xref: /linux/arch/powerpc/kernel/align.c (revision 3932b9ca55b0be314a36d3e84faff3e823c081f5)
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 #include <asm/disassemble.h>
29 
30 struct aligninfo {
31 	unsigned char len;
32 	unsigned char flags;
33 };
34 
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 /*
57  * The PowerPC stores certain bits of the instruction that caused the
58  * alignment exception in the DSISR register.  This array maps those
59  * bits to information about the operand length and what the
60  * instruction would do.
61  */
62 static struct aligninfo aligninfo[128] = {
63 	{ 4, LD },		/* 00 0 0000: lwz / lwarx */
64 	INVALID,		/* 00 0 0001 */
65 	{ 4, ST },		/* 00 0 0010: stw */
66 	INVALID,		/* 00 0 0011 */
67 	{ 2, LD },		/* 00 0 0100: lhz */
68 	{ 2, LD+SE },		/* 00 0 0101: lha */
69 	{ 2, ST },		/* 00 0 0110: sth */
70 	{ 4, LD+M },		/* 00 0 0111: lmw */
71 	{ 4, LD+F+S },		/* 00 0 1000: lfs */
72 	{ 8, LD+F },		/* 00 0 1001: lfd */
73 	{ 4, ST+F+S },		/* 00 0 1010: stfs */
74 	{ 8, ST+F },		/* 00 0 1011: stfd */
75 	{ 16, LD },		/* 00 0 1100: lq */
76 	{ 8, LD },		/* 00 0 1101: ld/ldu/lwa */
77 	INVALID,		/* 00 0 1110 */
78 	{ 8, ST },		/* 00 0 1111: std/stdu */
79 	{ 4, LD+U },		/* 00 1 0000: lwzu */
80 	INVALID,		/* 00 1 0001 */
81 	{ 4, ST+U },		/* 00 1 0010: stwu */
82 	INVALID,		/* 00 1 0011 */
83 	{ 2, LD+U },		/* 00 1 0100: lhzu */
84 	{ 2, LD+SE+U },		/* 00 1 0101: lhau */
85 	{ 2, ST+U },		/* 00 1 0110: sthu */
86 	{ 4, ST+M },		/* 00 1 0111: stmw */
87 	{ 4, LD+F+S+U },	/* 00 1 1000: lfsu */
88 	{ 8, LD+F+U },		/* 00 1 1001: lfdu */
89 	{ 4, ST+F+S+U },	/* 00 1 1010: stfsu */
90 	{ 8, ST+F+U },		/* 00 1 1011: stfdu */
91 	{ 16, LD+F },		/* 00 1 1100: lfdp */
92 	INVALID,		/* 00 1 1101 */
93 	{ 16, ST+F },		/* 00 1 1110: stfdp */
94 	INVALID,		/* 00 1 1111 */
95 	{ 8, LD },		/* 01 0 0000: ldx */
96 	INVALID,		/* 01 0 0001 */
97 	{ 8, ST },		/* 01 0 0010: stdx */
98 	INVALID,		/* 01 0 0011 */
99 	INVALID,		/* 01 0 0100 */
100 	{ 4, LD+SE },		/* 01 0 0101: lwax */
101 	INVALID,		/* 01 0 0110 */
102 	INVALID,		/* 01 0 0111 */
103 	{ 4, LD+M+HARD+SX },	/* 01 0 1000: lswx */
104 	{ 4, LD+M+HARD },	/* 01 0 1001: lswi */
105 	{ 4, ST+M+HARD+SX },	/* 01 0 1010: stswx */
106 	{ 4, ST+M+HARD },	/* 01 0 1011: stswi */
107 	INVALID,		/* 01 0 1100 */
108 	{ 8, LD+U },		/* 01 0 1101: ldu */
109 	INVALID,		/* 01 0 1110 */
110 	{ 8, ST+U },		/* 01 0 1111: stdu */
111 	{ 8, LD+U },		/* 01 1 0000: ldux */
112 	INVALID,		/* 01 1 0001 */
113 	{ 8, ST+U },		/* 01 1 0010: stdux */
114 	INVALID,		/* 01 1 0011 */
115 	INVALID,		/* 01 1 0100 */
116 	{ 4, LD+SE+U },		/* 01 1 0101: lwaux */
117 	INVALID,		/* 01 1 0110 */
118 	INVALID,		/* 01 1 0111 */
119 	INVALID,		/* 01 1 1000 */
120 	INVALID,		/* 01 1 1001 */
121 	INVALID,		/* 01 1 1010 */
122 	INVALID,		/* 01 1 1011 */
123 	INVALID,		/* 01 1 1100 */
124 	INVALID,		/* 01 1 1101 */
125 	INVALID,		/* 01 1 1110 */
126 	INVALID,		/* 01 1 1111 */
127 	INVALID,		/* 10 0 0000 */
128 	INVALID,		/* 10 0 0001 */
129 	INVALID,		/* 10 0 0010: stwcx. */
130 	INVALID,		/* 10 0 0011 */
131 	INVALID,		/* 10 0 0100 */
132 	INVALID,		/* 10 0 0101 */
133 	INVALID,		/* 10 0 0110 */
134 	INVALID,		/* 10 0 0111 */
135 	{ 4, LD+SW },		/* 10 0 1000: lwbrx */
136 	INVALID,		/* 10 0 1001 */
137 	{ 4, ST+SW },		/* 10 0 1010: stwbrx */
138 	INVALID,		/* 10 0 1011 */
139 	{ 2, LD+SW },		/* 10 0 1100: lhbrx */
140 	{ 4, LD+SE },		/* 10 0 1101  lwa */
141 	{ 2, ST+SW },		/* 10 0 1110: sthbrx */
142 	{ 16, ST },		/* 10 0 1111: stq */
143 	INVALID,		/* 10 1 0000 */
144 	INVALID,		/* 10 1 0001 */
145 	INVALID,		/* 10 1 0010 */
146 	INVALID,		/* 10 1 0011 */
147 	INVALID,		/* 10 1 0100 */
148 	INVALID,		/* 10 1 0101 */
149 	INVALID,		/* 10 1 0110 */
150 	INVALID,		/* 10 1 0111 */
151 	INVALID,		/* 10 1 1000 */
152 	INVALID,		/* 10 1 1001 */
153 	INVALID,		/* 10 1 1010 */
154 	INVALID,		/* 10 1 1011 */
155 	INVALID,		/* 10 1 1100 */
156 	INVALID,		/* 10 1 1101 */
157 	INVALID,		/* 10 1 1110 */
158 	{ 0, ST+HARD },		/* 10 1 1111: dcbz */
159 	{ 4, LD },		/* 11 0 0000: lwzx */
160 	INVALID,		/* 11 0 0001 */
161 	{ 4, ST },		/* 11 0 0010: stwx */
162 	INVALID,		/* 11 0 0011 */
163 	{ 2, LD },		/* 11 0 0100: lhzx */
164 	{ 2, LD+SE },		/* 11 0 0101: lhax */
165 	{ 2, ST },		/* 11 0 0110: sthx */
166 	INVALID,		/* 11 0 0111 */
167 	{ 4, LD+F+S },		/* 11 0 1000: lfsx */
168 	{ 8, LD+F },		/* 11 0 1001: lfdx */
169 	{ 4, ST+F+S },		/* 11 0 1010: stfsx */
170 	{ 8, ST+F },		/* 11 0 1011: stfdx */
171 	{ 16, LD+F },		/* 11 0 1100: lfdpx */
172 	{ 4, LD+F+SE },		/* 11 0 1101: lfiwax */
173 	{ 16, ST+F },		/* 11 0 1110: stfdpx */
174 	{ 4, ST+F },		/* 11 0 1111: stfiwx */
175 	{ 4, LD+U },		/* 11 1 0000: lwzux */
176 	INVALID,		/* 11 1 0001 */
177 	{ 4, ST+U },		/* 11 1 0010: stwux */
178 	INVALID,		/* 11 1 0011 */
179 	{ 2, LD+U },		/* 11 1 0100: lhzux */
180 	{ 2, LD+SE+U },		/* 11 1 0101: lhaux */
181 	{ 2, ST+U },		/* 11 1 0110: sthux */
182 	INVALID,		/* 11 1 0111 */
183 	{ 4, LD+F+S+U },	/* 11 1 1000: lfsux */
184 	{ 8, LD+F+U },		/* 11 1 1001: lfdux */
185 	{ 4, ST+F+S+U },	/* 11 1 1010: stfsux */
186 	{ 8, ST+F+U },		/* 11 1 1011: stfdux */
187 	INVALID,		/* 11 1 1100 */
188 	{ 4, LD+F },		/* 11 1 1101: lfiwzx */
189 	INVALID,		/* 11 1 1110 */
190 	INVALID,		/* 11 1 1111 */
191 };
192 
193 /*
194  * The dcbz (data cache block zero) instruction
195  * gives an alignment fault if used on non-cacheable
196  * memory.  We handle the fault mainly for the
197  * case when we are running with the cache disabled
198  * for debugging.
199  */
200 static int emulate_dcbz(struct pt_regs *regs, unsigned char __user *addr)
201 {
202 	long __user *p;
203 	int i, size;
204 
205 #ifdef __powerpc64__
206 	size = ppc64_caches.dline_size;
207 #else
208 	size = L1_CACHE_BYTES;
209 #endif
210 	p = (long __user *) (regs->dar & -size);
211 	if (user_mode(regs) && !access_ok(VERIFY_WRITE, p, size))
212 		return -EFAULT;
213 	for (i = 0; i < size / sizeof(long); ++i)
214 		if (__put_user_inatomic(0, p+i))
215 			return -EFAULT;
216 	return 1;
217 }
218 
219 /*
220  * Emulate load & store multiple instructions
221  * On 64-bit machines, these instructions only affect/use the
222  * bottom 4 bytes of each register, and the loads clear the
223  * top 4 bytes of the affected register.
224  */
225 #ifdef __BIG_ENDIAN__
226 #ifdef CONFIG_PPC64
227 #define REG_BYTE(rp, i)		*((u8 *)((rp) + ((i) >> 2)) + ((i) & 3) + 4)
228 #else
229 #define REG_BYTE(rp, i)		*((u8 *)(rp) + (i))
230 #endif
231 #endif
232 
233 #ifdef __LITTLE_ENDIAN__
234 #define REG_BYTE(rp, i)		(*(((u8 *)((rp) + ((i)>>2)) + ((i)&3))))
235 #endif
236 
237 #define SWIZ_PTR(p)		((unsigned char __user *)((p) ^ swiz))
238 
239 static int emulate_multiple(struct pt_regs *regs, unsigned char __user *addr,
240 			    unsigned int reg, unsigned int nb,
241 			    unsigned int flags, unsigned int instr,
242 			    unsigned long swiz)
243 {
244 	unsigned long *rptr;
245 	unsigned int nb0, i, bswiz;
246 	unsigned long p;
247 
248 	/*
249 	 * We do not try to emulate 8 bytes multiple as they aren't really
250 	 * available in our operating environments and we don't try to
251 	 * emulate multiples operations in kernel land as they should never
252 	 * be used/generated there at least not on unaligned boundaries
253 	 */
254 	if (unlikely((nb > 4) || !user_mode(regs)))
255 		return 0;
256 
257 	/* lmw, stmw, lswi/x, stswi/x */
258 	nb0 = 0;
259 	if (flags & HARD) {
260 		if (flags & SX) {
261 			nb = regs->xer & 127;
262 			if (nb == 0)
263 				return 1;
264 		} else {
265 			unsigned long pc = regs->nip ^ (swiz & 4);
266 
267 			if (__get_user_inatomic(instr,
268 						(unsigned int __user *)pc))
269 				return -EFAULT;
270 			if (swiz == 0 && (flags & SW))
271 				instr = cpu_to_le32(instr);
272 			nb = (instr >> 11) & 0x1f;
273 			if (nb == 0)
274 				nb = 32;
275 		}
276 		if (nb + reg * 4 > 128) {
277 			nb0 = nb + reg * 4 - 128;
278 			nb = 128 - reg * 4;
279 		}
280 #ifdef __LITTLE_ENDIAN__
281 		/*
282 		 *  String instructions are endian neutral but the code
283 		 *  below is not.  Force byte swapping on so that the
284 		 *  effects of swizzling are undone in the load/store
285 		 *  loops below.
286 		 */
287 		flags ^= SW;
288 #endif
289 	} else {
290 		/* lwm, stmw */
291 		nb = (32 - reg) * 4;
292 	}
293 
294 	if (!access_ok((flags & ST ? VERIFY_WRITE: VERIFY_READ), addr, nb+nb0))
295 		return -EFAULT;	/* bad address */
296 
297 	rptr = &regs->gpr[reg];
298 	p = (unsigned long) addr;
299 	bswiz = (flags & SW)? 3: 0;
300 
301 	if (!(flags & ST)) {
302 		/*
303 		 * This zeroes the top 4 bytes of the affected registers
304 		 * in 64-bit mode, and also zeroes out any remaining
305 		 * bytes of the last register for lsw*.
306 		 */
307 		memset(rptr, 0, ((nb + 3) / 4) * sizeof(unsigned long));
308 		if (nb0 > 0)
309 			memset(&regs->gpr[0], 0,
310 			       ((nb0 + 3) / 4) * sizeof(unsigned long));
311 
312 		for (i = 0; i < nb; ++i, ++p)
313 			if (__get_user_inatomic(REG_BYTE(rptr, i ^ bswiz),
314 						SWIZ_PTR(p)))
315 				return -EFAULT;
316 		if (nb0 > 0) {
317 			rptr = &regs->gpr[0];
318 			addr += nb;
319 			for (i = 0; i < nb0; ++i, ++p)
320 				if (__get_user_inatomic(REG_BYTE(rptr,
321 								 i ^ bswiz),
322 							SWIZ_PTR(p)))
323 					return -EFAULT;
324 		}
325 
326 	} else {
327 		for (i = 0; i < nb; ++i, ++p)
328 			if (__put_user_inatomic(REG_BYTE(rptr, i ^ bswiz),
329 						SWIZ_PTR(p)))
330 				return -EFAULT;
331 		if (nb0 > 0) {
332 			rptr = &regs->gpr[0];
333 			addr += nb;
334 			for (i = 0; i < nb0; ++i, ++p)
335 				if (__put_user_inatomic(REG_BYTE(rptr,
336 								 i ^ bswiz),
337 							SWIZ_PTR(p)))
338 					return -EFAULT;
339 		}
340 	}
341 	return 1;
342 }
343 
344 /*
345  * Emulate floating-point pair loads and stores.
346  * Only POWER6 has these instructions, and it does true little-endian,
347  * so we don't need the address swizzling.
348  */
349 static int emulate_fp_pair(unsigned char __user *addr, unsigned int reg,
350 			   unsigned int flags)
351 {
352 	char *ptr0 = (char *) &current->thread.TS_FPR(reg);
353 	char *ptr1 = (char *) &current->thread.TS_FPR(reg+1);
354 	int i, ret, sw = 0;
355 
356 	if (reg & 1)
357 		return 0;	/* invalid form: FRS/FRT must be even */
358 	if (flags & SW)
359 		sw = 7;
360 	ret = 0;
361 	for (i = 0; i < 8; ++i) {
362 		if (!(flags & ST)) {
363 			ret |= __get_user(ptr0[i^sw], addr + i);
364 			ret |= __get_user(ptr1[i^sw], addr + i + 8);
365 		} else {
366 			ret |= __put_user(ptr0[i^sw], addr + i);
367 			ret |= __put_user(ptr1[i^sw], addr + i + 8);
368 		}
369 	}
370 	if (ret)
371 		return -EFAULT;
372 	return 1;	/* exception handled and fixed up */
373 }
374 
375 #ifdef CONFIG_PPC64
376 static int emulate_lq_stq(struct pt_regs *regs, unsigned char __user *addr,
377 			  unsigned int reg, unsigned int flags)
378 {
379 	char *ptr0 = (char *)&regs->gpr[reg];
380 	char *ptr1 = (char *)&regs->gpr[reg+1];
381 	int i, ret, sw = 0;
382 
383 	if (reg & 1)
384 		return 0;	/* invalid form: GPR must be even */
385 	if (flags & SW)
386 		sw = 7;
387 	ret = 0;
388 	for (i = 0; i < 8; ++i) {
389 		if (!(flags & ST)) {
390 			ret |= __get_user(ptr0[i^sw], addr + i);
391 			ret |= __get_user(ptr1[i^sw], addr + i + 8);
392 		} else {
393 			ret |= __put_user(ptr0[i^sw], addr + i);
394 			ret |= __put_user(ptr1[i^sw], addr + i + 8);
395 		}
396 	}
397 	if (ret)
398 		return -EFAULT;
399 	return 1;	/* exception handled and fixed up */
400 }
401 #endif /* CONFIG_PPC64 */
402 
403 #ifdef CONFIG_SPE
404 
405 static struct aligninfo spe_aligninfo[32] = {
406 	{ 8, LD+E8 },		/* 0 00 00: evldd[x] */
407 	{ 8, LD+E4 },		/* 0 00 01: evldw[x] */
408 	{ 8, LD },		/* 0 00 10: evldh[x] */
409 	INVALID,		/* 0 00 11 */
410 	{ 2, LD },		/* 0 01 00: evlhhesplat[x] */
411 	INVALID,		/* 0 01 01 */
412 	{ 2, LD },		/* 0 01 10: evlhhousplat[x] */
413 	{ 2, LD+SE },		/* 0 01 11: evlhhossplat[x] */
414 	{ 4, LD },		/* 0 10 00: evlwhe[x] */
415 	INVALID,		/* 0 10 01 */
416 	{ 4, LD },		/* 0 10 10: evlwhou[x] */
417 	{ 4, LD+SE },		/* 0 10 11: evlwhos[x] */
418 	{ 4, LD+E4 },		/* 0 11 00: evlwwsplat[x] */
419 	INVALID,		/* 0 11 01 */
420 	{ 4, LD },		/* 0 11 10: evlwhsplat[x] */
421 	INVALID,		/* 0 11 11 */
422 
423 	{ 8, ST+E8 },		/* 1 00 00: evstdd[x] */
424 	{ 8, ST+E4 },		/* 1 00 01: evstdw[x] */
425 	{ 8, ST },		/* 1 00 10: evstdh[x] */
426 	INVALID,		/* 1 00 11 */
427 	INVALID,		/* 1 01 00 */
428 	INVALID,		/* 1 01 01 */
429 	INVALID,		/* 1 01 10 */
430 	INVALID,		/* 1 01 11 */
431 	{ 4, ST },		/* 1 10 00: evstwhe[x] */
432 	INVALID,		/* 1 10 01 */
433 	{ 4, ST },		/* 1 10 10: evstwho[x] */
434 	INVALID,		/* 1 10 11 */
435 	{ 4, ST+E4 },		/* 1 11 00: evstwwe[x] */
436 	INVALID,		/* 1 11 01 */
437 	{ 4, ST+E4 },		/* 1 11 10: evstwwo[x] */
438 	INVALID,		/* 1 11 11 */
439 };
440 
441 #define	EVLDD		0x00
442 #define	EVLDW		0x01
443 #define	EVLDH		0x02
444 #define	EVLHHESPLAT	0x04
445 #define	EVLHHOUSPLAT	0x06
446 #define	EVLHHOSSPLAT	0x07
447 #define	EVLWHE		0x08
448 #define	EVLWHOU		0x0A
449 #define	EVLWHOS		0x0B
450 #define	EVLWWSPLAT	0x0C
451 #define	EVLWHSPLAT	0x0E
452 #define	EVSTDD		0x10
453 #define	EVSTDW		0x11
454 #define	EVSTDH		0x12
455 #define	EVSTWHE		0x18
456 #define	EVSTWHO		0x1A
457 #define	EVSTWWE		0x1C
458 #define	EVSTWWO		0x1E
459 
460 /*
461  * Emulate SPE loads and stores.
462  * Only Book-E has these instructions, and it does true little-endian,
463  * so we don't need the address swizzling.
464  */
465 static int emulate_spe(struct pt_regs *regs, unsigned int reg,
466 		       unsigned int instr)
467 {
468 	int ret;
469 	union {
470 		u64 ll;
471 		u32 w[2];
472 		u16 h[4];
473 		u8 v[8];
474 	} data, temp;
475 	unsigned char __user *p, *addr;
476 	unsigned long *evr = &current->thread.evr[reg];
477 	unsigned int nb, flags;
478 
479 	instr = (instr >> 1) & 0x1f;
480 
481 	/* DAR has the operand effective address */
482 	addr = (unsigned char __user *)regs->dar;
483 
484 	nb = spe_aligninfo[instr].len;
485 	flags = spe_aligninfo[instr].flags;
486 
487 	/* Verify the address of the operand */
488 	if (unlikely(user_mode(regs) &&
489 		     !access_ok((flags & ST ? VERIFY_WRITE : VERIFY_READ),
490 				addr, nb)))
491 		return -EFAULT;
492 
493 	/* userland only */
494 	if (unlikely(!user_mode(regs)))
495 		return 0;
496 
497 	flush_spe_to_thread(current);
498 
499 	/* If we are loading, get the data from user space, else
500 	 * get it from register values
501 	 */
502 	if (flags & ST) {
503 		data.ll = 0;
504 		switch (instr) {
505 		case EVSTDD:
506 		case EVSTDW:
507 		case EVSTDH:
508 			data.w[0] = *evr;
509 			data.w[1] = regs->gpr[reg];
510 			break;
511 		case EVSTWHE:
512 			data.h[2] = *evr >> 16;
513 			data.h[3] = regs->gpr[reg] >> 16;
514 			break;
515 		case EVSTWHO:
516 			data.h[2] = *evr & 0xffff;
517 			data.h[3] = regs->gpr[reg] & 0xffff;
518 			break;
519 		case EVSTWWE:
520 			data.w[1] = *evr;
521 			break;
522 		case EVSTWWO:
523 			data.w[1] = regs->gpr[reg];
524 			break;
525 		default:
526 			return -EINVAL;
527 		}
528 	} else {
529 		temp.ll = data.ll = 0;
530 		ret = 0;
531 		p = addr;
532 
533 		switch (nb) {
534 		case 8:
535 			ret |= __get_user_inatomic(temp.v[0], p++);
536 			ret |= __get_user_inatomic(temp.v[1], p++);
537 			ret |= __get_user_inatomic(temp.v[2], p++);
538 			ret |= __get_user_inatomic(temp.v[3], p++);
539 		case 4:
540 			ret |= __get_user_inatomic(temp.v[4], p++);
541 			ret |= __get_user_inatomic(temp.v[5], p++);
542 		case 2:
543 			ret |= __get_user_inatomic(temp.v[6], p++);
544 			ret |= __get_user_inatomic(temp.v[7], p++);
545 			if (unlikely(ret))
546 				return -EFAULT;
547 		}
548 
549 		switch (instr) {
550 		case EVLDD:
551 		case EVLDW:
552 		case EVLDH:
553 			data.ll = temp.ll;
554 			break;
555 		case EVLHHESPLAT:
556 			data.h[0] = temp.h[3];
557 			data.h[2] = temp.h[3];
558 			break;
559 		case EVLHHOUSPLAT:
560 		case EVLHHOSSPLAT:
561 			data.h[1] = temp.h[3];
562 			data.h[3] = temp.h[3];
563 			break;
564 		case EVLWHE:
565 			data.h[0] = temp.h[2];
566 			data.h[2] = temp.h[3];
567 			break;
568 		case EVLWHOU:
569 		case EVLWHOS:
570 			data.h[1] = temp.h[2];
571 			data.h[3] = temp.h[3];
572 			break;
573 		case EVLWWSPLAT:
574 			data.w[0] = temp.w[1];
575 			data.w[1] = temp.w[1];
576 			break;
577 		case EVLWHSPLAT:
578 			data.h[0] = temp.h[2];
579 			data.h[1] = temp.h[2];
580 			data.h[2] = temp.h[3];
581 			data.h[3] = temp.h[3];
582 			break;
583 		default:
584 			return -EINVAL;
585 		}
586 	}
587 
588 	if (flags & SW) {
589 		switch (flags & 0xf0) {
590 		case E8:
591 			data.ll = swab64(data.ll);
592 			break;
593 		case E4:
594 			data.w[0] = swab32(data.w[0]);
595 			data.w[1] = swab32(data.w[1]);
596 			break;
597 		/* Its half word endian */
598 		default:
599 			data.h[0] = swab16(data.h[0]);
600 			data.h[1] = swab16(data.h[1]);
601 			data.h[2] = swab16(data.h[2]);
602 			data.h[3] = swab16(data.h[3]);
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 #ifdef CONFIG_VSX
641 /*
642  * Emulate VSX instructions...
643  */
644 static int emulate_vsx(unsigned char __user *addr, unsigned int reg,
645 		       unsigned int areg, struct pt_regs *regs,
646 		       unsigned int flags, unsigned int length,
647 		       unsigned int elsize)
648 {
649 	char *ptr;
650 	unsigned long *lptr;
651 	int ret = 0;
652 	int sw = 0;
653 	int i, j;
654 
655 	/* userland only */
656 	if (unlikely(!user_mode(regs)))
657 		return 0;
658 
659 	flush_vsx_to_thread(current);
660 
661 	if (reg < 32)
662 		ptr = (char *) &current->thread.fp_state.fpr[reg][0];
663 	else
664 		ptr = (char *) &current->thread.vr_state.vr[reg - 32];
665 
666 	lptr = (unsigned long *) ptr;
667 
668 #ifdef __LITTLE_ENDIAN__
669 	if (flags & SW) {
670 		elsize = length;
671 		sw = length-1;
672 	} else {
673 		/*
674 		 * The elements are BE ordered, even in LE mode, so process
675 		 * them in reverse order.
676 		 */
677 		addr += length - elsize;
678 
679 		/* 8 byte memory accesses go in the top 8 bytes of the VR */
680 		if (length == 8)
681 			ptr += 8;
682 	}
683 #else
684 	if (flags & SW)
685 		sw = elsize-1;
686 #endif
687 
688 	for (j = 0; j < length; j += elsize) {
689 		for (i = 0; i < elsize; ++i) {
690 			if (flags & ST)
691 				ret |= __put_user(ptr[i^sw], addr + i);
692 			else
693 				ret |= __get_user(ptr[i^sw], addr + i);
694 		}
695 		ptr  += elsize;
696 #ifdef __LITTLE_ENDIAN__
697 		addr -= elsize;
698 #else
699 		addr += elsize;
700 #endif
701 	}
702 
703 #ifdef __BIG_ENDIAN__
704 #define VSX_HI 0
705 #define VSX_LO 1
706 #else
707 #define VSX_HI 1
708 #define VSX_LO 0
709 #endif
710 
711 	if (!ret) {
712 		if (flags & U)
713 			regs->gpr[areg] = regs->dar;
714 
715 		/* Splat load copies the same data to top and bottom 8 bytes */
716 		if (flags & SPLT)
717 			lptr[VSX_LO] = lptr[VSX_HI];
718 		/* For 8 byte loads, zero the low 8 bytes */
719 		else if (!(flags & ST) && (8 == length))
720 			lptr[VSX_LO] = 0;
721 	} else
722 		return -EFAULT;
723 
724 	return 1;
725 }
726 #endif
727 
728 /*
729  * Called on alignment exception. Attempts to fixup
730  *
731  * Return 1 on success
732  * Return 0 if unable to handle the interrupt
733  * Return -EFAULT if data address is bad
734  */
735 
736 int fix_alignment(struct pt_regs *regs)
737 {
738 	unsigned int instr, nb, flags, instruction = 0;
739 	unsigned int reg, areg;
740 	unsigned int dsisr;
741 	unsigned char __user *addr;
742 	unsigned long p, swiz;
743 	int ret, i;
744 	union data {
745 		u64 ll;
746 		double dd;
747 		unsigned char v[8];
748 		struct {
749 #ifdef __LITTLE_ENDIAN__
750 			int	 low32;
751 			unsigned hi32;
752 #else
753 			unsigned hi32;
754 			int	 low32;
755 #endif
756 		} x32;
757 		struct {
758 #ifdef __LITTLE_ENDIAN__
759 			short	      low16;
760 			unsigned char hi48[6];
761 #else
762 			unsigned char hi48[6];
763 			short	      low16;
764 #endif
765 		} x16;
766 	} data;
767 
768 	/*
769 	 * We require a complete register set, if not, then our assembly
770 	 * is broken
771 	 */
772 	CHECK_FULL_REGS(regs);
773 
774 	dsisr = regs->dsisr;
775 
776 	/* Some processors don't provide us with a DSISR we can use here,
777 	 * let's make one up from the instruction
778 	 */
779 	if (cpu_has_feature(CPU_FTR_NODSISRALIGN)) {
780 		unsigned long pc = regs->nip;
781 
782 		if (cpu_has_feature(CPU_FTR_PPC_LE) && (regs->msr & MSR_LE))
783 			pc ^= 4;
784 		if (unlikely(__get_user_inatomic(instr,
785 						 (unsigned int __user *)pc)))
786 			return -EFAULT;
787 		if (cpu_has_feature(CPU_FTR_REAL_LE) && (regs->msr & MSR_LE))
788 			instr = cpu_to_le32(instr);
789 		dsisr = make_dsisr(instr);
790 		instruction = instr;
791 	}
792 
793 	/* extract the operation and registers from the dsisr */
794 	reg = (dsisr >> 5) & 0x1f;	/* source/dest register */
795 	areg = dsisr & 0x1f;		/* register to update */
796 
797 #ifdef CONFIG_SPE
798 	if ((instr >> 26) == 0x4) {
799 		PPC_WARN_ALIGNMENT(spe, regs);
800 		return emulate_spe(regs, reg, instr);
801 	}
802 #endif
803 
804 	instr = (dsisr >> 10) & 0x7f;
805 	instr |= (dsisr >> 13) & 0x60;
806 
807 	/* Lookup the operation in our table */
808 	nb = aligninfo[instr].len;
809 	flags = aligninfo[instr].flags;
810 
811 	/* ldbrx/stdbrx overlap lfs/stfs in the DSISR unfortunately */
812 	if (IS_XFORM(instruction) && ((instruction >> 1) & 0x3ff) == 532) {
813 		nb = 8;
814 		flags = LD+SW;
815 	} else if (IS_XFORM(instruction) &&
816 		   ((instruction >> 1) & 0x3ff) == 660) {
817 		nb = 8;
818 		flags = ST+SW;
819 	}
820 
821 	/* Byteswap little endian loads and stores */
822 	swiz = 0;
823 	if ((regs->msr & MSR_LE) != (MSR_KERNEL & MSR_LE)) {
824 		flags ^= SW;
825 #ifdef __BIG_ENDIAN__
826 		/*
827 		 * So-called "PowerPC little endian" mode works by
828 		 * swizzling addresses rather than by actually doing
829 		 * any byte-swapping.  To emulate this, we XOR each
830 		 * byte address with 7.  We also byte-swap, because
831 		 * the processor's address swizzling depends on the
832 		 * operand size (it xors the address with 7 for bytes,
833 		 * 6 for halfwords, 4 for words, 0 for doublewords) but
834 		 * we will xor with 7 and load/store each byte separately.
835 		 */
836 		if (cpu_has_feature(CPU_FTR_PPC_LE))
837 			swiz = 7;
838 #endif
839 	}
840 
841 	/* DAR has the operand effective address */
842 	addr = (unsigned char __user *)regs->dar;
843 
844 #ifdef CONFIG_VSX
845 	if ((instruction & 0xfc00003e) == 0x7c000018) {
846 		unsigned int elsize;
847 
848 		/* Additional register addressing bit (64 VSX vs 32 FPR/GPR) */
849 		reg |= (instruction & 0x1) << 5;
850 		/* Simple inline decoder instead of a table */
851 		/* VSX has only 8 and 16 byte memory accesses */
852 		nb = 8;
853 		if (instruction & 0x200)
854 			nb = 16;
855 
856 		/* Vector stores in little-endian mode swap individual
857 		   elements, so process them separately */
858 		elsize = 4;
859 		if (instruction & 0x80)
860 			elsize = 8;
861 
862 		flags = 0;
863 		if ((regs->msr & MSR_LE) != (MSR_KERNEL & MSR_LE))
864 			flags |= SW;
865 		if (instruction & 0x100)
866 			flags |= ST;
867 		if (instruction & 0x040)
868 			flags |= U;
869 		/* splat load needs a special decoder */
870 		if ((instruction & 0x400) == 0){
871 			flags |= SPLT;
872 			nb = 8;
873 		}
874 		PPC_WARN_ALIGNMENT(vsx, regs);
875 		return emulate_vsx(addr, reg, areg, regs, flags, nb, elsize);
876 	}
877 #endif
878 	/* A size of 0 indicates an instruction we don't support, with
879 	 * the exception of DCBZ which is handled as a special case here
880 	 */
881 	if (instr == DCBZ) {
882 		PPC_WARN_ALIGNMENT(dcbz, regs);
883 		return emulate_dcbz(regs, addr);
884 	}
885 	if (unlikely(nb == 0))
886 		return 0;
887 
888 	/* Load/Store Multiple instructions are handled in their own
889 	 * function
890 	 */
891 	if (flags & M) {
892 		PPC_WARN_ALIGNMENT(multiple, regs);
893 		return emulate_multiple(regs, addr, reg, nb,
894 					flags, instr, swiz);
895 	}
896 
897 	/* Verify the address of the operand */
898 	if (unlikely(user_mode(regs) &&
899 		     !access_ok((flags & ST ? VERIFY_WRITE : VERIFY_READ),
900 				addr, nb)))
901 		return -EFAULT;
902 
903 	/* Force the fprs into the save area so we can reference them */
904 	if (flags & F) {
905 		/* userland only */
906 		if (unlikely(!user_mode(regs)))
907 			return 0;
908 		flush_fp_to_thread(current);
909 	}
910 
911 	if ((nb == 16)) {
912 		if (flags & F) {
913 			/* Special case for 16-byte FP loads and stores */
914 			PPC_WARN_ALIGNMENT(fp_pair, regs);
915 			return emulate_fp_pair(addr, reg, flags);
916 		} else {
917 #ifdef CONFIG_PPC64
918 			/* Special case for 16-byte loads and stores */
919 			PPC_WARN_ALIGNMENT(lq_stq, regs);
920 			return emulate_lq_stq(regs, addr, reg, flags);
921 #else
922 			return 0;
923 #endif
924 		}
925 	}
926 
927 	PPC_WARN_ALIGNMENT(unaligned, regs);
928 
929 	/* If we are loading, get the data from user space, else
930 	 * get it from register values
931 	 */
932 	if (!(flags & ST)) {
933 		unsigned int start = 0;
934 
935 		switch (nb) {
936 		case 4:
937 			start = offsetof(union data, x32.low32);
938 			break;
939 		case 2:
940 			start = offsetof(union data, x16.low16);
941 			break;
942 		}
943 
944 		data.ll = 0;
945 		ret = 0;
946 		p = (unsigned long)addr;
947 
948 		for (i = 0; i < nb; i++)
949 			ret |= __get_user_inatomic(data.v[start + i],
950 						   SWIZ_PTR(p++));
951 
952 		if (unlikely(ret))
953 			return -EFAULT;
954 
955 	} else if (flags & F) {
956 		data.ll = current->thread.TS_FPR(reg);
957 		if (flags & S) {
958 			/* Single-precision FP store requires conversion... */
959 #ifdef CONFIG_PPC_FPU
960 			preempt_disable();
961 			enable_kernel_fp();
962 			cvt_df(&data.dd, (float *)&data.x32.low32);
963 			preempt_enable();
964 #else
965 			return 0;
966 #endif
967 		}
968 	} else
969 		data.ll = regs->gpr[reg];
970 
971 	if (flags & SW) {
972 		switch (nb) {
973 		case 8:
974 			data.ll = swab64(data.ll);
975 			break;
976 		case 4:
977 			data.x32.low32 = swab32(data.x32.low32);
978 			break;
979 		case 2:
980 			data.x16.low16 = swab16(data.x16.low16);
981 			break;
982 		}
983 	}
984 
985 	/* Perform other misc operations like sign extension
986 	 * or floating point single precision conversion
987 	 */
988 	switch (flags & ~(U|SW)) {
989 	case LD+SE:	/* sign extending integer loads */
990 	case LD+F+SE:	/* sign extend for lfiwax */
991 		if ( nb == 2 )
992 			data.ll = data.x16.low16;
993 		else	/* nb must be 4 */
994 			data.ll = data.x32.low32;
995 		break;
996 
997 	/* Single-precision FP load requires conversion... */
998 	case LD+F+S:
999 #ifdef CONFIG_PPC_FPU
1000 		preempt_disable();
1001 		enable_kernel_fp();
1002 		cvt_fd((float *)&data.x32.low32, &data.dd);
1003 		preempt_enable();
1004 #else
1005 		return 0;
1006 #endif
1007 		break;
1008 	}
1009 
1010 	/* Store result to memory or update registers */
1011 	if (flags & ST) {
1012 		unsigned int start = 0;
1013 
1014 		switch (nb) {
1015 		case 4:
1016 			start = offsetof(union data, x32.low32);
1017 			break;
1018 		case 2:
1019 			start = offsetof(union data, x16.low16);
1020 			break;
1021 		}
1022 
1023 		ret = 0;
1024 		p = (unsigned long)addr;
1025 
1026 		for (i = 0; i < nb; i++)
1027 			ret |= __put_user_inatomic(data.v[start + i],
1028 						   SWIZ_PTR(p++));
1029 
1030 		if (unlikely(ret))
1031 			return -EFAULT;
1032 	} else if (flags & F)
1033 		current->thread.TS_FPR(reg) = data.ll;
1034 	else
1035 		regs->gpr[reg] = data.ll;
1036 
1037 	/* Update RA as needed */
1038 	if (flags & U)
1039 		regs->gpr[areg] = regs->dar;
1040 
1041 	return 1;
1042 }
1043