xref: /linux/arch/powerpc/kernel/align.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
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/cache.h>
25 #include <asm/cputable.h>
26 #include <asm/emulated_ops.h>
27 #include <asm/switch_to.h>
28 
29 struct aligninfo {
30 	unsigned char len;
31 	unsigned char flags;
32 };
33 
34 #define IS_XFORM(inst)	(((inst) >> 26) == 31)
35 #define IS_DSFORM(inst)	(((inst) >> 26) >= 56)
36 
37 #define INVALID	{ 0, 0 }
38 
39 /* Bits in the flags field */
40 #define LD	0	/* load */
41 #define ST	1	/* store */
42 #define SE	2	/* sign-extend value, or FP ld/st as word */
43 #define F	4	/* to/from fp regs */
44 #define U	8	/* update index register */
45 #define M	0x10	/* multiple load/store */
46 #define SW	0x20	/* byte swap */
47 #define S	0x40	/* single-precision fp or... */
48 #define SX	0x40	/* ... byte count in XER */
49 #define HARD	0x80	/* string, stwcx. */
50 #define E4	0x40	/* SPE endianness is word */
51 #define E8	0x80	/* SPE endianness is double word */
52 #define SPLT	0x80	/* VSX SPLAT load */
53 
54 /* DSISR bits reported for a DCBZ instruction: */
55 #define DCBZ	0x5f	/* 8xx/82xx dcbz faults when cache not enabled */
56 
57 #define SWAP(a, b)	(t = (a), (a) = (b), (b) = t)
58 
59 /*
60  * The PowerPC stores certain bits of the instruction that caused the
61  * alignment exception in the DSISR register.  This array maps those
62  * bits to information about the operand length and what the
63  * instruction would do.
64  */
65 static struct aligninfo aligninfo[128] = {
66 	{ 4, LD },		/* 00 0 0000: lwz / lwarx */
67 	INVALID,		/* 00 0 0001 */
68 	{ 4, ST },		/* 00 0 0010: stw */
69 	INVALID,		/* 00 0 0011 */
70 	{ 2, LD },		/* 00 0 0100: lhz */
71 	{ 2, LD+SE },		/* 00 0 0101: lha */
72 	{ 2, ST },		/* 00 0 0110: sth */
73 	{ 4, LD+M },		/* 00 0 0111: lmw */
74 	{ 4, LD+F+S },		/* 00 0 1000: lfs */
75 	{ 8, LD+F },		/* 00 0 1001: lfd */
76 	{ 4, ST+F+S },		/* 00 0 1010: stfs */
77 	{ 8, ST+F },		/* 00 0 1011: stfd */
78 	INVALID,		/* 00 0 1100 */
79 	{ 8, LD },		/* 00 0 1101: ld/ldu/lwa */
80 	INVALID,		/* 00 0 1110 */
81 	{ 8, ST },		/* 00 0 1111: std/stdu */
82 	{ 4, LD+U },		/* 00 1 0000: lwzu */
83 	INVALID,		/* 00 1 0001 */
84 	{ 4, ST+U },		/* 00 1 0010: stwu */
85 	INVALID,		/* 00 1 0011 */
86 	{ 2, LD+U },		/* 00 1 0100: lhzu */
87 	{ 2, LD+SE+U },		/* 00 1 0101: lhau */
88 	{ 2, ST+U },		/* 00 1 0110: sthu */
89 	{ 4, ST+M },		/* 00 1 0111: stmw */
90 	{ 4, LD+F+S+U },	/* 00 1 1000: lfsu */
91 	{ 8, LD+F+U },		/* 00 1 1001: lfdu */
92 	{ 4, ST+F+S+U },	/* 00 1 1010: stfsu */
93 	{ 8, ST+F+U },		/* 00 1 1011: stfdu */
94 	{ 16, LD+F },		/* 00 1 1100: lfdp */
95 	INVALID,		/* 00 1 1101 */
96 	{ 16, ST+F },		/* 00 1 1110: stfdp */
97 	INVALID,		/* 00 1 1111 */
98 	{ 8, LD },		/* 01 0 0000: ldx */
99 	INVALID,		/* 01 0 0001 */
100 	{ 8, ST },		/* 01 0 0010: stdx */
101 	INVALID,		/* 01 0 0011 */
102 	INVALID,		/* 01 0 0100 */
103 	{ 4, LD+SE },		/* 01 0 0101: lwax */
104 	INVALID,		/* 01 0 0110 */
105 	INVALID,		/* 01 0 0111 */
106 	{ 4, LD+M+HARD+SX },	/* 01 0 1000: lswx */
107 	{ 4, LD+M+HARD },	/* 01 0 1001: lswi */
108 	{ 4, ST+M+HARD+SX },	/* 01 0 1010: stswx */
109 	{ 4, ST+M+HARD },	/* 01 0 1011: stswi */
110 	INVALID,		/* 01 0 1100 */
111 	{ 8, LD+U },		/* 01 0 1101: ldu */
112 	INVALID,		/* 01 0 1110 */
113 	{ 8, ST+U },		/* 01 0 1111: stdu */
114 	{ 8, LD+U },		/* 01 1 0000: ldux */
115 	INVALID,		/* 01 1 0001 */
116 	{ 8, ST+U },		/* 01 1 0010: stdux */
117 	INVALID,		/* 01 1 0011 */
118 	INVALID,		/* 01 1 0100 */
119 	{ 4, LD+SE+U },		/* 01 1 0101: lwaux */
120 	INVALID,		/* 01 1 0110 */
121 	INVALID,		/* 01 1 0111 */
122 	INVALID,		/* 01 1 1000 */
123 	INVALID,		/* 01 1 1001 */
124 	INVALID,		/* 01 1 1010 */
125 	INVALID,		/* 01 1 1011 */
126 	INVALID,		/* 01 1 1100 */
127 	INVALID,		/* 01 1 1101 */
128 	INVALID,		/* 01 1 1110 */
129 	INVALID,		/* 01 1 1111 */
130 	INVALID,		/* 10 0 0000 */
131 	INVALID,		/* 10 0 0001 */
132 	INVALID,		/* 10 0 0010: stwcx. */
133 	INVALID,		/* 10 0 0011 */
134 	INVALID,		/* 10 0 0100 */
135 	INVALID,		/* 10 0 0101 */
136 	INVALID,		/* 10 0 0110 */
137 	INVALID,		/* 10 0 0111 */
138 	{ 4, LD+SW },		/* 10 0 1000: lwbrx */
139 	INVALID,		/* 10 0 1001 */
140 	{ 4, ST+SW },		/* 10 0 1010: stwbrx */
141 	INVALID,		/* 10 0 1011 */
142 	{ 2, LD+SW },		/* 10 0 1100: lhbrx */
143 	{ 4, LD+SE },		/* 10 0 1101  lwa */
144 	{ 2, ST+SW },		/* 10 0 1110: sthbrx */
145 	INVALID,		/* 10 0 1111 */
146 	INVALID,		/* 10 1 0000 */
147 	INVALID,		/* 10 1 0001 */
148 	INVALID,		/* 10 1 0010 */
149 	INVALID,		/* 10 1 0011 */
150 	INVALID,		/* 10 1 0100 */
151 	INVALID,		/* 10 1 0101 */
152 	INVALID,		/* 10 1 0110 */
153 	INVALID,		/* 10 1 0111 */
154 	INVALID,		/* 10 1 1000 */
155 	INVALID,		/* 10 1 1001 */
156 	INVALID,		/* 10 1 1010 */
157 	INVALID,		/* 10 1 1011 */
158 	INVALID,		/* 10 1 1100 */
159 	INVALID,		/* 10 1 1101 */
160 	INVALID,		/* 10 1 1110 */
161 	{ 0, ST+HARD },		/* 10 1 1111: dcbz */
162 	{ 4, LD },		/* 11 0 0000: lwzx */
163 	INVALID,		/* 11 0 0001 */
164 	{ 4, ST },		/* 11 0 0010: stwx */
165 	INVALID,		/* 11 0 0011 */
166 	{ 2, LD },		/* 11 0 0100: lhzx */
167 	{ 2, LD+SE },		/* 11 0 0101: lhax */
168 	{ 2, ST },		/* 11 0 0110: sthx */
169 	INVALID,		/* 11 0 0111 */
170 	{ 4, LD+F+S },		/* 11 0 1000: lfsx */
171 	{ 8, LD+F },		/* 11 0 1001: lfdx */
172 	{ 4, ST+F+S },		/* 11 0 1010: stfsx */
173 	{ 8, ST+F },		/* 11 0 1011: stfdx */
174 	{ 16, LD+F },		/* 11 0 1100: lfdpx */
175 	{ 4, LD+F+SE },		/* 11 0 1101: lfiwax */
176 	{ 16, ST+F },		/* 11 0 1110: stfdpx */
177 	{ 4, ST+F },		/* 11 0 1111: stfiwx */
178 	{ 4, LD+U },		/* 11 1 0000: lwzux */
179 	INVALID,		/* 11 1 0001 */
180 	{ 4, ST+U },		/* 11 1 0010: stwux */
181 	INVALID,		/* 11 1 0011 */
182 	{ 2, LD+U },		/* 11 1 0100: lhzux */
183 	{ 2, LD+SE+U },		/* 11 1 0101: lhaux */
184 	{ 2, ST+U },		/* 11 1 0110: sthux */
185 	INVALID,		/* 11 1 0111 */
186 	{ 4, LD+F+S+U },	/* 11 1 1000: lfsux */
187 	{ 8, LD+F+U },		/* 11 1 1001: lfdux */
188 	{ 4, ST+F+S+U },	/* 11 1 1010: stfsux */
189 	{ 8, ST+F+U },		/* 11 1 1011: stfdux */
190 	INVALID,		/* 11 1 1100 */
191 	{ 4, LD+F },		/* 11 1 1101: lfiwzx */
192 	INVALID,		/* 11 1 1110 */
193 	INVALID,		/* 11 1 1111 */
194 };
195 
196 /*
197  * Create a DSISR value from the instruction
198  */
199 static inline unsigned make_dsisr(unsigned instr)
200 {
201 	unsigned dsisr;
202 
203 
204 	/* bits  6:15 --> 22:31 */
205 	dsisr = (instr & 0x03ff0000) >> 16;
206 
207 	if (IS_XFORM(instr)) {
208 		/* bits 29:30 --> 15:16 */
209 		dsisr |= (instr & 0x00000006) << 14;
210 		/* bit     25 -->    17 */
211 		dsisr |= (instr & 0x00000040) << 8;
212 		/* bits 21:24 --> 18:21 */
213 		dsisr |= (instr & 0x00000780) << 3;
214 	} else {
215 		/* bit      5 -->    17 */
216 		dsisr |= (instr & 0x04000000) >> 12;
217 		/* bits  1: 4 --> 18:21 */
218 		dsisr |= (instr & 0x78000000) >> 17;
219 		/* bits 30:31 --> 12:13 */
220 		if (IS_DSFORM(instr))
221 			dsisr |= (instr & 0x00000003) << 18;
222 	}
223 
224 	return dsisr;
225 }
226 
227 /*
228  * The dcbz (data cache block zero) instruction
229  * gives an alignment fault if used on non-cacheable
230  * memory.  We handle the fault mainly for the
231  * case when we are running with the cache disabled
232  * for debugging.
233  */
234 static int emulate_dcbz(struct pt_regs *regs, unsigned char __user *addr)
235 {
236 	long __user *p;
237 	int i, size;
238 
239 #ifdef __powerpc64__
240 	size = ppc64_caches.dline_size;
241 #else
242 	size = L1_CACHE_BYTES;
243 #endif
244 	p = (long __user *) (regs->dar & -size);
245 	if (user_mode(regs) && !access_ok(VERIFY_WRITE, p, size))
246 		return -EFAULT;
247 	for (i = 0; i < size / sizeof(long); ++i)
248 		if (__put_user_inatomic(0, p+i))
249 			return -EFAULT;
250 	return 1;
251 }
252 
253 /*
254  * Emulate load & store multiple instructions
255  * On 64-bit machines, these instructions only affect/use the
256  * bottom 4 bytes of each register, and the loads clear the
257  * top 4 bytes of the affected register.
258  */
259 #ifdef CONFIG_PPC64
260 #define REG_BYTE(rp, i)		*((u8 *)((rp) + ((i) >> 2)) + ((i) & 3) + 4)
261 #else
262 #define REG_BYTE(rp, i)		*((u8 *)(rp) + (i))
263 #endif
264 
265 #define SWIZ_PTR(p)		((unsigned char __user *)((p) ^ swiz))
266 
267 static int emulate_multiple(struct pt_regs *regs, unsigned char __user *addr,
268 			    unsigned int reg, unsigned int nb,
269 			    unsigned int flags, unsigned int instr,
270 			    unsigned long swiz)
271 {
272 	unsigned long *rptr;
273 	unsigned int nb0, i, bswiz;
274 	unsigned long p;
275 
276 	/*
277 	 * We do not try to emulate 8 bytes multiple as they aren't really
278 	 * available in our operating environments and we don't try to
279 	 * emulate multiples operations in kernel land as they should never
280 	 * be used/generated there at least not on unaligned boundaries
281 	 */
282 	if (unlikely((nb > 4) || !user_mode(regs)))
283 		return 0;
284 
285 	/* lmw, stmw, lswi/x, stswi/x */
286 	nb0 = 0;
287 	if (flags & HARD) {
288 		if (flags & SX) {
289 			nb = regs->xer & 127;
290 			if (nb == 0)
291 				return 1;
292 		} else {
293 			unsigned long pc = regs->nip ^ (swiz & 4);
294 
295 			if (__get_user_inatomic(instr,
296 						(unsigned int __user *)pc))
297 				return -EFAULT;
298 			if (swiz == 0 && (flags & SW))
299 				instr = cpu_to_le32(instr);
300 			nb = (instr >> 11) & 0x1f;
301 			if (nb == 0)
302 				nb = 32;
303 		}
304 		if (nb + reg * 4 > 128) {
305 			nb0 = nb + reg * 4 - 128;
306 			nb = 128 - reg * 4;
307 		}
308 	} else {
309 		/* lwm, stmw */
310 		nb = (32 - reg) * 4;
311 	}
312 
313 	if (!access_ok((flags & ST ? VERIFY_WRITE: VERIFY_READ), addr, nb+nb0))
314 		return -EFAULT;	/* bad address */
315 
316 	rptr = &regs->gpr[reg];
317 	p = (unsigned long) addr;
318 	bswiz = (flags & SW)? 3: 0;
319 
320 	if (!(flags & ST)) {
321 		/*
322 		 * This zeroes the top 4 bytes of the affected registers
323 		 * in 64-bit mode, and also zeroes out any remaining
324 		 * bytes of the last register for lsw*.
325 		 */
326 		memset(rptr, 0, ((nb + 3) / 4) * sizeof(unsigned long));
327 		if (nb0 > 0)
328 			memset(&regs->gpr[0], 0,
329 			       ((nb0 + 3) / 4) * sizeof(unsigned long));
330 
331 		for (i = 0; i < nb; ++i, ++p)
332 			if (__get_user_inatomic(REG_BYTE(rptr, i ^ bswiz),
333 						SWIZ_PTR(p)))
334 				return -EFAULT;
335 		if (nb0 > 0) {
336 			rptr = &regs->gpr[0];
337 			addr += nb;
338 			for (i = 0; i < nb0; ++i, ++p)
339 				if (__get_user_inatomic(REG_BYTE(rptr,
340 								 i ^ bswiz),
341 							SWIZ_PTR(p)))
342 					return -EFAULT;
343 		}
344 
345 	} else {
346 		for (i = 0; i < nb; ++i, ++p)
347 			if (__put_user_inatomic(REG_BYTE(rptr, i ^ bswiz),
348 						SWIZ_PTR(p)))
349 				return -EFAULT;
350 		if (nb0 > 0) {
351 			rptr = &regs->gpr[0];
352 			addr += nb;
353 			for (i = 0; i < nb0; ++i, ++p)
354 				if (__put_user_inatomic(REG_BYTE(rptr,
355 								 i ^ bswiz),
356 							SWIZ_PTR(p)))
357 					return -EFAULT;
358 		}
359 	}
360 	return 1;
361 }
362 
363 /*
364  * Emulate floating-point pair loads and stores.
365  * Only POWER6 has these instructions, and it does true little-endian,
366  * so we don't need the address swizzling.
367  */
368 static int emulate_fp_pair(unsigned char __user *addr, unsigned int reg,
369 			   unsigned int flags)
370 {
371 	char *ptr0 = (char *) &current->thread.TS_FPR(reg);
372 	char *ptr1 = (char *) &current->thread.TS_FPR(reg+1);
373 	int i, ret, sw = 0;
374 
375 	if (!(flags & F))
376 		return 0;
377 	if (reg & 1)
378 		return 0;	/* invalid form: FRS/FRT must be even */
379 	if (flags & SW)
380 		sw = 7;
381 	ret = 0;
382 	for (i = 0; i < 8; ++i) {
383 		if (!(flags & ST)) {
384 			ret |= __get_user(ptr0[i^sw], addr + i);
385 			ret |= __get_user(ptr1[i^sw], addr + i + 8);
386 		} else {
387 			ret |= __put_user(ptr0[i^sw], addr + i);
388 			ret |= __put_user(ptr1[i^sw], addr + i + 8);
389 		}
390 	}
391 	if (ret)
392 		return -EFAULT;
393 	return 1;	/* exception handled and fixed up */
394 }
395 
396 #ifdef CONFIG_SPE
397 
398 static struct aligninfo spe_aligninfo[32] = {
399 	{ 8, LD+E8 },		/* 0 00 00: evldd[x] */
400 	{ 8, LD+E4 },		/* 0 00 01: evldw[x] */
401 	{ 8, LD },		/* 0 00 10: evldh[x] */
402 	INVALID,		/* 0 00 11 */
403 	{ 2, LD },		/* 0 01 00: evlhhesplat[x] */
404 	INVALID,		/* 0 01 01 */
405 	{ 2, LD },		/* 0 01 10: evlhhousplat[x] */
406 	{ 2, LD+SE },		/* 0 01 11: evlhhossplat[x] */
407 	{ 4, LD },		/* 0 10 00: evlwhe[x] */
408 	INVALID,		/* 0 10 01 */
409 	{ 4, LD },		/* 0 10 10: evlwhou[x] */
410 	{ 4, LD+SE },		/* 0 10 11: evlwhos[x] */
411 	{ 4, LD+E4 },		/* 0 11 00: evlwwsplat[x] */
412 	INVALID,		/* 0 11 01 */
413 	{ 4, LD },		/* 0 11 10: evlwhsplat[x] */
414 	INVALID,		/* 0 11 11 */
415 
416 	{ 8, ST+E8 },		/* 1 00 00: evstdd[x] */
417 	{ 8, ST+E4 },		/* 1 00 01: evstdw[x] */
418 	{ 8, ST },		/* 1 00 10: evstdh[x] */
419 	INVALID,		/* 1 00 11 */
420 	INVALID,		/* 1 01 00 */
421 	INVALID,		/* 1 01 01 */
422 	INVALID,		/* 1 01 10 */
423 	INVALID,		/* 1 01 11 */
424 	{ 4, ST },		/* 1 10 00: evstwhe[x] */
425 	INVALID,		/* 1 10 01 */
426 	{ 4, ST },		/* 1 10 10: evstwho[x] */
427 	INVALID,		/* 1 10 11 */
428 	{ 4, ST+E4 },		/* 1 11 00: evstwwe[x] */
429 	INVALID,		/* 1 11 01 */
430 	{ 4, ST+E4 },		/* 1 11 10: evstwwo[x] */
431 	INVALID,		/* 1 11 11 */
432 };
433 
434 #define	EVLDD		0x00
435 #define	EVLDW		0x01
436 #define	EVLDH		0x02
437 #define	EVLHHESPLAT	0x04
438 #define	EVLHHOUSPLAT	0x06
439 #define	EVLHHOSSPLAT	0x07
440 #define	EVLWHE		0x08
441 #define	EVLWHOU		0x0A
442 #define	EVLWHOS		0x0B
443 #define	EVLWWSPLAT	0x0C
444 #define	EVLWHSPLAT	0x0E
445 #define	EVSTDD		0x10
446 #define	EVSTDW		0x11
447 #define	EVSTDH		0x12
448 #define	EVSTWHE		0x18
449 #define	EVSTWHO		0x1A
450 #define	EVSTWWE		0x1C
451 #define	EVSTWWO		0x1E
452 
453 /*
454  * Emulate SPE loads and stores.
455  * Only Book-E has these instructions, and it does true little-endian,
456  * so we don't need the address swizzling.
457  */
458 static int emulate_spe(struct pt_regs *regs, unsigned int reg,
459 		       unsigned int instr)
460 {
461 	int t, ret;
462 	union {
463 		u64 ll;
464 		u32 w[2];
465 		u16 h[4];
466 		u8 v[8];
467 	} data, temp;
468 	unsigned char __user *p, *addr;
469 	unsigned long *evr = &current->thread.evr[reg];
470 	unsigned int nb, flags;
471 
472 	instr = (instr >> 1) & 0x1f;
473 
474 	/* DAR has the operand effective address */
475 	addr = (unsigned char __user *)regs->dar;
476 
477 	nb = spe_aligninfo[instr].len;
478 	flags = spe_aligninfo[instr].flags;
479 
480 	/* Verify the address of the operand */
481 	if (unlikely(user_mode(regs) &&
482 		     !access_ok((flags & ST ? VERIFY_WRITE : VERIFY_READ),
483 				addr, nb)))
484 		return -EFAULT;
485 
486 	/* userland only */
487 	if (unlikely(!user_mode(regs)))
488 		return 0;
489 
490 	flush_spe_to_thread(current);
491 
492 	/* If we are loading, get the data from user space, else
493 	 * get it from register values
494 	 */
495 	if (flags & ST) {
496 		data.ll = 0;
497 		switch (instr) {
498 		case EVSTDD:
499 		case EVSTDW:
500 		case EVSTDH:
501 			data.w[0] = *evr;
502 			data.w[1] = regs->gpr[reg];
503 			break;
504 		case EVSTWHE:
505 			data.h[2] = *evr >> 16;
506 			data.h[3] = regs->gpr[reg] >> 16;
507 			break;
508 		case EVSTWHO:
509 			data.h[2] = *evr & 0xffff;
510 			data.h[3] = regs->gpr[reg] & 0xffff;
511 			break;
512 		case EVSTWWE:
513 			data.w[1] = *evr;
514 			break;
515 		case EVSTWWO:
516 			data.w[1] = regs->gpr[reg];
517 			break;
518 		default:
519 			return -EINVAL;
520 		}
521 	} else {
522 		temp.ll = data.ll = 0;
523 		ret = 0;
524 		p = addr;
525 
526 		switch (nb) {
527 		case 8:
528 			ret |= __get_user_inatomic(temp.v[0], p++);
529 			ret |= __get_user_inatomic(temp.v[1], p++);
530 			ret |= __get_user_inatomic(temp.v[2], p++);
531 			ret |= __get_user_inatomic(temp.v[3], p++);
532 		case 4:
533 			ret |= __get_user_inatomic(temp.v[4], p++);
534 			ret |= __get_user_inatomic(temp.v[5], p++);
535 		case 2:
536 			ret |= __get_user_inatomic(temp.v[6], p++);
537 			ret |= __get_user_inatomic(temp.v[7], p++);
538 			if (unlikely(ret))
539 				return -EFAULT;
540 		}
541 
542 		switch (instr) {
543 		case EVLDD:
544 		case EVLDW:
545 		case EVLDH:
546 			data.ll = temp.ll;
547 			break;
548 		case EVLHHESPLAT:
549 			data.h[0] = temp.h[3];
550 			data.h[2] = temp.h[3];
551 			break;
552 		case EVLHHOUSPLAT:
553 		case EVLHHOSSPLAT:
554 			data.h[1] = temp.h[3];
555 			data.h[3] = temp.h[3];
556 			break;
557 		case EVLWHE:
558 			data.h[0] = temp.h[2];
559 			data.h[2] = temp.h[3];
560 			break;
561 		case EVLWHOU:
562 		case EVLWHOS:
563 			data.h[1] = temp.h[2];
564 			data.h[3] = temp.h[3];
565 			break;
566 		case EVLWWSPLAT:
567 			data.w[0] = temp.w[1];
568 			data.w[1] = temp.w[1];
569 			break;
570 		case EVLWHSPLAT:
571 			data.h[0] = temp.h[2];
572 			data.h[1] = temp.h[2];
573 			data.h[2] = temp.h[3];
574 			data.h[3] = temp.h[3];
575 			break;
576 		default:
577 			return -EINVAL;
578 		}
579 	}
580 
581 	if (flags & SW) {
582 		switch (flags & 0xf0) {
583 		case E8:
584 			SWAP(data.v[0], data.v[7]);
585 			SWAP(data.v[1], data.v[6]);
586 			SWAP(data.v[2], data.v[5]);
587 			SWAP(data.v[3], data.v[4]);
588 			break;
589 		case E4:
590 
591 			SWAP(data.v[0], data.v[3]);
592 			SWAP(data.v[1], data.v[2]);
593 			SWAP(data.v[4], data.v[7]);
594 			SWAP(data.v[5], data.v[6]);
595 			break;
596 		/* Its half word endian */
597 		default:
598 			SWAP(data.v[0], data.v[1]);
599 			SWAP(data.v[2], data.v[3]);
600 			SWAP(data.v[4], data.v[5]);
601 			SWAP(data.v[6], data.v[7]);
602 			break;
603 		}
604 	}
605 
606 	if (flags & SE) {
607 		data.w[0] = (s16)data.h[1];
608 		data.w[1] = (s16)data.h[3];
609 	}
610 
611 	/* Store result to memory or update registers */
612 	if (flags & ST) {
613 		ret = 0;
614 		p = addr;
615 		switch (nb) {
616 		case 8:
617 			ret |= __put_user_inatomic(data.v[0], p++);
618 			ret |= __put_user_inatomic(data.v[1], p++);
619 			ret |= __put_user_inatomic(data.v[2], p++);
620 			ret |= __put_user_inatomic(data.v[3], p++);
621 		case 4:
622 			ret |= __put_user_inatomic(data.v[4], p++);
623 			ret |= __put_user_inatomic(data.v[5], p++);
624 		case 2:
625 			ret |= __put_user_inatomic(data.v[6], p++);
626 			ret |= __put_user_inatomic(data.v[7], p++);
627 		}
628 		if (unlikely(ret))
629 			return -EFAULT;
630 	} else {
631 		*evr = data.w[0];
632 		regs->gpr[reg] = data.w[1];
633 	}
634 
635 	return 1;
636 }
637 #endif /* CONFIG_SPE */
638 
639 #ifdef CONFIG_VSX
640 /*
641  * Emulate VSX instructions...
642  */
643 static int emulate_vsx(unsigned char __user *addr, unsigned int reg,
644 		       unsigned int areg, struct pt_regs *regs,
645 		       unsigned int flags, unsigned int length,
646 		       unsigned int elsize)
647 {
648 	char *ptr;
649 	unsigned long *lptr;
650 	int ret = 0;
651 	int sw = 0;
652 	int i, j;
653 
654 	flush_vsx_to_thread(current);
655 
656 	if (reg < 32)
657 		ptr = (char *) &current->thread.TS_FPR(reg);
658 	else
659 		ptr = (char *) &current->thread.vr[reg - 32];
660 
661 	lptr = (unsigned long *) ptr;
662 
663 	if (flags & SW)
664 		sw = elsize-1;
665 
666 	for (j = 0; j < length; j += elsize) {
667 		for (i = 0; i < elsize; ++i) {
668 			if (flags & ST)
669 				ret |= __put_user(ptr[i^sw], addr + i);
670 			else
671 				ret |= __get_user(ptr[i^sw], addr + i);
672 		}
673 		ptr  += elsize;
674 		addr += elsize;
675 	}
676 
677 	if (!ret) {
678 		if (flags & U)
679 			regs->gpr[areg] = regs->dar;
680 
681 		/* Splat load copies the same data to top and bottom 8 bytes */
682 		if (flags & SPLT)
683 			lptr[1] = lptr[0];
684 		/* For 8 byte loads, zero the top 8 bytes */
685 		else if (!(flags & ST) && (8 == length))
686 			lptr[1] = 0;
687 	} else
688 		return -EFAULT;
689 
690 	return 1;
691 }
692 #endif
693 
694 /*
695  * Called on alignment exception. Attempts to fixup
696  *
697  * Return 1 on success
698  * Return 0 if unable to handle the interrupt
699  * Return -EFAULT if data address is bad
700  */
701 
702 int fix_alignment(struct pt_regs *regs)
703 {
704 	unsigned int instr, nb, flags, instruction = 0;
705 	unsigned int reg, areg;
706 	unsigned int dsisr;
707 	unsigned char __user *addr;
708 	unsigned long p, swiz;
709 	int ret, t;
710 	union {
711 		u64 ll;
712 		double dd;
713 		unsigned char v[8];
714 		struct {
715 			unsigned hi32;
716 			int	 low32;
717 		} x32;
718 		struct {
719 			unsigned char hi48[6];
720 			short	      low16;
721 		} x16;
722 	} data;
723 
724 	/*
725 	 * We require a complete register set, if not, then our assembly
726 	 * is broken
727 	 */
728 	CHECK_FULL_REGS(regs);
729 
730 	dsisr = regs->dsisr;
731 
732 	/* Some processors don't provide us with a DSISR we can use here,
733 	 * let's make one up from the instruction
734 	 */
735 	if (cpu_has_feature(CPU_FTR_NODSISRALIGN)) {
736 		unsigned long pc = regs->nip;
737 
738 		if (cpu_has_feature(CPU_FTR_PPC_LE) && (regs->msr & MSR_LE))
739 			pc ^= 4;
740 		if (unlikely(__get_user_inatomic(instr,
741 						 (unsigned int __user *)pc)))
742 			return -EFAULT;
743 		if (cpu_has_feature(CPU_FTR_REAL_LE) && (regs->msr & MSR_LE))
744 			instr = cpu_to_le32(instr);
745 		dsisr = make_dsisr(instr);
746 		instruction = instr;
747 	}
748 
749 	/* extract the operation and registers from the dsisr */
750 	reg = (dsisr >> 5) & 0x1f;	/* source/dest register */
751 	areg = dsisr & 0x1f;		/* register to update */
752 
753 #ifdef CONFIG_SPE
754 	if ((instr >> 26) == 0x4) {
755 		PPC_WARN_ALIGNMENT(spe, regs);
756 		return emulate_spe(regs, reg, instr);
757 	}
758 #endif
759 
760 	instr = (dsisr >> 10) & 0x7f;
761 	instr |= (dsisr >> 13) & 0x60;
762 
763 	/* Lookup the operation in our table */
764 	nb = aligninfo[instr].len;
765 	flags = aligninfo[instr].flags;
766 
767 	/* Byteswap little endian loads and stores */
768 	swiz = 0;
769 	if (regs->msr & MSR_LE) {
770 		flags ^= SW;
771 		/*
772 		 * So-called "PowerPC little endian" mode works by
773 		 * swizzling addresses rather than by actually doing
774 		 * any byte-swapping.  To emulate this, we XOR each
775 		 * byte address with 7.  We also byte-swap, because
776 		 * the processor's address swizzling depends on the
777 		 * operand size (it xors the address with 7 for bytes,
778 		 * 6 for halfwords, 4 for words, 0 for doublewords) but
779 		 * we will xor with 7 and load/store each byte separately.
780 		 */
781 		if (cpu_has_feature(CPU_FTR_PPC_LE))
782 			swiz = 7;
783 	}
784 
785 	/* DAR has the operand effective address */
786 	addr = (unsigned char __user *)regs->dar;
787 
788 #ifdef CONFIG_VSX
789 	if ((instruction & 0xfc00003e) == 0x7c000018) {
790 		unsigned int elsize;
791 
792 		/* Additional register addressing bit (64 VSX vs 32 FPR/GPR) */
793 		reg |= (instruction & 0x1) << 5;
794 		/* Simple inline decoder instead of a table */
795 		/* VSX has only 8 and 16 byte memory accesses */
796 		nb = 8;
797 		if (instruction & 0x200)
798 			nb = 16;
799 
800 		/* Vector stores in little-endian mode swap individual
801 		   elements, so process them separately */
802 		elsize = 4;
803 		if (instruction & 0x80)
804 			elsize = 8;
805 
806 		flags = 0;
807 		if (regs->msr & MSR_LE)
808 			flags |= SW;
809 		if (instruction & 0x100)
810 			flags |= ST;
811 		if (instruction & 0x040)
812 			flags |= U;
813 		/* splat load needs a special decoder */
814 		if ((instruction & 0x400) == 0){
815 			flags |= SPLT;
816 			nb = 8;
817 		}
818 		PPC_WARN_ALIGNMENT(vsx, regs);
819 		return emulate_vsx(addr, reg, areg, regs, flags, nb, elsize);
820 	}
821 #endif
822 	/* A size of 0 indicates an instruction we don't support, with
823 	 * the exception of DCBZ which is handled as a special case here
824 	 */
825 	if (instr == DCBZ) {
826 		PPC_WARN_ALIGNMENT(dcbz, regs);
827 		return emulate_dcbz(regs, addr);
828 	}
829 	if (unlikely(nb == 0))
830 		return 0;
831 
832 	/* Load/Store Multiple instructions are handled in their own
833 	 * function
834 	 */
835 	if (flags & M) {
836 		PPC_WARN_ALIGNMENT(multiple, regs);
837 		return emulate_multiple(regs, addr, reg, nb,
838 					flags, instr, swiz);
839 	}
840 
841 	/* Verify the address of the operand */
842 	if (unlikely(user_mode(regs) &&
843 		     !access_ok((flags & ST ? VERIFY_WRITE : VERIFY_READ),
844 				addr, nb)))
845 		return -EFAULT;
846 
847 	/* Force the fprs into the save area so we can reference them */
848 	if (flags & F) {
849 		/* userland only */
850 		if (unlikely(!user_mode(regs)))
851 			return 0;
852 		flush_fp_to_thread(current);
853 	}
854 
855 	/* Special case for 16-byte FP loads and stores */
856 	if (nb == 16) {
857 		PPC_WARN_ALIGNMENT(fp_pair, regs);
858 		return emulate_fp_pair(addr, reg, flags);
859 	}
860 
861 	PPC_WARN_ALIGNMENT(unaligned, regs);
862 
863 	/* If we are loading, get the data from user space, else
864 	 * get it from register values
865 	 */
866 	if (!(flags & ST)) {
867 		data.ll = 0;
868 		ret = 0;
869 		p = (unsigned long) addr;
870 		switch (nb) {
871 		case 8:
872 			ret |= __get_user_inatomic(data.v[0], SWIZ_PTR(p++));
873 			ret |= __get_user_inatomic(data.v[1], SWIZ_PTR(p++));
874 			ret |= __get_user_inatomic(data.v[2], SWIZ_PTR(p++));
875 			ret |= __get_user_inatomic(data.v[3], SWIZ_PTR(p++));
876 		case 4:
877 			ret |= __get_user_inatomic(data.v[4], SWIZ_PTR(p++));
878 			ret |= __get_user_inatomic(data.v[5], SWIZ_PTR(p++));
879 		case 2:
880 			ret |= __get_user_inatomic(data.v[6], SWIZ_PTR(p++));
881 			ret |= __get_user_inatomic(data.v[7], SWIZ_PTR(p++));
882 			if (unlikely(ret))
883 				return -EFAULT;
884 		}
885 	} else if (flags & F) {
886 		data.dd = current->thread.TS_FPR(reg);
887 		if (flags & S) {
888 			/* Single-precision FP store requires conversion... */
889 #ifdef CONFIG_PPC_FPU
890 			preempt_disable();
891 			enable_kernel_fp();
892 			cvt_df(&data.dd, (float *)&data.v[4]);
893 			preempt_enable();
894 #else
895 			return 0;
896 #endif
897 		}
898 	} else
899 		data.ll = regs->gpr[reg];
900 
901 	if (flags & SW) {
902 		switch (nb) {
903 		case 8:
904 			SWAP(data.v[0], data.v[7]);
905 			SWAP(data.v[1], data.v[6]);
906 			SWAP(data.v[2], data.v[5]);
907 			SWAP(data.v[3], data.v[4]);
908 			break;
909 		case 4:
910 			SWAP(data.v[4], data.v[7]);
911 			SWAP(data.v[5], data.v[6]);
912 			break;
913 		case 2:
914 			SWAP(data.v[6], data.v[7]);
915 			break;
916 		}
917 	}
918 
919 	/* Perform other misc operations like sign extension
920 	 * or floating point single precision conversion
921 	 */
922 	switch (flags & ~(U|SW)) {
923 	case LD+SE:	/* sign extending integer loads */
924 	case LD+F+SE:	/* sign extend for lfiwax */
925 		if ( nb == 2 )
926 			data.ll = data.x16.low16;
927 		else	/* nb must be 4 */
928 			data.ll = data.x32.low32;
929 		break;
930 
931 	/* Single-precision FP load requires conversion... */
932 	case LD+F+S:
933 #ifdef CONFIG_PPC_FPU
934 		preempt_disable();
935 		enable_kernel_fp();
936 		cvt_fd((float *)&data.v[4], &data.dd);
937 		preempt_enable();
938 #else
939 		return 0;
940 #endif
941 		break;
942 	}
943 
944 	/* Store result to memory or update registers */
945 	if (flags & ST) {
946 		ret = 0;
947 		p = (unsigned long) addr;
948 		switch (nb) {
949 		case 8:
950 			ret |= __put_user_inatomic(data.v[0], SWIZ_PTR(p++));
951 			ret |= __put_user_inatomic(data.v[1], SWIZ_PTR(p++));
952 			ret |= __put_user_inatomic(data.v[2], SWIZ_PTR(p++));
953 			ret |= __put_user_inatomic(data.v[3], SWIZ_PTR(p++));
954 		case 4:
955 			ret |= __put_user_inatomic(data.v[4], SWIZ_PTR(p++));
956 			ret |= __put_user_inatomic(data.v[5], SWIZ_PTR(p++));
957 		case 2:
958 			ret |= __put_user_inatomic(data.v[6], SWIZ_PTR(p++));
959 			ret |= __put_user_inatomic(data.v[7], SWIZ_PTR(p++));
960 		}
961 		if (unlikely(ret))
962 			return -EFAULT;
963 	} else if (flags & F)
964 		current->thread.TS_FPR(reg) = data.dd;
965 	else
966 		regs->gpr[reg] = data.ll;
967 
968 	/* Update RA as needed */
969 	if (flags & U)
970 		regs->gpr[areg] = regs->dar;
971 
972 	return 1;
973 }
974