1 /*    $Id: unaligned.c,v 1.10 2002/09/22 02:21:05 tausq Exp $
2  *
3  *    Unaligned memory access handler
4  *
5  *    Copyright (C) 2001 Randolph Chung <tausq@debian.org>
6  *    Significantly tweaked by LaMont Jones <lamont@debian.org>
7  *
8  *    This program is free software; you can redistribute it and/or modify
9  *    it under the terms of the GNU General Public License as published by
10  *    the Free Software Foundation; either version 2, or (at your option)
11  *    any later version.
12  *
13  *    This program is distributed in the hope that it will be useful,
14  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
15  *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  *    GNU General Public License for more details.
17  *
18  *    You should have received a copy of the GNU General Public License
19  *    along with this program; if not, write to the Free Software
20  *    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21  *
22  */
23 
24 #include <linux/config.h>
25 #include <linux/sched.h>
26 #include <linux/kernel.h>
27 #include <linux/string.h>
28 #include <linux/errno.h>
29 #include <linux/ptrace.h>
30 #include <linux/timer.h>
31 #include <linux/mm.h>
32 #include <linux/smp.h>
33 #include <linux/smp_lock.h>
34 #include <linux/spinlock.h>
35 #include <linux/init.h>
36 #include <linux/interrupt.h>
37 #include <asm/system.h>
38 #include <asm/uaccess.h>
39 #include <asm/io.h>
40 #include <asm/irq.h>
41 #include <asm/atomic.h>
42 
43 #include <asm/smp.h>
44 #include <asm/pdc.h>
45 
46 /* #define DEBUG_UNALIGNED 1 */
47 
48 #ifdef DEBUG_UNALIGNED
49 #define DPRINTF(fmt, args...) do { printk(KERN_DEBUG "%s:%d:%s ", __FILE__, __LINE__, __FUNCTION__ ); printk(KERN_DEBUG fmt, ##args ); } while (0)
50 #else
51 #define DPRINTF(fmt, args...)
52 #endif
53 
54 #ifdef __LP64__
55 #define RFMT "%016lx"
56 #else
57 #define RFMT "%08lx"
58 #endif
59 
60 /* 1111 1100 0000 0000 0001 0011 1100 0000 */
61 #define OPCODE1(a,b,c)	((a)<<26|(b)<<12|(c)<<6)
62 #define OPCODE2(a,b)	((a)<<26|(b)<<1)
63 #define OPCODE3(a,b)	((a)<<26|(b)<<2)
64 #define OPCODE4(a)	((a)<<26)
65 #define OPCODE1_MASK	OPCODE1(0x3f,1,0xf)
66 #define OPCODE2_MASK 	OPCODE2(0x3f,1)
67 #define OPCODE3_MASK	OPCODE3(0x3f,1)
68 #define OPCODE4_MASK    OPCODE4(0x3f)
69 
70 /* skip LDB - never unaligned (index) */
71 #define OPCODE_LDH_I	OPCODE1(0x03,0,0x1)
72 #define OPCODE_LDW_I	OPCODE1(0x03,0,0x2)
73 #define OPCODE_LDD_I	OPCODE1(0x03,0,0x3)
74 #define OPCODE_LDDA_I	OPCODE1(0x03,0,0x4)
75 #define OPCODE_LDCD_I	OPCODE1(0x03,0,0x5)
76 #define OPCODE_LDWA_I	OPCODE1(0x03,0,0x6)
77 #define OPCODE_LDCW_I	OPCODE1(0x03,0,0x7)
78 /* skip LDB - never unaligned (short) */
79 #define OPCODE_LDH_S	OPCODE1(0x03,1,0x1)
80 #define OPCODE_LDW_S	OPCODE1(0x03,1,0x2)
81 #define OPCODE_LDD_S	OPCODE1(0x03,1,0x3)
82 #define OPCODE_LDDA_S	OPCODE1(0x03,1,0x4)
83 #define OPCODE_LDCD_S	OPCODE1(0x03,1,0x5)
84 #define OPCODE_LDWA_S	OPCODE1(0x03,1,0x6)
85 #define OPCODE_LDCW_S	OPCODE1(0x03,1,0x7)
86 /* skip STB - never unaligned */
87 #define OPCODE_STH	OPCODE1(0x03,1,0x9)
88 #define OPCODE_STW	OPCODE1(0x03,1,0xa)
89 #define OPCODE_STD	OPCODE1(0x03,1,0xb)
90 /* skip STBY - never unaligned */
91 /* skip STDBY - never unaligned */
92 #define OPCODE_STWA	OPCODE1(0x03,1,0xe)
93 #define OPCODE_STDA	OPCODE1(0x03,1,0xf)
94 
95 #define OPCODE_LDD_L	OPCODE2(0x14,0)
96 #define OPCODE_FLDD_L	OPCODE2(0x14,1)
97 #define OPCODE_STD_L	OPCODE2(0x1c,0)
98 #define OPCODE_FSTD_L	OPCODE2(0x1c,1)
99 
100 #define OPCODE_LDW_M	OPCODE3(0x17,1)
101 #define OPCODE_FLDW_L	OPCODE3(0x17,0)
102 #define OPCODE_FSTW_L	OPCODE3(0x1f,0)
103 #define OPCODE_STW_M	OPCODE3(0x1f,1)
104 
105 #define OPCODE_LDH_L    OPCODE4(0x11)
106 #define OPCODE_LDW_L    OPCODE4(0x12)
107 #define OPCODE_LDWM     OPCODE4(0x13)
108 #define OPCODE_STH_L    OPCODE4(0x19)
109 #define OPCODE_STW_L    OPCODE4(0x1A)
110 #define OPCODE_STWM     OPCODE4(0x1B)
111 
112 #define MAJOR_OP(i) (((i)>>26)&0x3f)
113 #define R1(i) (((i)>>21)&0x1f)
114 #define R2(i) (((i)>>16)&0x1f)
115 #define R3(i) ((i)&0x1f)
116 #define IM(i,n) (((i)>>1&((1<<(n-1))-1))|((i)&1?((0-1L)<<(n-1)):0))
117 #define IM5_2(i) IM((i)>>16,5)
118 #define IM5_3(i) IM((i),5)
119 #define IM14(i) IM((i),14)
120 
121 int unaligned_enabled = 1;
122 
123 void die_if_kernel (char *str, struct pt_regs *regs, long err);
124 
emulate_ldh(struct pt_regs * regs,int toreg)125 static int emulate_ldh(struct pt_regs *regs, int toreg)
126 {
127 	unsigned long saddr = regs->ior;
128 	unsigned long val = 0;
129 
130 	DPRINTF("load " RFMT ":" RFMT " to r%d for 2 bytes\n",
131 		regs->isr, regs->ior, toreg);
132 
133 	__asm__ __volatile__  (
134 "	mtsp	%3, %%sr1\n"
135 "	ldbs	0(%%sr1,%2), %%r20\n"
136 "	ldbs	1(%%sr1,%2), %0\n"
137 	"depw	%%r20, 23, 24, %0\n"
138 	: "=r" (val)
139 	: "0" (val), "r" (saddr), "r" (regs->isr)
140 	: "r20" );
141 
142 	DPRINTF("val = 0x" RFMT "\n", val);
143 
144 	if (toreg)
145 		regs->gr[toreg] = val;
146 
147 	return 0;
148 }
emulate_ldw(struct pt_regs * regs,int toreg)149 static int emulate_ldw(struct pt_regs *regs, int toreg)
150 {
151 	unsigned long saddr = regs->ior;
152 	unsigned long val = 0;
153 
154 	DPRINTF("load " RFMT ":" RFMT " to r%d for 4 bytes\n",
155 		regs->isr, regs->ior, toreg);
156 
157 	__asm__ __volatile__  (
158 "	zdep	%2,28,2,%%r19\n"		/* r19=(ofs&3)*8 */
159 "	mtsp	%3, %%sr1\n"
160 "	depw	%%r0,31,2,%2\n"
161 "	ldw	0(%%sr1,%2),%0\n"
162 "	ldw	4(%%sr1,%2),%%r20\n"
163 "	subi	32,%%r19,%%r19\n"
164 "	mtctl	%%r19,11\n"
165 "	vshd	%0,%%r20,%0\n"
166 	: "=r" (val)
167 	: "0" (val), "r" (saddr), "r" (regs->isr)
168 	: "r19", "r20" );
169 
170 	DPRINTF("val = 0x" RFMT "\n", val);
171 
172 	if (toreg)
173 		regs->gr[toreg] = val;
174 
175 	return 0;
176 }
177 #ifdef __LP64__
emulate_ldd(struct pt_regs * regs,int toreg)178 static int emulate_ldd(struct pt_regs *regs, int toreg)
179 {
180 	unsigned long saddr = regs->ior;
181 	unsigned long val = 0;
182 
183 	DPRINTF("load " RFMT ":" RFMT " to r%d for 8 bytes\n",
184 		regs->isr, regs->ior, toreg);
185 
186 	__asm__ __volatile__  (
187 "	depd,z	%2,60,3,%%r19\n"		/* r19=(ofs&7)*8 */
188 "	mtsp	%3, %%sr1\n"
189 "	depd	%%r0,63,3,%2\n"
190 "	ldd	0(%%sr1,%2),%0\n"
191 "	ldd	8(%%sr1,%2),%%r20\n"
192 "	subi	64,%%r19,%%r19\n"
193 "	mtsar	%%r19\n"
194 "	shrpd	%0,%%r20,%%sar,%0\n"
195 	: "=r" (val)
196 	: "0" (val), "r" (saddr), "r" (regs->isr)
197 	: "r19", "r20" );
198 
199 	DPRINTF("val = 0x" RFMT "\n", val);
200 
201 	if (toreg)
202 		regs->gr[toreg] = val;
203 
204 	return 0;
205 }
206 #endif
207 
emulate_sth(struct pt_regs * regs,int frreg)208 static int emulate_sth(struct pt_regs *regs, int frreg)
209 {
210 	unsigned long val = regs->gr[frreg];
211 	if (!frreg)
212 		val = 0;
213 
214 	DPRINTF("store r%d (0x" RFMT ") to " RFMT ":" RFMT " for 2 bytes\n", frreg,
215 		regs->gr[frreg], regs->isr, regs->ior);
216 
217 	__asm__ __volatile__ (
218 "	mtsp %2, %%sr1\n"
219 "	extrw,u %0, 23, 8, %%r19\n"
220 "	stb %0, 1(%%sr1, %1)\n"
221 "	stb %%r19, 0(%%sr1, %1)\n"
222 	:
223 	: "r" (val), "r" (regs->ior), "r" (regs->isr)
224 	: "r19" );
225 
226 	return 0;
227 }
emulate_stw(struct pt_regs * regs,int frreg)228 static int emulate_stw(struct pt_regs *regs, int frreg)
229 {
230 	unsigned long val = regs->gr[frreg];
231 	if (!frreg)
232 		val = 0;
233 
234 	DPRINTF("store r%d (0x" RFMT ") to " RFMT ":" RFMT " for 4 bytes\n", frreg,
235 		regs->gr[frreg], regs->isr, regs->ior);
236 
237 
238 	__asm__ __volatile__ (
239 "	mtsp %2, %%sr1\n"
240 "	zdep	%1, 28, 2, %%r19\n"
241 "	dep	%%r0, 31, 2, %1\n"
242 "	mtsar	%%r19\n"
243 "	depwi,z	-2, %%sar, 32, %%r19\n"
244 "	ldw	0(%%sr1,%1),%%r20\n"
245 "	ldw	4(%%sr1,%1),%%r21\n"
246 "	vshd	%%r0, %0, %%r22\n"
247 "	vshd	%0, %%r0, %%r1\n"
248 "	and	%%r20, %%r19, %%r20\n"
249 "	andcm	%%r21, %%r19, %%r21\n"
250 "	or	%%r22, %%r20, %%r20\n"
251 "	or	%%r1, %%r21, %%r21\n"
252 "	stw	%%r20,0(%%sr1,%1)\n"
253 "	stw	%%r21,4(%%sr1,%1)\n"
254 	:
255 	: "r" (val), "r" (regs->ior), "r" (regs->isr)
256 	: "r19", "r20", "r21", "r22", "r1" );
257 
258 	return 0;
259 }
260 #ifdef __LP64__
emulate_std(struct pt_regs * regs,int frreg)261 static int emulate_std(struct pt_regs *regs, int frreg)
262 {
263 	unsigned long val = regs->gr[frreg];
264 	if (!frreg)
265 		val = 0;
266 
267 	DPRINTF("store r%d (0x" RFMT ") to " RFMT ":" RFMT " for 8 bytes\n", frreg,
268 		regs->gr[frreg], regs->isr, regs->ior);
269 
270 
271 	__asm__ __volatile__ (
272 "	mtsp %2, %%sr1\n"
273 "	depd,z	%1, 60, 3, %%r19\n"
274 "	depd	%%r0, 63, 3, %1\n"
275 "	mtsar	%%r19\n"
276 "	depdi,z	-2, %%sar, 64, %%r19\n"
277 "	ldd	0(%%sr1,%1),%%r20\n"
278 "	ldd	8(%%sr1,%1),%%r21\n"
279 "	shrpd	%%r0, %0, %%sar, %%r22\n"
280 "	shrpd	%0, %%r0, %%sar, %%r1\n"
281 "	and	%%r20, %%r19, %%r20\n"
282 "	andcm	%%r21, %%r19, %%r21\n"
283 "	or	%%r22, %%r20, %%r20\n"
284 "	or	%%r1, %%r21, %%r21\n"
285 "	std	%%r20,0(%%sr1,%1)\n"
286 "	std	%%r21,8(%%sr1,%1)\n"
287 	:
288 	: "r" (val), "r" (regs->ior), "r" (regs->isr)
289 	: "r19", "r20", "r21", "r22", "r1" );
290 
291 	return 0;
292 }
293 #endif
294 
handle_unaligned(struct pt_regs * regs)295 void handle_unaligned(struct pt_regs *regs)
296 {
297 	unsigned long unaligned_count = 0;
298 	unsigned long last_time = 0;
299 	unsigned long newbase = regs->gr[R1(regs->iir)];
300 	int modify = 0;
301 	int ret = -1;
302 	struct siginfo si;
303 
304 	/* if the unaligned access is inside the kernel:
305 	 *   if the access is caused by a syscall, then we fault the calling
306 	 *     user process
307 	 *   otherwise we halt the kernel
308 	 */
309 	if (!user_mode(regs))
310 	{
311 		const struct exception_table_entry *fix;
312 
313 		/* see if the offending code have its own
314 		 * exception handler
315 		 */
316 
317 		fix = search_exception_table(regs->iaoq[0]);
318 		if (fix)
319 		{
320 			/* lower bits of fix->skip are flags
321 			 * upper bits are the handler addr
322 			 */
323 			if (fix->skip & 1)
324 				regs->gr[8] = -EFAULT;
325 			if (fix->skip & 2)
326 				regs->gr[9] = 0;
327 
328 			regs->iaoq[0] += ((fix->skip) & ~3);
329 			regs->iaoq[1] = regs->iaoq[0] + 4;
330 			regs->gr[0] &= ~PSW_B;
331 
332 			return;
333 		}
334 	}
335 
336 	/* log a message with pacing */
337 	if (user_mode(regs))
338 	{
339 		if (unaligned_count > 5 && jiffies - last_time > 5*HZ)
340 		{
341 			unaligned_count = 0;
342 			last_time = jiffies;
343 		}
344 		if (++unaligned_count < 5)
345 		{
346 			char buf[256];
347 			sprintf(buf, "%s(%d): unaligned access to 0x" RFMT " at ip=0x" RFMT "\n",
348 				current->comm, current->pid, regs->ior, regs->iaoq[0]);
349 			printk(KERN_WARNING "%s", buf);
350 #ifdef DEBUG_UNALIGNED
351 			show_regs(regs);
352 #endif
353 		}
354 	}
355 
356 	if (!unaligned_enabled)
357 		goto force_sigbus;
358 
359 	/* handle modification - OK, it's ugly, see the instruction manual */
360 	switch (MAJOR_OP(regs->iir))
361 	{
362 	case 0x03:
363 	case 0x09:
364 	case 0x0b:
365 		if (regs->iir&0x20)
366 		{
367 			modify = 1;
368 			if (regs->iir&0x1000)		/* short loads */
369 				if (regs->iir&0x200)
370 					newbase += IM5_3(regs->iir);
371 				else
372 					newbase += IM5_2(regs->iir);
373 			else if (regs->iir&0x2000)	/* scaled indexed */
374 			{
375 				int shift=0;
376 				switch (regs->iir & OPCODE1_MASK)
377 				{
378 				case OPCODE_LDH_I:
379 					shift= 1; break;
380 				case OPCODE_LDW_I:
381 					shift= 2; break;
382 				case OPCODE_LDD_I:
383 				case OPCODE_LDDA_I:
384 					shift= 3; break;
385 				}
386 				newbase += (R2(regs->iir)?regs->gr[R2(regs->iir)]:0)<<shift;
387 			} else				/* simple indexed */
388 				newbase += (R2(regs->iir)?regs->gr[R2(regs->iir)]:0);
389 		}
390 		break;
391 	case 0x13:
392 	case 0x1b:
393 		modify = 1;
394 		newbase += IM14(regs->iir);
395 		break;
396 	case 0x14:
397 	case 0x1c:
398 		if (regs->iir&8)
399 		{
400 			modify = 1;
401 			newbase += IM14(regs->iir&~0xe);
402 		}
403 		break;
404 	case 0x16:
405 	case 0x1e:
406 		modify = 1;
407 		newbase += IM14(regs->iir&6);
408 		break;
409 	case 0x17:
410 	case 0x1f:
411 		if (regs->iir&4)
412 		{
413 			modify = 1;
414 			newbase += IM14(regs->iir&~4);
415 		}
416 		break;
417 	}
418 
419 	if (regs->isr != regs->sr[7])
420 	{
421 		printk(KERN_CRIT "isr verification failed (isr: " RFMT ", sr7: " RFMT "\n",
422 			regs->isr, regs->sr[7]);
423 
424 		/* don't kill him though, since he has appropriate access to the page, or we
425 		 * would never have gotten here.
426 		 */
427 	}
428 
429 	/* TODO: make this cleaner... */
430 	switch (regs->iir & OPCODE1_MASK)
431 	{
432 	case OPCODE_LDH_I:
433 	case OPCODE_LDH_S:
434 		ret = emulate_ldh(regs, R3(regs->iir));
435 		break;
436 
437 	case OPCODE_LDW_I:
438 	case OPCODE_LDWA_I:
439 	case OPCODE_LDW_S:
440 	case OPCODE_LDWA_S:
441 		ret = emulate_ldw(regs, R3(regs->iir));
442 		break;
443 
444 	case OPCODE_STH:
445 		ret = emulate_sth(regs, R2(regs->iir));
446 		break;
447 
448 	case OPCODE_STW:
449 	case OPCODE_STWA:
450 		ret = emulate_stw(regs, R2(regs->iir));
451 		break;
452 
453 #ifdef __LP64__
454 	case OPCODE_LDD_I:
455 	case OPCODE_LDDA_I:
456 	case OPCODE_LDD_S:
457 	case OPCODE_LDDA_S:
458 		ret = emulate_ldd(regs, R3(regs->iir));
459 		break;
460 
461 	case OPCODE_STD:
462 	case OPCODE_STDA:
463 		ret = emulate_std(regs, R2(regs->iir));
464 		break;
465 #endif
466 
467 	case OPCODE_LDCD_I:
468 	case OPCODE_LDCW_I:
469 	case OPCODE_LDCD_S:
470 	case OPCODE_LDCW_S:
471 		ret = -1;	/* "undefined", but lets kill them. */
472 		break;
473 	}
474 #ifdef __LP64__
475 	switch (regs->iir & OPCODE2_MASK)
476 	{
477 	case OPCODE_LDD_L:
478 	case OPCODE_FLDD_L:
479 		ret = emulate_ldd(regs, R2(regs->iir));
480 		break;
481 
482 	case OPCODE_STD_L:
483 	case OPCODE_FSTD_L:
484 		ret = emulate_std(regs, R2(regs->iir));
485 		break;
486 	}
487 #endif
488 	switch (regs->iir & OPCODE3_MASK)
489 	{
490 	case OPCODE_LDW_M:
491 	case OPCODE_FLDW_L:
492 		ret = emulate_ldw(regs, R2(regs->iir));
493 		break;
494 
495 	case OPCODE_FSTW_L:
496 	case OPCODE_STW_M:
497 		ret = emulate_stw(regs, R2(regs->iir));
498 		break;
499 	}
500 	switch (regs->iir & OPCODE4_MASK)
501 	{
502 	case OPCODE_LDH_L:
503 		ret = emulate_ldh(regs, R2(regs->iir));
504 		break;
505 	case OPCODE_LDW_L:
506 	case OPCODE_LDWM:
507 		ret = emulate_ldw(regs, R2(regs->iir));
508 		break;
509 	case OPCODE_STH_L:
510 		ret = emulate_sth(regs, R2(regs->iir));
511 		break;
512 	case OPCODE_STW_L:
513 	case OPCODE_STWM:
514 		ret = emulate_stw(regs, R2(regs->iir));
515 		break;
516 	}
517 	/* XXX LJ - need to handle float load/store */
518 
519 	if (modify && R1(regs->iir))
520 		regs->gr[R1(regs->iir)] = newbase;
521 
522 
523 	if (ret < 0)
524 		printk(KERN_CRIT "Not-handled unaligned insn 0x%08lx\n", regs->iir);
525 
526 	DPRINTF("ret = %d\n", ret);
527 
528 	if (ret)
529 	{
530 		printk(KERN_CRIT "Unaligned handler failed, ret = %d\n", ret);
531 		die_if_kernel("Unaligned data reference", regs, 28);
532 force_sigbus:
533 		/* couldn't handle it ... */
534 		si.si_signo = SIGBUS;
535 		si.si_errno = 0;
536 		si.si_code = BUS_ADRALN;
537 		si.si_addr = (void *)regs->ior;
538 		force_sig_info(SIGBUS, &si, current);
539 
540 		return;
541 	}
542 
543 	/* else we handled it, advance the PC.... */
544 	regs->iaoq[0] = regs->iaoq[1];
545 	regs->iaoq[1] = regs->iaoq[0] + 4;
546 }
547 
548 /*
549  * NB: check_unaligned() is only used for PCXS processors right
550  * now, so we only check for PA1.1 encodings at this point.
551  */
552 
553 int
check_unaligned(struct pt_regs * regs)554 check_unaligned(struct pt_regs *regs)
555 {
556 	unsigned long align_mask;
557 
558 	/* Get alignment mask */
559 
560 	align_mask = 0UL;
561 	switch (regs->iir & OPCODE1_MASK) {
562 
563 	case OPCODE_LDH_I:
564 	case OPCODE_LDH_S:
565 	case OPCODE_STH:
566 		align_mask = 1UL;
567 		break;
568 
569 	case OPCODE_LDW_I:
570 	case OPCODE_LDWA_I:
571 	case OPCODE_LDW_S:
572 	case OPCODE_LDWA_S:
573 	case OPCODE_STW:
574 	case OPCODE_STWA:
575 		align_mask = 3UL;
576 		break;
577 
578 	default:
579 		switch (regs->iir & OPCODE4_MASK) {
580 		case OPCODE_LDH_L:
581 		case OPCODE_STH_L:
582 			align_mask = 1UL;
583 			break;
584 		case OPCODE_LDW_L:
585 		case OPCODE_LDWM:
586 		case OPCODE_STW_L:
587 		case OPCODE_STWM:
588 			align_mask = 3UL;
589 			break;
590 		}
591 		break;
592 	}
593 
594 	return (int)(regs->ior & align_mask);
595 }
596 
597