xref: /openbmc/linux/arch/powerpc/kernel/align.c (revision 96de0e252cedffad61b3cb5e05662c591898e69a)
1 /* align.c - handle alignment exceptions for the Power PC.
2  *
3  * Copyright (c) 1996 Paul Mackerras <paulus@cs.anu.edu.au>
4  * Copyright (c) 1998-1999 TiVo, Inc.
5  *   PowerPC 403GCX modifications.
6  * Copyright (c) 1999 Grant Erickson <grant@lcse.umn.edu>
7  *   PowerPC 403GCX/405GP modifications.
8  * Copyright (c) 2001-2002 PPC64 team, IBM Corp
9  *   64-bit and Power4 support
10  * Copyright (c) 2005 Benjamin Herrenschmidt, IBM Corp
11  *                    <benh@kernel.crashing.org>
12  *   Merge ppc32 and ppc64 implementations
13  *
14  * This program is free software; you can redistribute it and/or
15  * modify it under the terms of the GNU General Public License
16  * as published by the Free Software Foundation; either version
17  * 2 of the License, or (at your option) any later version.
18  */
19 
20 #include <linux/kernel.h>
21 #include <linux/mm.h>
22 #include <asm/processor.h>
23 #include <asm/uaccess.h>
24 #include <asm/system.h>
25 #include <asm/cache.h>
26 #include <asm/cputable.h>
27 
28 struct aligninfo {
29 	unsigned char len;
30 	unsigned char flags;
31 };
32 
33 #define IS_XFORM(inst)	(((inst) >> 26) == 31)
34 #define IS_DSFORM(inst)	(((inst) >> 26) >= 56)
35 
36 #define INVALID	{ 0, 0 }
37 
38 /* Bits in the flags field */
39 #define LD	0	/* load */
40 #define ST	1	/* store */
41 #define SE	2	/* sign-extend value, or FP ld/st as word */
42 #define F	4	/* to/from fp regs */
43 #define U	8	/* update index register */
44 #define M	0x10	/* multiple load/store */
45 #define SW	0x20	/* byte swap */
46 #define S	0x40	/* single-precision fp or... */
47 #define SX	0x40	/* ... byte count in XER */
48 #define HARD	0x80	/* string, stwcx. */
49 #define E4	0x40	/* SPE endianness is word */
50 #define E8	0x80	/* SPE endianness is double word */
51 
52 /* DSISR bits reported for a DCBZ instruction: */
53 #define DCBZ	0x5f	/* 8xx/82xx dcbz faults when cache not enabled */
54 
55 #define SWAP(a, b)	(t = (a), (a) = (b), (b) = t)
56 
57 /*
58  * The PowerPC stores certain bits of the instruction that caused the
59  * alignment exception in the DSISR register.  This array maps those
60  * bits to information about the operand length and what the
61  * instruction would do.
62  */
63 static struct aligninfo aligninfo[128] = {
64 	{ 4, LD },		/* 00 0 0000: lwz / lwarx */
65 	INVALID,		/* 00 0 0001 */
66 	{ 4, ST },		/* 00 0 0010: stw */
67 	INVALID,		/* 00 0 0011 */
68 	{ 2, LD },		/* 00 0 0100: lhz */
69 	{ 2, LD+SE },		/* 00 0 0101: lha */
70 	{ 2, ST },		/* 00 0 0110: sth */
71 	{ 4, LD+M },		/* 00 0 0111: lmw */
72 	{ 4, LD+F+S },		/* 00 0 1000: lfs */
73 	{ 8, LD+F },		/* 00 0 1001: lfd */
74 	{ 4, ST+F+S },		/* 00 0 1010: stfs */
75 	{ 8, ST+F },		/* 00 0 1011: stfd */
76 	INVALID,		/* 00 0 1100 */
77 	{ 8, LD },		/* 00 0 1101: ld/ldu/lwa */
78 	INVALID,		/* 00 0 1110 */
79 	{ 8, ST },		/* 00 0 1111: std/stdu */
80 	{ 4, LD+U },		/* 00 1 0000: lwzu */
81 	INVALID,		/* 00 1 0001 */
82 	{ 4, ST+U },		/* 00 1 0010: stwu */
83 	INVALID,		/* 00 1 0011 */
84 	{ 2, LD+U },		/* 00 1 0100: lhzu */
85 	{ 2, LD+SE+U },		/* 00 1 0101: lhau */
86 	{ 2, ST+U },		/* 00 1 0110: sthu */
87 	{ 4, ST+M },		/* 00 1 0111: stmw */
88 	{ 4, LD+F+S+U },	/* 00 1 1000: lfsu */
89 	{ 8, LD+F+U },		/* 00 1 1001: lfdu */
90 	{ 4, ST+F+S+U },	/* 00 1 1010: stfsu */
91 	{ 8, ST+F+U },		/* 00 1 1011: stfdu */
92 	{ 16, LD+F },		/* 00 1 1100: lfdp */
93 	INVALID,		/* 00 1 1101 */
94 	{ 16, ST+F },		/* 00 1 1110: stfdp */
95 	INVALID,		/* 00 1 1111 */
96 	{ 8, LD },		/* 01 0 0000: ldx */
97 	INVALID,		/* 01 0 0001 */
98 	{ 8, ST },		/* 01 0 0010: stdx */
99 	INVALID,		/* 01 0 0011 */
100 	INVALID,		/* 01 0 0100 */
101 	{ 4, LD+SE },		/* 01 0 0101: lwax */
102 	INVALID,		/* 01 0 0110 */
103 	INVALID,		/* 01 0 0111 */
104 	{ 4, LD+M+HARD+SX },	/* 01 0 1000: lswx */
105 	{ 4, LD+M+HARD },	/* 01 0 1001: lswi */
106 	{ 4, ST+M+HARD+SX },	/* 01 0 1010: stswx */
107 	{ 4, ST+M+HARD },	/* 01 0 1011: stswi */
108 	INVALID,		/* 01 0 1100 */
109 	{ 8, LD+U },		/* 01 0 1101: ldu */
110 	INVALID,		/* 01 0 1110 */
111 	{ 8, ST+U },		/* 01 0 1111: stdu */
112 	{ 8, LD+U },		/* 01 1 0000: ldux */
113 	INVALID,		/* 01 1 0001 */
114 	{ 8, ST+U },		/* 01 1 0010: stdux */
115 	INVALID,		/* 01 1 0011 */
116 	INVALID,		/* 01 1 0100 */
117 	{ 4, LD+SE+U },		/* 01 1 0101: lwaux */
118 	INVALID,		/* 01 1 0110 */
119 	INVALID,		/* 01 1 0111 */
120 	INVALID,		/* 01 1 1000 */
121 	INVALID,		/* 01 1 1001 */
122 	INVALID,		/* 01 1 1010 */
123 	INVALID,		/* 01 1 1011 */
124 	INVALID,		/* 01 1 1100 */
125 	INVALID,		/* 01 1 1101 */
126 	INVALID,		/* 01 1 1110 */
127 	INVALID,		/* 01 1 1111 */
128 	INVALID,		/* 10 0 0000 */
129 	INVALID,		/* 10 0 0001 */
130 	INVALID,		/* 10 0 0010: stwcx. */
131 	INVALID,		/* 10 0 0011 */
132 	INVALID,		/* 10 0 0100 */
133 	INVALID,		/* 10 0 0101 */
134 	INVALID,		/* 10 0 0110 */
135 	INVALID,		/* 10 0 0111 */
136 	{ 4, LD+SW },		/* 10 0 1000: lwbrx */
137 	INVALID,		/* 10 0 1001 */
138 	{ 4, ST+SW },		/* 10 0 1010: stwbrx */
139 	INVALID,		/* 10 0 1011 */
140 	{ 2, LD+SW },		/* 10 0 1100: lhbrx */
141 	{ 4, LD+SE },		/* 10 0 1101  lwa */
142 	{ 2, ST+SW },		/* 10 0 1110: sthbrx */
143 	INVALID,		/* 10 0 1111 */
144 	INVALID,		/* 10 1 0000 */
145 	INVALID,		/* 10 1 0001 */
146 	INVALID,		/* 10 1 0010 */
147 	INVALID,		/* 10 1 0011 */
148 	INVALID,		/* 10 1 0100 */
149 	INVALID,		/* 10 1 0101 */
150 	INVALID,		/* 10 1 0110 */
151 	INVALID,		/* 10 1 0111 */
152 	INVALID,		/* 10 1 1000 */
153 	INVALID,		/* 10 1 1001 */
154 	INVALID,		/* 10 1 1010 */
155 	INVALID,		/* 10 1 1011 */
156 	INVALID,		/* 10 1 1100 */
157 	INVALID,		/* 10 1 1101 */
158 	INVALID,		/* 10 1 1110 */
159 	{ 0, ST+HARD },		/* 10 1 1111: dcbz */
160 	{ 4, LD },		/* 11 0 0000: lwzx */
161 	INVALID,		/* 11 0 0001 */
162 	{ 4, ST },		/* 11 0 0010: stwx */
163 	INVALID,		/* 11 0 0011 */
164 	{ 2, LD },		/* 11 0 0100: lhzx */
165 	{ 2, LD+SE },		/* 11 0 0101: lhax */
166 	{ 2, ST },		/* 11 0 0110: sthx */
167 	INVALID,		/* 11 0 0111 */
168 	{ 4, LD+F+S },		/* 11 0 1000: lfsx */
169 	{ 8, LD+F },		/* 11 0 1001: lfdx */
170 	{ 4, ST+F+S },		/* 11 0 1010: stfsx */
171 	{ 8, ST+F },		/* 11 0 1011: stfdx */
172 	{ 16, LD+F },		/* 11 0 1100: lfdpx */
173 	{ 4, LD+F+SE },		/* 11 0 1101: lfiwax */
174 	{ 16, ST+F },		/* 11 0 1110: stfdpx */
175 	{ 4, ST+F },		/* 11 0 1111: stfiwx */
176 	{ 4, LD+U },		/* 11 1 0000: lwzux */
177 	INVALID,		/* 11 1 0001 */
178 	{ 4, ST+U },		/* 11 1 0010: stwux */
179 	INVALID,		/* 11 1 0011 */
180 	{ 2, LD+U },		/* 11 1 0100: lhzux */
181 	{ 2, LD+SE+U },		/* 11 1 0101: lhaux */
182 	{ 2, ST+U },		/* 11 1 0110: sthux */
183 	INVALID,		/* 11 1 0111 */
184 	{ 4, LD+F+S+U },	/* 11 1 1000: lfsux */
185 	{ 8, LD+F+U },		/* 11 1 1001: lfdux */
186 	{ 4, ST+F+S+U },	/* 11 1 1010: stfsux */
187 	{ 8, ST+F+U },		/* 11 1 1011: stfdux */
188 	INVALID,		/* 11 1 1100 */
189 	INVALID,		/* 11 1 1101 */
190 	INVALID,		/* 11 1 1110 */
191 	INVALID,		/* 11 1 1111 */
192 };
193 
194 /*
195  * Create a DSISR value from the instruction
196  */
197 static inline unsigned make_dsisr(unsigned instr)
198 {
199 	unsigned dsisr;
200 
201 
202 	/* bits  6:15 --> 22:31 */
203 	dsisr = (instr & 0x03ff0000) >> 16;
204 
205 	if (IS_XFORM(instr)) {
206 		/* bits 29:30 --> 15:16 */
207 		dsisr |= (instr & 0x00000006) << 14;
208 		/* bit     25 -->    17 */
209 		dsisr |= (instr & 0x00000040) << 8;
210 		/* bits 21:24 --> 18:21 */
211 		dsisr |= (instr & 0x00000780) << 3;
212 	} else {
213 		/* bit      5 -->    17 */
214 		dsisr |= (instr & 0x04000000) >> 12;
215 		/* bits  1: 4 --> 18:21 */
216 		dsisr |= (instr & 0x78000000) >> 17;
217 		/* bits 30:31 --> 12:13 */
218 		if (IS_DSFORM(instr))
219 			dsisr |= (instr & 0x00000003) << 18;
220 	}
221 
222 	return dsisr;
223 }
224 
225 /*
226  * The dcbz (data cache block zero) instruction
227  * gives an alignment fault if used on non-cacheable
228  * memory.  We handle the fault mainly for the
229  * case when we are running with the cache disabled
230  * for debugging.
231  */
232 static int emulate_dcbz(struct pt_regs *regs, unsigned char __user *addr)
233 {
234 	long __user *p;
235 	int i, size;
236 
237 #ifdef __powerpc64__
238 	size = ppc64_caches.dline_size;
239 #else
240 	size = L1_CACHE_BYTES;
241 #endif
242 	p = (long __user *) (regs->dar & -size);
243 	if (user_mode(regs) && !access_ok(VERIFY_WRITE, p, size))
244 		return -EFAULT;
245 	for (i = 0; i < size / sizeof(long); ++i)
246 		if (__put_user_inatomic(0, p+i))
247 			return -EFAULT;
248 	return 1;
249 }
250 
251 /*
252  * Emulate load & store multiple instructions
253  * On 64-bit machines, these instructions only affect/use the
254  * bottom 4 bytes of each register, and the loads clear the
255  * top 4 bytes of the affected register.
256  */
257 #ifdef CONFIG_PPC64
258 #define REG_BYTE(rp, i)		*((u8 *)((rp) + ((i) >> 2)) + ((i) & 3) + 4)
259 #else
260 #define REG_BYTE(rp, i)		*((u8 *)(rp) + (i))
261 #endif
262 
263 #define SWIZ_PTR(p)		((unsigned char __user *)((p) ^ swiz))
264 
265 static int emulate_multiple(struct pt_regs *regs, unsigned char __user *addr,
266 			    unsigned int reg, unsigned int nb,
267 			    unsigned int flags, unsigned int instr,
268 			    unsigned long swiz)
269 {
270 	unsigned long *rptr;
271 	unsigned int nb0, i, bswiz;
272 	unsigned long p;
273 
274 	/*
275 	 * We do not try to emulate 8 bytes multiple as they aren't really
276 	 * available in our operating environments and we don't try to
277 	 * emulate multiples operations in kernel land as they should never
278 	 * be used/generated there at least not on unaligned boundaries
279 	 */
280 	if (unlikely((nb > 4) || !user_mode(regs)))
281 		return 0;
282 
283 	/* lmw, stmw, lswi/x, stswi/x */
284 	nb0 = 0;
285 	if (flags & HARD) {
286 		if (flags & SX) {
287 			nb = regs->xer & 127;
288 			if (nb == 0)
289 				return 1;
290 		} else {
291 			unsigned long pc = regs->nip ^ (swiz & 4);
292 
293 			if (__get_user_inatomic(instr,
294 						(unsigned int __user *)pc))
295 				return -EFAULT;
296 			if (swiz == 0 && (flags & SW))
297 				instr = cpu_to_le32(instr);
298 			nb = (instr >> 11) & 0x1f;
299 			if (nb == 0)
300 				nb = 32;
301 		}
302 		if (nb + reg * 4 > 128) {
303 			nb0 = nb + reg * 4 - 128;
304 			nb = 128 - reg * 4;
305 		}
306 	} else {
307 		/* lwm, stmw */
308 		nb = (32 - reg) * 4;
309 	}
310 
311 	if (!access_ok((flags & ST ? VERIFY_WRITE: VERIFY_READ), addr, nb+nb0))
312 		return -EFAULT;	/* bad address */
313 
314 	rptr = &regs->gpr[reg];
315 	p = (unsigned long) addr;
316 	bswiz = (flags & SW)? 3: 0;
317 
318 	if (!(flags & ST)) {
319 		/*
320 		 * This zeroes the top 4 bytes of the affected registers
321 		 * in 64-bit mode, and also zeroes out any remaining
322 		 * bytes of the last register for lsw*.
323 		 */
324 		memset(rptr, 0, ((nb + 3) / 4) * sizeof(unsigned long));
325 		if (nb0 > 0)
326 			memset(&regs->gpr[0], 0,
327 			       ((nb0 + 3) / 4) * sizeof(unsigned long));
328 
329 		for (i = 0; i < nb; ++i, ++p)
330 			if (__get_user_inatomic(REG_BYTE(rptr, i ^ bswiz),
331 						SWIZ_PTR(p)))
332 				return -EFAULT;
333 		if (nb0 > 0) {
334 			rptr = &regs->gpr[0];
335 			addr += nb;
336 			for (i = 0; i < nb0; ++i, ++p)
337 				if (__get_user_inatomic(REG_BYTE(rptr,
338 								 i ^ bswiz),
339 							SWIZ_PTR(p)))
340 					return -EFAULT;
341 		}
342 
343 	} else {
344 		for (i = 0; i < nb; ++i, ++p)
345 			if (__put_user_inatomic(REG_BYTE(rptr, i ^ bswiz),
346 						SWIZ_PTR(p)))
347 				return -EFAULT;
348 		if (nb0 > 0) {
349 			rptr = &regs->gpr[0];
350 			addr += nb;
351 			for (i = 0; i < nb0; ++i, ++p)
352 				if (__put_user_inatomic(REG_BYTE(rptr,
353 								 i ^ bswiz),
354 							SWIZ_PTR(p)))
355 					return -EFAULT;
356 		}
357 	}
358 	return 1;
359 }
360 
361 /*
362  * Emulate floating-point pair loads and stores.
363  * Only POWER6 has these instructions, and it does true little-endian,
364  * so we don't need the address swizzling.
365  */
366 static int emulate_fp_pair(struct pt_regs *regs, unsigned char __user *addr,
367 			   unsigned int reg, unsigned int flags)
368 {
369 	char *ptr = (char *) &current->thread.fpr[reg];
370 	int i, ret;
371 
372 	if (!(flags & F))
373 		return 0;
374 	if (reg & 1)
375 		return 0;	/* invalid form: FRS/FRT must be even */
376 	if (!(flags & SW)) {
377 		/* not byte-swapped - easy */
378 		if (!(flags & ST))
379 			ret = __copy_from_user(ptr, addr, 16);
380 		else
381 			ret = __copy_to_user(addr, ptr, 16);
382 	} else {
383 		/* each FPR value is byte-swapped separately */
384 		ret = 0;
385 		for (i = 0; i < 16; ++i) {
386 			if (!(flags & ST))
387 				ret |= __get_user(ptr[i^7], addr + i);
388 			else
389 				ret |= __put_user(ptr[i^7], addr + i);
390 		}
391 	}
392 	if (ret)
393 		return -EFAULT;
394 	return 1;	/* exception handled and fixed up */
395 }
396 
397 #ifdef CONFIG_SPE
398 
399 static struct aligninfo spe_aligninfo[32] = {
400 	{ 8, LD+E8 },		/* 0 00 00: evldd[x] */
401 	{ 8, LD+E4 },		/* 0 00 01: evldw[x] */
402 	{ 8, LD },		/* 0 00 10: evldh[x] */
403 	INVALID,		/* 0 00 11 */
404 	{ 2, LD },		/* 0 01 00: evlhhesplat[x] */
405 	INVALID,		/* 0 01 01 */
406 	{ 2, LD },		/* 0 01 10: evlhhousplat[x] */
407 	{ 2, LD+SE },		/* 0 01 11: evlhhossplat[x] */
408 	{ 4, LD },		/* 0 10 00: evlwhe[x] */
409 	INVALID,		/* 0 10 01 */
410 	{ 4, LD },		/* 0 10 10: evlwhou[x] */
411 	{ 4, LD+SE },		/* 0 10 11: evlwhos[x] */
412 	{ 4, LD+E4 },		/* 0 11 00: evlwwsplat[x] */
413 	INVALID,		/* 0 11 01 */
414 	{ 4, LD },		/* 0 11 10: evlwhsplat[x] */
415 	INVALID,		/* 0 11 11 */
416 
417 	{ 8, ST+E8 },		/* 1 00 00: evstdd[x] */
418 	{ 8, ST+E4 },		/* 1 00 01: evstdw[x] */
419 	{ 8, ST },		/* 1 00 10: evstdh[x] */
420 	INVALID,		/* 1 00 11 */
421 	INVALID,		/* 1 01 00 */
422 	INVALID,		/* 1 01 01 */
423 	INVALID,		/* 1 01 10 */
424 	INVALID,		/* 1 01 11 */
425 	{ 4, ST },		/* 1 10 00: evstwhe[x] */
426 	INVALID,		/* 1 10 01 */
427 	{ 4, ST },		/* 1 10 10: evstwho[x] */
428 	INVALID,		/* 1 10 11 */
429 	{ 4, ST+E4 },		/* 1 11 00: evstwwe[x] */
430 	INVALID,		/* 1 11 01 */
431 	{ 4, ST+E4 },		/* 1 11 10: evstwwo[x] */
432 	INVALID,		/* 1 11 11 */
433 };
434 
435 #define	EVLDD		0x00
436 #define	EVLDW		0x01
437 #define	EVLDH		0x02
438 #define	EVLHHESPLAT	0x04
439 #define	EVLHHOUSPLAT	0x06
440 #define	EVLHHOSSPLAT	0x07
441 #define	EVLWHE		0x08
442 #define	EVLWHOU		0x0A
443 #define	EVLWHOS		0x0B
444 #define	EVLWWSPLAT	0x0C
445 #define	EVLWHSPLAT	0x0E
446 #define	EVSTDD		0x10
447 #define	EVSTDW		0x11
448 #define	EVSTDH		0x12
449 #define	EVSTWHE		0x18
450 #define	EVSTWHO		0x1A
451 #define	EVSTWWE		0x1C
452 #define	EVSTWWO		0x1E
453 
454 /*
455  * Emulate SPE loads and stores.
456  * Only Book-E has these instructions, and it does true little-endian,
457  * so we don't need the address swizzling.
458  */
459 static int emulate_spe(struct pt_regs *regs, unsigned int reg,
460 		       unsigned int instr)
461 {
462 	int t, ret;
463 	union {
464 		u64 ll;
465 		u32 w[2];
466 		u16 h[4];
467 		u8 v[8];
468 	} data, temp;
469 	unsigned char __user *p, *addr;
470 	unsigned long *evr = &current->thread.evr[reg];
471 	unsigned int nb, flags;
472 
473 	instr = (instr >> 1) & 0x1f;
474 
475 	/* DAR has the operand effective address */
476 	addr = (unsigned char __user *)regs->dar;
477 
478 	nb = spe_aligninfo[instr].len;
479 	flags = spe_aligninfo[instr].flags;
480 
481 	/* Verify the address of the operand */
482 	if (unlikely(user_mode(regs) &&
483 		     !access_ok((flags & ST ? VERIFY_WRITE : VERIFY_READ),
484 				addr, nb)))
485 		return -EFAULT;
486 
487 	/* userland only */
488 	if (unlikely(!user_mode(regs)))
489 		return 0;
490 
491 	flush_spe_to_thread(current);
492 
493 	/* If we are loading, get the data from user space, else
494 	 * get it from register values
495 	 */
496 	if (flags & ST) {
497 		data.ll = 0;
498 		switch (instr) {
499 		case EVSTDD:
500 		case EVSTDW:
501 		case EVSTDH:
502 			data.w[0] = *evr;
503 			data.w[1] = regs->gpr[reg];
504 			break;
505 		case EVSTWHE:
506 			data.h[2] = *evr >> 16;
507 			data.h[3] = regs->gpr[reg] >> 16;
508 			break;
509 		case EVSTWHO:
510 			data.h[2] = *evr & 0xffff;
511 			data.h[3] = regs->gpr[reg] & 0xffff;
512 			break;
513 		case EVSTWWE:
514 			data.w[1] = *evr;
515 			break;
516 		case EVSTWWO:
517 			data.w[1] = regs->gpr[reg];
518 			break;
519 		default:
520 			return -EINVAL;
521 		}
522 	} else {
523 		temp.ll = data.ll = 0;
524 		ret = 0;
525 		p = addr;
526 
527 		switch (nb) {
528 		case 8:
529 			ret |= __get_user_inatomic(temp.v[0], p++);
530 			ret |= __get_user_inatomic(temp.v[1], p++);
531 			ret |= __get_user_inatomic(temp.v[2], p++);
532 			ret |= __get_user_inatomic(temp.v[3], p++);
533 		case 4:
534 			ret |= __get_user_inatomic(temp.v[4], p++);
535 			ret |= __get_user_inatomic(temp.v[5], p++);
536 		case 2:
537 			ret |= __get_user_inatomic(temp.v[6], p++);
538 			ret |= __get_user_inatomic(temp.v[7], p++);
539 			if (unlikely(ret))
540 				return -EFAULT;
541 		}
542 
543 		switch (instr) {
544 		case EVLDD:
545 		case EVLDW:
546 		case EVLDH:
547 			data.ll = temp.ll;
548 			break;
549 		case EVLHHESPLAT:
550 			data.h[0] = temp.h[3];
551 			data.h[2] = temp.h[3];
552 			break;
553 		case EVLHHOUSPLAT:
554 		case EVLHHOSSPLAT:
555 			data.h[1] = temp.h[3];
556 			data.h[3] = temp.h[3];
557 			break;
558 		case EVLWHE:
559 			data.h[0] = temp.h[2];
560 			data.h[2] = temp.h[3];
561 			break;
562 		case EVLWHOU:
563 		case EVLWHOS:
564 			data.h[1] = temp.h[2];
565 			data.h[3] = temp.h[3];
566 			break;
567 		case EVLWWSPLAT:
568 			data.w[0] = temp.w[1];
569 			data.w[1] = temp.w[1];
570 			break;
571 		case EVLWHSPLAT:
572 			data.h[0] = temp.h[2];
573 			data.h[1] = temp.h[2];
574 			data.h[2] = temp.h[3];
575 			data.h[3] = temp.h[3];
576 			break;
577 		default:
578 			return -EINVAL;
579 		}
580 	}
581 
582 	if (flags & SW) {
583 		switch (flags & 0xf0) {
584 		case E8:
585 			SWAP(data.v[0], data.v[7]);
586 			SWAP(data.v[1], data.v[6]);
587 			SWAP(data.v[2], data.v[5]);
588 			SWAP(data.v[3], data.v[4]);
589 			break;
590 		case E4:
591 
592 			SWAP(data.v[0], data.v[3]);
593 			SWAP(data.v[1], data.v[2]);
594 			SWAP(data.v[4], data.v[7]);
595 			SWAP(data.v[5], data.v[6]);
596 			break;
597 		/* Its half word endian */
598 		default:
599 			SWAP(data.v[0], data.v[1]);
600 			SWAP(data.v[2], data.v[3]);
601 			SWAP(data.v[4], data.v[5]);
602 			SWAP(data.v[6], data.v[7]);
603 			break;
604 		}
605 	}
606 
607 	if (flags & SE) {
608 		data.w[0] = (s16)data.h[1];
609 		data.w[1] = (s16)data.h[3];
610 	}
611 
612 	/* Store result to memory or update registers */
613 	if (flags & ST) {
614 		ret = 0;
615 		p = addr;
616 		switch (nb) {
617 		case 8:
618 			ret |= __put_user_inatomic(data.v[0], p++);
619 			ret |= __put_user_inatomic(data.v[1], p++);
620 			ret |= __put_user_inatomic(data.v[2], p++);
621 			ret |= __put_user_inatomic(data.v[3], p++);
622 		case 4:
623 			ret |= __put_user_inatomic(data.v[4], p++);
624 			ret |= __put_user_inatomic(data.v[5], p++);
625 		case 2:
626 			ret |= __put_user_inatomic(data.v[6], p++);
627 			ret |= __put_user_inatomic(data.v[7], p++);
628 		}
629 		if (unlikely(ret))
630 			return -EFAULT;
631 	} else {
632 		*evr = data.w[0];
633 		regs->gpr[reg] = data.w[1];
634 	}
635 
636 	return 1;
637 }
638 #endif /* CONFIG_SPE */
639 
640 /*
641  * Called on alignment exception. Attempts to fixup
642  *
643  * Return 1 on success
644  * Return 0 if unable to handle the interrupt
645  * Return -EFAULT if data address is bad
646  */
647 
648 int fix_alignment(struct pt_regs *regs)
649 {
650 	unsigned int instr, nb, flags;
651 	unsigned int reg, areg;
652 	unsigned int dsisr;
653 	unsigned char __user *addr;
654 	unsigned long p, swiz;
655 	int ret, t;
656 	union {
657 		u64 ll;
658 		double dd;
659 		unsigned char v[8];
660 		struct {
661 			unsigned hi32;
662 			int	 low32;
663 		} x32;
664 		struct {
665 			unsigned char hi48[6];
666 			short	      low16;
667 		} x16;
668 	} data;
669 
670 	/*
671 	 * We require a complete register set, if not, then our assembly
672 	 * is broken
673 	 */
674 	CHECK_FULL_REGS(regs);
675 
676 	dsisr = regs->dsisr;
677 
678 	/* Some processors don't provide us with a DSISR we can use here,
679 	 * let's make one up from the instruction
680 	 */
681 	if (cpu_has_feature(CPU_FTR_NODSISRALIGN)) {
682 		unsigned long pc = regs->nip;
683 
684 		if (cpu_has_feature(CPU_FTR_PPC_LE) && (regs->msr & MSR_LE))
685 			pc ^= 4;
686 		if (unlikely(__get_user_inatomic(instr,
687 						 (unsigned int __user *)pc)))
688 			return -EFAULT;
689 		if (cpu_has_feature(CPU_FTR_REAL_LE) && (regs->msr & MSR_LE))
690 			instr = cpu_to_le32(instr);
691 		dsisr = make_dsisr(instr);
692 	}
693 
694 	/* extract the operation and registers from the dsisr */
695 	reg = (dsisr >> 5) & 0x1f;	/* source/dest register */
696 	areg = dsisr & 0x1f;		/* register to update */
697 
698 #ifdef CONFIG_SPE
699 	if ((instr >> 26) == 0x4)
700 		return emulate_spe(regs, reg, instr);
701 #endif
702 
703 	instr = (dsisr >> 10) & 0x7f;
704 	instr |= (dsisr >> 13) & 0x60;
705 
706 	/* Lookup the operation in our table */
707 	nb = aligninfo[instr].len;
708 	flags = aligninfo[instr].flags;
709 
710 	/* Byteswap little endian loads and stores */
711 	swiz = 0;
712 	if (regs->msr & MSR_LE) {
713 		flags ^= SW;
714 		/*
715 		 * So-called "PowerPC little endian" mode works by
716 		 * swizzling addresses rather than by actually doing
717 		 * any byte-swapping.  To emulate this, we XOR each
718 		 * byte address with 7.  We also byte-swap, because
719 		 * the processor's address swizzling depends on the
720 		 * operand size (it xors the address with 7 for bytes,
721 		 * 6 for halfwords, 4 for words, 0 for doublewords) but
722 		 * we will xor with 7 and load/store each byte separately.
723 		 */
724 		if (cpu_has_feature(CPU_FTR_PPC_LE))
725 			swiz = 7;
726 	}
727 
728 	/* DAR has the operand effective address */
729 	addr = (unsigned char __user *)regs->dar;
730 
731 	/* A size of 0 indicates an instruction we don't support, with
732 	 * the exception of DCBZ which is handled as a special case here
733 	 */
734 	if (instr == DCBZ)
735 		return emulate_dcbz(regs, addr);
736 	if (unlikely(nb == 0))
737 		return 0;
738 
739 	/* Load/Store Multiple instructions are handled in their own
740 	 * function
741 	 */
742 	if (flags & M)
743 		return emulate_multiple(regs, addr, reg, nb,
744 					flags, instr, swiz);
745 
746 	/* Verify the address of the operand */
747 	if (unlikely(user_mode(regs) &&
748 		     !access_ok((flags & ST ? VERIFY_WRITE : VERIFY_READ),
749 				addr, nb)))
750 		return -EFAULT;
751 
752 	/* Force the fprs into the save area so we can reference them */
753 	if (flags & F) {
754 		/* userland only */
755 		if (unlikely(!user_mode(regs)))
756 			return 0;
757 		flush_fp_to_thread(current);
758 	}
759 
760 	/* Special case for 16-byte FP loads and stores */
761 	if (nb == 16)
762 		return emulate_fp_pair(regs, addr, reg, flags);
763 
764 	/* If we are loading, get the data from user space, else
765 	 * get it from register values
766 	 */
767 	if (!(flags & ST)) {
768 		data.ll = 0;
769 		ret = 0;
770 		p = (unsigned long) addr;
771 		switch (nb) {
772 		case 8:
773 			ret |= __get_user_inatomic(data.v[0], SWIZ_PTR(p++));
774 			ret |= __get_user_inatomic(data.v[1], SWIZ_PTR(p++));
775 			ret |= __get_user_inatomic(data.v[2], SWIZ_PTR(p++));
776 			ret |= __get_user_inatomic(data.v[3], SWIZ_PTR(p++));
777 		case 4:
778 			ret |= __get_user_inatomic(data.v[4], SWIZ_PTR(p++));
779 			ret |= __get_user_inatomic(data.v[5], SWIZ_PTR(p++));
780 		case 2:
781 			ret |= __get_user_inatomic(data.v[6], SWIZ_PTR(p++));
782 			ret |= __get_user_inatomic(data.v[7], SWIZ_PTR(p++));
783 			if (unlikely(ret))
784 				return -EFAULT;
785 		}
786 	} else if (flags & F) {
787 		data.dd = current->thread.fpr[reg];
788 		if (flags & S) {
789 			/* Single-precision FP store requires conversion... */
790 #ifdef CONFIG_PPC_FPU
791 			preempt_disable();
792 			enable_kernel_fp();
793 			cvt_df(&data.dd, (float *)&data.v[4], &current->thread);
794 			preempt_enable();
795 #else
796 			return 0;
797 #endif
798 		}
799 	} else
800 		data.ll = regs->gpr[reg];
801 
802 	if (flags & SW) {
803 		switch (nb) {
804 		case 8:
805 			SWAP(data.v[0], data.v[7]);
806 			SWAP(data.v[1], data.v[6]);
807 			SWAP(data.v[2], data.v[5]);
808 			SWAP(data.v[3], data.v[4]);
809 			break;
810 		case 4:
811 			SWAP(data.v[4], data.v[7]);
812 			SWAP(data.v[5], data.v[6]);
813 			break;
814 		case 2:
815 			SWAP(data.v[6], data.v[7]);
816 			break;
817 		}
818 	}
819 
820 	/* Perform other misc operations like sign extension
821 	 * or floating point single precision conversion
822 	 */
823 	switch (flags & ~(U|SW)) {
824 	case LD+SE:	/* sign extending integer loads */
825 	case LD+F+SE:	/* sign extend for lfiwax */
826 		if ( nb == 2 )
827 			data.ll = data.x16.low16;
828 		else	/* nb must be 4 */
829 			data.ll = data.x32.low32;
830 		break;
831 
832 	/* Single-precision FP load requires conversion... */
833 	case LD+F+S:
834 #ifdef CONFIG_PPC_FPU
835 		preempt_disable();
836 		enable_kernel_fp();
837 		cvt_fd((float *)&data.v[4], &data.dd, &current->thread);
838 		preempt_enable();
839 #else
840 		return 0;
841 #endif
842 		break;
843 	}
844 
845 	/* Store result to memory or update registers */
846 	if (flags & ST) {
847 		ret = 0;
848 		p = (unsigned long) addr;
849 		switch (nb) {
850 		case 8:
851 			ret |= __put_user_inatomic(data.v[0], SWIZ_PTR(p++));
852 			ret |= __put_user_inatomic(data.v[1], SWIZ_PTR(p++));
853 			ret |= __put_user_inatomic(data.v[2], SWIZ_PTR(p++));
854 			ret |= __put_user_inatomic(data.v[3], SWIZ_PTR(p++));
855 		case 4:
856 			ret |= __put_user_inatomic(data.v[4], SWIZ_PTR(p++));
857 			ret |= __put_user_inatomic(data.v[5], SWIZ_PTR(p++));
858 		case 2:
859 			ret |= __put_user_inatomic(data.v[6], SWIZ_PTR(p++));
860 			ret |= __put_user_inatomic(data.v[7], SWIZ_PTR(p++));
861 		}
862 		if (unlikely(ret))
863 			return -EFAULT;
864 	} else if (flags & F)
865 		current->thread.fpr[reg] = data.dd;
866 	else
867 		regs->gpr[reg] = data.ll;
868 
869 	/* Update RA as needed */
870 	if (flags & U)
871 		regs->gpr[areg] = regs->dar;
872 
873 	return 1;
874 }
875