg)
552 {
553 	return 0;
554 }
555