xref: /openbmc/linux/arch/powerpc/kernel/align.c (revision 8fa5723aa7e053d498336b48448b292fc2e0458b)
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 #define SPLT	0x80	/* VSX SPLAT load */
52 
53 /* DSISR bits reported for a DCBZ instruction: */
54 #define DCBZ	0x5f	/* 8xx/82xx dcbz faults when cache not enabled */
55 
56 #define SWAP(a, b)	(t = (a), (a) = (b), (b) = t)
57 
58 /*
59  * The PowerPC stores certain bits of the instruction that caused the
60  * alignment exception in the DSISR register.  This array maps those
61  * bits to information about the operand length and what the
62  * instruction would do.
63  */
64 static struct aligninfo aligninfo[128] = {
65 	{ 4, LD },		/* 00 0 0000: lwz / lwarx */
66 	INVALID,		/* 00 0 0001 */
67 	{ 4, ST },		/* 00 0 0010: stw */
68 	INVALID,		/* 00 0 0011 */
69 	{ 2, LD },		/* 00 0 0100: lhz */
70 	{ 2, LD+SE },		/* 00 0 0101: lha */
71 	{ 2, ST },		/* 00 0 0110: sth */
72 	{ 4, LD+M },		/* 00 0 0111: lmw */
73 	{ 4, LD+F+S },		/* 00 0 1000: lfs */
74 	{ 8, LD+F },		/* 00 0 1001: lfd */
75 	{ 4, ST+F+S },		/* 00 0 1010: stfs */
76 	{ 8, ST+F },		/* 00 0 1011: stfd */
77 	INVALID,		/* 00 0 1100 */
78 	{ 8, LD },		/* 00 0 1101: ld/ldu/lwa */
79 	INVALID,		/* 00 0 1110 */
80 	{ 8, ST },		/* 00 0 1111: std/stdu */
81 	{ 4, LD+U },		/* 00 1 0000: lwzu */
82 	INVALID,		/* 00 1 0001 */
83 	{ 4, ST+U },		/* 00 1 0010: stwu */
84 	INVALID,		/* 00 1 0011 */
85 	{ 2, LD+U },		/* 00 1 0100: lhzu */
86 	{ 2, LD+SE+U },		/* 00 1 0101: lhau */
87 	{ 2, ST+U },		/* 00 1 0110: sthu */
88 	{ 4, ST+M },		/* 00 1 0111: stmw */
89 	{ 4, LD+F+S+U },	/* 00 1 1000: lfsu */
90 	{ 8, LD+F+U },		/* 00 1 1001: lfdu */
91 	{ 4, ST+F+S+U },	/* 00 1 1010: stfsu */
92 	{ 8, ST+F+U },		/* 00 1 1011: stfdu */
93 	{ 16, LD+F },		/* 00 1 1100: lfdp */
94 	INVALID,		/* 00 1 1101 */
95 	{ 16, ST+F },		/* 00 1 1110: stfdp */
96 	INVALID,		/* 00 1 1111 */
97 	{ 8, LD },		/* 01 0 0000: ldx */
98 	INVALID,		/* 01 0 0001 */
99 	{ 8, ST },		/* 01 0 0010: stdx */
100 	INVALID,		/* 01 0 0011 */
101 	INVALID,		/* 01 0 0100 */
102 	{ 4, LD+SE },		/* 01 0 0101: lwax */
103 	INVALID,		/* 01 0 0110 */
104 	INVALID,		/* 01 0 0111 */
105 	{ 4, LD+M+HARD+SX },	/* 01 0 1000: lswx */
106 	{ 4, LD+M+HARD },	/* 01 0 1001: lswi */
107 	{ 4, ST+M+HARD+SX },	/* 01 0 1010: stswx */
108 	{ 4, ST+M+HARD },	/* 01 0 1011: stswi */
109 	INVALID,		/* 01 0 1100 */
110 	{ 8, LD+U },		/* 01 0 1101: ldu */
111 	INVALID,		/* 01 0 1110 */
112 	{ 8, ST+U },		/* 01 0 1111: stdu */
113 	{ 8, LD+U },		/* 01 1 0000: ldux */
114 	INVALID,		/* 01 1 0001 */
115 	{ 8, ST+U },		/* 01 1 0010: stdux */
116 	INVALID,		/* 01 1 0011 */
117 	INVALID,		/* 01 1 0100 */
118 	{ 4, LD+SE+U },		/* 01 1 0101: lwaux */
119 	INVALID,		/* 01 1 0110 */
120 	INVALID,		/* 01 1 0111 */
121 	INVALID,		/* 01 1 1000 */
122 	INVALID,		/* 01 1 1001 */
123 	INVALID,		/* 01 1 1010 */
124 	INVALID,		/* 01 1 1011 */
125 	INVALID,		/* 01 1 1100 */
126 	INVALID,		/* 01 1 1101 */
127 	INVALID,		/* 01 1 1110 */
128 	INVALID,		/* 01 1 1111 */
129 	INVALID,		/* 10 0 0000 */
130 	INVALID,		/* 10 0 0001 */
131 	INVALID,		/* 10 0 0010: stwcx. */
132 	INVALID,		/* 10 0 0011 */
133 	INVALID,		/* 10 0 0100 */
134 	INVALID,		/* 10 0 0101 */
135 	INVALID,		/* 10 0 0110 */
136 	INVALID,		/* 10 0 0111 */
137 	{ 4, LD+SW },		/* 10 0 1000: lwbrx */
138 	INVALID,		/* 10 0 1001 */
139 	{ 4, ST+SW },		/* 10 0 1010: stwbrx */
140 	INVALID,		/* 10 0 1011 */
141 	{ 2, LD+SW },		/* 10 0 1100: lhbrx */
142 	{ 4, LD+SE },		/* 10 0 1101  lwa */
143 	{ 2, ST+SW },		/* 10 0 1110: sthbrx */
144 	INVALID,		/* 10 0 1111 */
145 	INVALID,		/* 10 1 0000 */
146 	INVALID,		/* 10 1 0001 */
147 	INVALID,		/* 10 1 0010 */
148 	INVALID,		/* 10 1 0011 */
149 	INVALID,		/* 10 1 0100 */
150 	INVALID,		/* 10 1 0101 */
151 	INVALID,		/* 10 1 0110 */
152 	INVALID,		/* 10 1 0111 */
153 	INVALID,		/* 10 1 1000 */
154 	INVALID,		/* 10 1 1001 */
155 	INVALID,		/* 10 1 1010 */
156 	INVALID,		/* 10 1 1011 */
157 	INVALID,		/* 10 1 1100 */
158 	INVALID,		/* 10 1 1101 */
159 	INVALID,		/* 10 1 1110 */
160 	{ 0, ST+HARD },		/* 10 1 1111: dcbz */
161 	{ 4, LD },		/* 11 0 0000: lwzx */
162 	INVALID,		/* 11 0 0001 */
163 	{ 4, ST },		/* 11 0 0010: stwx */
164 	INVALID,		/* 11 0 0011 */
165 	{ 2, LD },		/* 11 0 0100: lhzx */
166 	{ 2, LD+SE },		/* 11 0 0101: lhax */
167 	{ 2, ST },		/* 11 0 0110: sthx */
168 	INVALID,		/* 11 0 0111 */
169 	{ 4, LD+F+S },		/* 11 0 1000: lfsx */
170 	{ 8, LD+F },		/* 11 0 1001: lfdx */
171 	{ 4, ST+F+S },		/* 11 0 1010: stfsx */
172 	{ 8, ST+F },		/* 11 0 1011: stfdx */
173 	{ 16, LD+F },		/* 11 0 1100: lfdpx */
174 	{ 4, LD+F+SE },		/* 11 0 1101: lfiwax */
175 	{ 16, ST+F },		/* 11 0 1110: stfdpx */
176 	{ 4, ST+F },		/* 11 0 1111: stfiwx */
177 	{ 4, LD+U },		/* 11 1 0000: lwzux */
178 	INVALID,		/* 11 1 0001 */
179 	{ 4, ST+U },		/* 11 1 0010: stwux */
180 	INVALID,		/* 11 1 0011 */
181 	{ 2, LD+U },		/* 11 1 0100: lhzux */
182 	{ 2, LD+SE+U },		/* 11 1 0101: lhaux */
183 	{ 2, ST+U },		/* 11 1 0110: sthux */
184 	INVALID,		/* 11 1 0111 */
185 	{ 4, LD+F+S+U },	/* 11 1 1000: lfsux */
186 	{ 8, LD+F+U },		/* 11 1 1001: lfdux */
187 	{ 4, ST+F+S+U },	/* 11 1 1010: stfsux */
188 	{ 8, ST+F+U },		/* 11 1 1011: stfdux */
189 	INVALID,		/* 11 1 1100 */
190 	INVALID,		/* 11 1 1101 */
191 	INVALID,		/* 11 1 1110 */
192 	INVALID,		/* 11 1 1111 */
193 };
194 
195 /*
196  * Create a DSISR value from the instruction
197  */
198 static inline unsigned make_dsisr(unsigned instr)
199 {
200 	unsigned dsisr;
201 
202 
203 	/* bits  6:15 --> 22:31 */
204 	dsisr = (instr & 0x03ff0000) >> 16;
205 
206 	if (IS_XFORM(instr)) {
207 		/* bits 29:30 --> 15:16 */
208 		dsisr |= (instr & 0x00000006) << 14;
209 		/* bit     25 -->    17 */
210 		dsisr |= (instr & 0x00000040) << 8;
211 		/* bits 21:24 --> 18:21 */
212 		dsisr |= (instr & 0x00000780) << 3;
213 	} else {
214 		/* bit      5 -->    17 */
215 		dsisr |= (instr & 0x04000000) >> 12;
216 		/* bits  1: 4 --> 18:21 */
217 		dsisr |= (instr & 0x78000000) >> 17;
218 		/* bits 30:31 --> 12:13 */
219 		if (IS_DSFORM(instr))
220 			dsisr |= (instr & 0x00000003) << 18;
221 	}
222 
223 	return dsisr;
224 }
225 
226 /*
227  * The dcbz (data cache block zero) instruction
228  * gives an alignment fault if used on non-cacheable
229  * memory.  We handle the fault mainly for the
230  * case when we are running with the cache disabled
231  * for debugging.
232  */
233 static int emulate_dcbz(struct pt_regs *regs, unsigned char __user *addr)
234 {
235 	long __user *p;
236 	int i, size;
237 
238 #ifdef __powerpc64__
239 	size = ppc64_caches.dline_size;
240 #else
241 	size = L1_CACHE_BYTES;
242 #endif
243 	p = (long __user *) (regs->dar & -size);
244 	if (user_mode(regs) && !access_ok(VERIFY_WRITE, p, size))
245 		return -EFAULT;
246 	for (i = 0; i < size / sizeof(long); ++i)
247 		if (__put_user_inatomic(0, p+i))
248 			return -EFAULT;
249 	return 1;
250 }
251 
252 /*
253  * Emulate load & store multiple instructions
254  * On 64-bit machines, these instructions only affect/use the
255  * bottom 4 bytes of each register, and the loads clear the
256  * top 4 bytes of the affected register.
257  */
258 #ifdef CONFIG_PPC64
259 #define REG_BYTE(rp, i)		*((u8 *)((rp) + ((i) >> 2)) + ((i) & 3) + 4)
260 #else
261 #define REG_BYTE(rp, i)		*((u8 *)(rp) + (i))
262 #endif
263 
264 #define SWIZ_PTR(p)		((unsigned char __user *)((p) ^ swiz))
265 
266 static int emulate_multiple(struct pt_regs *regs, unsigned char __user *addr,
267 			    unsigned int reg, unsigned int nb,
268 			    unsigned int flags, unsigned int instr,
269 			    unsigned long swiz)
270 {
271 	unsigned long *rptr;
272 	unsigned int nb0, i, bswiz;
273 	unsigned long p;
274 
275 	/*
276 	 * We do not try to emulate 8 bytes multiple as they aren't really
277 	 * available in our operating environments and we don't try to
278 	 * emulate multiples operations in kernel land as they should never
279 	 * be used/generated there at least not on unaligned boundaries
280 	 */
281 	if (unlikely((nb > 4) || !user_mode(regs)))
282 		return 0;
283 
284 	/* lmw, stmw, lswi/x, stswi/x */
285 	nb0 = 0;
286 	if (flags & HARD) {
287 		if (flags & SX) {
288 			nb = regs->xer & 127;
289 			if (nb == 0)
290 				return 1;
291 		} else {
292 			unsigned long pc = regs->nip ^ (swiz & 4);
293 
294 			if (__get_user_inatomic(instr,
295 						(unsigned int __user *)pc))
296 				return -EFAULT;
297 			if (swiz == 0 && (flags & SW))
298 				instr = cpu_to_le32(instr);
299 			nb = (instr >> 11) & 0x1f;
300 			if (nb == 0)
301 				nb = 32;
302 		}
303 		if (nb + reg * 4 > 128) {
304 			nb0 = nb + reg * 4 - 128;
305 			nb = 128 - reg * 4;
306 		}
307 	} else {
308 		/* lwm, stmw */
309 		nb = (32 - reg) * 4;
310 	}
311 
312 	if (!access_ok((flags & ST ? VERIFY_WRITE: VERIFY_READ), addr, nb+nb0))
313 		return -EFAULT;	/* bad address */
314 
315 	rptr = &regs->gpr[reg];
316 	p = (unsigned long) addr;
317 	bswiz = (flags & SW)? 3: 0;
318 
319 	if (!(flags & ST)) {
320 		/*
321 		 * This zeroes the top 4 bytes of the affected registers
322 		 * in 64-bit mode, and also zeroes out any remaining
323 		 * bytes of the last register for lsw*.
324 		 */
325 		memset(rptr, 0, ((nb + 3) / 4) * sizeof(unsigned long));
326 		if (nb0 > 0)
327 			memset(&regs->gpr[0], 0,
328 			       ((nb0 + 3) / 4) * sizeof(unsigned long));
329 
330 		for (i = 0; i < nb; ++i, ++p)
331 			if (__get_user_inatomic(REG_BYTE(rptr, i ^ bswiz),
332 						SWIZ_PTR(p)))
333 				return -EFAULT;
334 		if (nb0 > 0) {
335 			rptr = &regs->gpr[0];
336 			addr += nb;
337 			for (i = 0; i < nb0; ++i, ++p)
338 				if (__get_user_inatomic(REG_BYTE(rptr,
339 								 i ^ bswiz),
340 							SWIZ_PTR(p)))
341 					return -EFAULT;
342 		}
343 
344 	} else {
345 		for (i = 0; i < nb; ++i, ++p)
346 			if (__put_user_inatomic(REG_BYTE(rptr, i ^ bswiz),
347 						SWIZ_PTR(p)))
348 				return -EFAULT;
349 		if (nb0 > 0) {
350 			rptr = &regs->gpr[0];
351 			addr += nb;
352 			for (i = 0; i < nb0; ++i, ++p)
353 				if (__put_user_inatomic(REG_BYTE(rptr,
354 								 i ^ bswiz),
355 							SWIZ_PTR(p)))
356 					return -EFAULT;
357 		}
358 	}
359 	return 1;
360 }
361 
362 /*
363  * Emulate floating-point pair loads and stores.
364  * Only POWER6 has these instructions, and it does true little-endian,
365  * so we don't need the address swizzling.
366  */
367 static int emulate_fp_pair(unsigned char __user *addr, unsigned int reg,
368 			   unsigned int flags)
369 {
370 	char *ptr = (char *) &current->thread.TS_FPR(reg);
371 	int i, ret;
372 
373 	if (!(flags & F))
374 		return 0;
375 	if (reg & 1)
376 		return 0;	/* invalid form: FRS/FRT must be even */
377 	if (!(flags & SW)) {
378 		/* not byte-swapped - easy */
379 		if (!(flags & ST))
380 			ret = __copy_from_user(ptr, addr, 16);
381 		else
382 			ret = __copy_to_user(addr, ptr, 16);
383 	} else {
384 		/* each FPR value is byte-swapped separately */
385 		ret = 0;
386 		for (i = 0; i < 16; ++i) {
387 			if (!(flags & ST))
388 				ret |= __get_user(ptr[i^7], addr + i);
389 			else
390 				ret |= __put_user(ptr[i^7], addr + i);
391 		}
392 	}
393 	if (ret)
394 		return -EFAULT;
395 	return 1;	/* exception handled and fixed up */
396 }
397 
398 #ifdef CONFIG_SPE
399 
400 static struct aligninfo spe_aligninfo[32] = {
401 	{ 8, LD+E8 },		/* 0 00 00: evldd[x] */
402 	{ 8, LD+E4 },		/* 0 00 01: evldw[x] */
403 	{ 8, LD },		/* 0 00 10: evldh[x] */
404 	INVALID,		/* 0 00 11 */
405 	{ 2, LD },		/* 0 01 00: evlhhesplat[x] */
406 	INVALID,		/* 0 01 01 */
407 	{ 2, LD },		/* 0 01 10: evlhhousplat[x] */
408 	{ 2, LD+SE },		/* 0 01 11: evlhhossplat[x] */
409 	{ 4, LD },		/* 0 10 00: evlwhe[x] */
410 	INVALID,		/* 0 10 01 */
411 	{ 4, LD },		/* 0 10 10: evlwhou[x] */
412 	{ 4, LD+SE },		/* 0 10 11: evlwhos[x] */
413 	{ 4, LD+E4 },		/* 0 11 00: evlwwsplat[x] */
414 	INVALID,		/* 0 11 01 */
415 	{ 4, LD },		/* 0 11 10: evlwhsplat[x] */
416 	INVALID,		/* 0 11 11 */
417 
418 	{ 8, ST+E8 },		/* 1 00 00: evstdd[x] */
419 	{ 8, ST+E4 },		/* 1 00 01: evstdw[x] */
420 	{ 8, ST },		/* 1 00 10: evstdh[x] */
421 	INVALID,		/* 1 00 11 */
422 	INVALID,		/* 1 01 00 */
423 	INVALID,		/* 1 01 01 */
424 	INVALID,		/* 1 01 10 */
425 	INVALID,		/* 1 01 11 */
426 	{ 4, ST },		/* 1 10 00: evstwhe[x] */
427 	INVALID,		/* 1 10 01 */
428 	{ 4, ST },		/* 1 10 10: evstwho[x] */
429 	INVALID,		/* 1 10 11 */
430 	{ 4, ST+E4 },		/* 1 11 00: evstwwe[x] */
431 	INVALID,		/* 1 11 01 */
432 	{ 4, ST+E4 },		/* 1 11 10: evstwwo[x] */
433 	INVALID,		/* 1 11 11 */
434 };
435 
436 #define	EVLDD		0x00
437 #define	EVLDW		0x01
438 #define	EVLDH		0x02
439 #define	EVLHHESPLAT	0x04
440 #define	EVLHHOUSPLAT	0x06
441 #define	EVLHHOSSPLAT	0x07
442 #define	EVLWHE		0x08
443 #define	EVLWHOU		0x0A
444 #define	EVLWHOS		0x0B
445 #define	EVLWWSPLAT	0x0C
446 #define	EVLWHSPLAT	0x0E
447 #define	EVSTDD		0x10
448 #define	EVSTDW		0x11
449 #define	EVSTDH		0x12
450 #define	EVSTWHE		0x18
451 #define	EVSTWHO		0x1A
452 #define	EVSTWWE		0x1C
453 #define	EVSTWWO		0x1E
454 
455 /*
456  * Emulate SPE loads and stores.
457  * Only Book-E has these instructions, and it does true little-endian,
458  * so we don't need the address swizzling.
459  */
460 static int emulate_spe(struct pt_regs *regs, unsigned int reg,
461 		       unsigned int instr)
462 {
463 	int t, ret;
464 	union {
465 		u64 ll;
466 		u32 w[2];
467 		u16 h[4];
468 		u8 v[8];
469 	} data, temp;
470 	unsigned char __user *p, *addr;
471 	unsigned long *evr = &current->thread.evr[reg];
472 	unsigned int nb, flags;
473 
474 	instr = (instr >> 1) & 0x1f;
475 
476 	/* DAR has the operand effective address */
477 	addr = (unsigned char __user *)regs->dar;
478 
479 	nb = spe_aligninfo[instr].len;
480 	flags = spe_aligninfo[instr].flags;
481 
482 	/* Verify the address of the operand */
483 	if (unlikely(user_mode(regs) &&
484 		     !access_ok((flags & ST ? VERIFY_WRITE : VERIFY_READ),
485 				addr, nb)))
486 		return -EFAULT;
487 
488 	/* userland only */
489 	if (unlikely(!user_mode(regs)))
490 		return 0;
491 
492 	flush_spe_to_thread(current);
493 
494 	/* If we are loading, get the data from user space, else
495 	 * get it from register values
496 	 */
497 	if (flags & ST) {
498 		data.ll = 0;
499 		switch (instr) {
500 		case EVSTDD:
501 		case EVSTDW:
502 		case EVSTDH:
503 			data.w[0] = *evr;
504 			data.w[1] = regs->gpr[reg];
505 			break;
506 		case EVSTWHE:
507 			data.h[2] = *evr >> 16;
508 			data.h[3] = regs->gpr[reg] >> 16;
509 			break;
510 		case EVSTWHO:
511 			data.h[2] = *evr & 0xffff;
512 			data.h[3] = regs->gpr[reg] & 0xffff;
513 			break;
514 		case EVSTWWE:
515 			data.w[1] = *evr;
516 			break;
517 		case EVSTWWO:
518 			data.w[1] = regs->gpr[reg];
519 			break;
520 		default:
521 			return -EINVAL;
522 		}
523 	} else {
524 		temp.ll = data.ll = 0;
525 		ret = 0;
526 		p = addr;
527 
528 		switch (nb) {
529 		case 8:
530 			ret |= __get_user_inatomic(temp.v[0], p++);
531 			ret |= __get_user_inatomic(temp.v[1], p++);
532 			ret |= __get_user_inatomic(temp.v[2], p++);
533 			ret |= __get_user_inatomic(temp.v[3], p++);
534 		case 4:
535 			ret |= __get_user_inatomic(temp.v[4], p++);
536 			ret |= __get_user_inatomic(temp.v[5], p++);
537 		case 2:
538 			ret |= __get_user_inatomic(temp.v[6], p++);
539 			ret |= __get_user_inatomic(temp.v[7], p++);
540 			if (unlikely(ret))
541 				return -EFAULT;
542 		}
543 
544 		switch (instr) {
545 		case EVLDD:
546 		case EVLDW:
547 		case EVLDH:
548 			data.ll = temp.ll;
549 			break;
550 		case EVLHHESPLAT:
551 			data.h[0] = temp.h[3];
552 			data.h[2] = temp.h[3];
553 			break;
554 		case EVLHHOUSPLAT:
555 		case EVLHHOSSPLAT:
556 			data.h[1] = temp.h[3];
557 			data.h[3] = temp.h[3];
558 			break;
559 		case EVLWHE:
560 			data.h[0] = temp.h[2];
561 			data.h[2] = temp.h[3];
562 			break;
563 		case EVLWHOU:
564 		case EVLWHOS:
565 			data.h[1] = temp.h[2];
566 			data.h[3] = temp.h[3];
567 			break;
568 		case EVLWWSPLAT:
569 			data.w[0] = temp.w[1];
570 			data.w[1] = temp.w[1];
571 			break;
572 		case EVLWHSPLAT:
573 			data.h[0] = temp.h[2];
574 			data.h[1] = temp.h[2];
575 			data.h[2] = temp.h[3];
576 			data.h[3] = temp.h[3];
577 			break;
578 		default:
579 			return -EINVAL;
580 		}
581 	}
582 
583 	if (flags & SW) {
584 		switch (flags & 0xf0) {
585 		case E8:
586 			SWAP(data.v[0], data.v[7]);
587 			SWAP(data.v[1], data.v[6]);
588 			SWAP(data.v[2], data.v[5]);
589 			SWAP(data.v[3], data.v[4]);
590 			break;
591 		case E4:
592 
593 			SWAP(data.v[0], data.v[3]);
594 			SWAP(data.v[1], data.v[2]);
595 			SWAP(data.v[4], data.v[7]);
596 			SWAP(data.v[5], data.v[6]);
597 			break;
598 		/* Its half word endian */
599 		default:
600 			SWAP(data.v[0], data.v[1]);
601 			SWAP(data.v[2], data.v[3]);
602 			SWAP(data.v[4], data.v[5]);
603 			SWAP(data.v[6], data.v[7]);
604 			break;
605 		}
606 	}
607 
608 	if (flags & SE) {
609 		data.w[0] = (s16)data.h[1];
610 		data.w[1] = (s16)data.h[3];
611 	}
612 
613 	/* Store result to memory or update registers */
614 	if (flags & ST) {
615 		ret = 0;
616 		p = addr;
617 		switch (nb) {
618 		case 8:
619 			ret |= __put_user_inatomic(data.v[0], p++);
620 			ret |= __put_user_inatomic(data.v[1], p++);
621 			ret |= __put_user_inatomic(data.v[2], p++);
622 			ret |= __put_user_inatomic(data.v[3], p++);
623 		case 4:
624 			ret |= __put_user_inatomic(data.v[4], p++);
625 			ret |= __put_user_inatomic(data.v[5], p++);
626 		case 2:
627 			ret |= __put_user_inatomic(data.v[6], p++);
628 			ret |= __put_user_inatomic(data.v[7], p++);
629 		}
630 		if (unlikely(ret))
631 			return -EFAULT;
632 	} else {
633 		*evr = data.w[0];
634 		regs->gpr[reg] = data.w[1];
635 	}
636 
637 	return 1;
638 }
639 #endif /* CONFIG_SPE */
640 
641 #ifdef CONFIG_VSX
642 /*
643  * Emulate VSX instructions...
644  */
645 static int emulate_vsx(unsigned char __user *addr, unsigned int reg,
646 		       unsigned int areg, struct pt_regs *regs,
647 		       unsigned int flags, unsigned int length)
648 {
649 	char *ptr = (char *) &current->thread.TS_FPR(reg);
650 	int ret = 0;
651 
652 	flush_vsx_to_thread(current);
653 
654 	if (flags & ST)
655 		ret = __copy_to_user(addr, ptr, length);
656         else {
657 		if (flags & SPLT){
658 			ret = __copy_from_user(ptr, addr, length);
659 			ptr += length;
660 		}
661 		ret |= __copy_from_user(ptr, addr, length);
662 	}
663 	if (flags & U)
664 		regs->gpr[areg] = regs->dar;
665 	if (ret)
666 		return -EFAULT;
667 	return 1;
668 }
669 #endif
670 
671 /*
672  * Called on alignment exception. Attempts to fixup
673  *
674  * Return 1 on success
675  * Return 0 if unable to handle the interrupt
676  * Return -EFAULT if data address is bad
677  */
678 
679 int fix_alignment(struct pt_regs *regs)
680 {
681 	unsigned int instr, nb, flags, instruction = 0;
682 	unsigned int reg, areg;
683 	unsigned int dsisr;
684 	unsigned char __user *addr;
685 	unsigned long p, swiz;
686 	int ret, t;
687 	union {
688 		u64 ll;
689 		double dd;
690 		unsigned char v[8];
691 		struct {
692 			unsigned hi32;
693 			int	 low32;
694 		} x32;
695 		struct {
696 			unsigned char hi48[6];
697 			short	      low16;
698 		} x16;
699 	} data;
700 
701 	/*
702 	 * We require a complete register set, if not, then our assembly
703 	 * is broken
704 	 */
705 	CHECK_FULL_REGS(regs);
706 
707 	dsisr = regs->dsisr;
708 
709 	/* Some processors don't provide us with a DSISR we can use here,
710 	 * let's make one up from the instruction
711 	 */
712 	if (cpu_has_feature(CPU_FTR_NODSISRALIGN)) {
713 		unsigned long pc = regs->nip;
714 
715 		if (cpu_has_feature(CPU_FTR_PPC_LE) && (regs->msr & MSR_LE))
716 			pc ^= 4;
717 		if (unlikely(__get_user_inatomic(instr,
718 						 (unsigned int __user *)pc)))
719 			return -EFAULT;
720 		if (cpu_has_feature(CPU_FTR_REAL_LE) && (regs->msr & MSR_LE))
721 			instr = cpu_to_le32(instr);
722 		dsisr = make_dsisr(instr);
723 		instruction = instr;
724 	}
725 
726 	/* extract the operation and registers from the dsisr */
727 	reg = (dsisr >> 5) & 0x1f;	/* source/dest register */
728 	areg = dsisr & 0x1f;		/* register to update */
729 
730 #ifdef CONFIG_SPE
731 	if ((instr >> 26) == 0x4)
732 		return emulate_spe(regs, reg, instr);
733 #endif
734 
735 	instr = (dsisr >> 10) & 0x7f;
736 	instr |= (dsisr >> 13) & 0x60;
737 
738 	/* Lookup the operation in our table */
739 	nb = aligninfo[instr].len;
740 	flags = aligninfo[instr].flags;
741 
742 	/* Byteswap little endian loads and stores */
743 	swiz = 0;
744 	if (regs->msr & MSR_LE) {
745 		flags ^= SW;
746 		/*
747 		 * So-called "PowerPC little endian" mode works by
748 		 * swizzling addresses rather than by actually doing
749 		 * any byte-swapping.  To emulate this, we XOR each
750 		 * byte address with 7.  We also byte-swap, because
751 		 * the processor's address swizzling depends on the
752 		 * operand size (it xors the address with 7 for bytes,
753 		 * 6 for halfwords, 4 for words, 0 for doublewords) but
754 		 * we will xor with 7 and load/store each byte separately.
755 		 */
756 		if (cpu_has_feature(CPU_FTR_PPC_LE))
757 			swiz = 7;
758 	}
759 
760 	/* DAR has the operand effective address */
761 	addr = (unsigned char __user *)regs->dar;
762 
763 #ifdef CONFIG_VSX
764 	if ((instruction & 0xfc00003e) == 0x7c000018) {
765 		/* Additional register addressing bit (64 VSX vs 32 FPR/GPR */
766 		reg |= (instruction & 0x1) << 5;
767 		/* Simple inline decoder instead of a table */
768 		if (instruction & 0x200)
769 			nb = 16;
770 		else if (instruction & 0x080)
771 			nb = 8;
772 		else
773 			nb = 4;
774 		flags = 0;
775 		if (instruction & 0x100)
776 			flags |= ST;
777 		if (instruction & 0x040)
778 			flags |= U;
779 		/* splat load needs a special decoder */
780 		if ((instruction & 0x400) == 0){
781 			flags |= SPLT;
782 			nb = 8;
783 		}
784 		return emulate_vsx(addr, reg, areg, regs, flags, nb);
785 	}
786 #endif
787 	/* A size of 0 indicates an instruction we don't support, with
788 	 * the exception of DCBZ which is handled as a special case here
789 	 */
790 	if (instr == DCBZ)
791 		return emulate_dcbz(regs, addr);
792 	if (unlikely(nb == 0))
793 		return 0;
794 
795 	/* Load/Store Multiple instructions are handled in their own
796 	 * function
797 	 */
798 	if (flags & M)
799 		return emulate_multiple(regs, addr, reg, nb,
800 					flags, instr, swiz);
801 
802 	/* Verify the address of the operand */
803 	if (unlikely(user_mode(regs) &&
804 		     !access_ok((flags & ST ? VERIFY_WRITE : VERIFY_READ),
805 				addr, nb)))
806 		return -EFAULT;
807 
808 	/* Force the fprs into the save area so we can reference them */
809 	if (flags & F) {
810 		/* userland only */
811 		if (unlikely(!user_mode(regs)))
812 			return 0;
813 		flush_fp_to_thread(current);
814 	}
815 
816 	/* Special case for 16-byte FP loads and stores */
817 	if (nb == 16)
818 		return emulate_fp_pair(addr, reg, flags);
819 
820 	/* If we are loading, get the data from user space, else
821 	 * get it from register values
822 	 */
823 	if (!(flags & ST)) {
824 		data.ll = 0;
825 		ret = 0;
826 		p = (unsigned long) addr;
827 		switch (nb) {
828 		case 8:
829 			ret |= __get_user_inatomic(data.v[0], SWIZ_PTR(p++));
830 			ret |= __get_user_inatomic(data.v[1], SWIZ_PTR(p++));
831 			ret |= __get_user_inatomic(data.v[2], SWIZ_PTR(p++));
832 			ret |= __get_user_inatomic(data.v[3], SWIZ_PTR(p++));
833 		case 4:
834 			ret |= __get_user_inatomic(data.v[4], SWIZ_PTR(p++));
835 			ret |= __get_user_inatomic(data.v[5], SWIZ_PTR(p++));
836 		case 2:
837 			ret |= __get_user_inatomic(data.v[6], SWIZ_PTR(p++));
838 			ret |= __get_user_inatomic(data.v[7], SWIZ_PTR(p++));
839 			if (unlikely(ret))
840 				return -EFAULT;
841 		}
842 	} else if (flags & F) {
843 		data.dd = current->thread.TS_FPR(reg);
844 		if (flags & S) {
845 			/* Single-precision FP store requires conversion... */
846 #ifdef CONFIG_PPC_FPU
847 			preempt_disable();
848 			enable_kernel_fp();
849 			cvt_df(&data.dd, (float *)&data.v[4], &current->thread);
850 			preempt_enable();
851 #else
852 			return 0;
853 #endif
854 		}
855 	} else
856 		data.ll = regs->gpr[reg];
857 
858 	if (flags & SW) {
859 		switch (nb) {
860 		case 8:
861 			SWAP(data.v[0], data.v[7]);
862 			SWAP(data.v[1], data.v[6]);
863 			SWAP(data.v[2], data.v[5]);
864 			SWAP(data.v[3], data.v[4]);
865 			break;
866 		case 4:
867 			SWAP(data.v[4], data.v[7]);
868 			SWAP(data.v[5], data.v[6]);
869 			break;
870 		case 2:
871 			SWAP(data.v[6], data.v[7]);
872 			break;
873 		}
874 	}
875 
876 	/* Perform other misc operations like sign extension
877 	 * or floating point single precision conversion
878 	 */
879 	switch (flags & ~(U|SW)) {
880 	case LD+SE:	/* sign extending integer loads */
881 	case LD+F+SE:	/* sign extend for lfiwax */
882 		if ( nb == 2 )
883 			data.ll = data.x16.low16;
884 		else	/* nb must be 4 */
885 			data.ll = data.x32.low32;
886 		break;
887 
888 	/* Single-precision FP load requires conversion... */
889 	case LD+F+S:
890 #ifdef CONFIG_PPC_FPU
891 		preempt_disable();
892 		enable_kernel_fp();
893 		cvt_fd((float *)&data.v[4], &data.dd, &current->thread);
894 		preempt_enable();
895 #else
896 		return 0;
897 #endif
898 		break;
899 	}
900 
901 	/* Store result to memory or update registers */
902 	if (flags & ST) {
903 		ret = 0;
904 		p = (unsigned long) addr;
905 		switch (nb) {
906 		case 8:
907 			ret |= __put_user_inatomic(data.v[0], SWIZ_PTR(p++));
908 			ret |= __put_user_inatomic(data.v[1], SWIZ_PTR(p++));
909 			ret |= __put_user_inatomic(data.v[2], SWIZ_PTR(p++));
910 			ret |= __put_user_inatomic(data.v[3], SWIZ_PTR(p++));
911 		case 4:
912 			ret |= __put_user_inatomic(data.v[4], SWIZ_PTR(p++));
913 			ret |= __put_user_inatomic(data.v[5], SWIZ_PTR(p++));
914 		case 2:
915 			ret |= __put_user_inatomic(data.v[6], SWIZ_PTR(p++));
916 			ret |= __put_user_inatomic(data.v[7], SWIZ_PTR(p++));
917 		}
918 		if (unlikely(ret))
919 			return -EFAULT;
920 	} else if (flags & F)
921 		current->thread.TS_FPR(reg) = data.dd;
922 	else
923 		regs->gpr[reg] = data.ll;
924 
925 	/* Update RA as needed */
926 	if (flags & U)
927 		regs->gpr[areg] = regs->dar;
928 
929 	return 1;
930 }
931