xref: /openbmc/linux/arch/powerpc/kernel/align.c (revision afb46f79)
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 /*
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 	{ 16, LD },		/* 00 0 1100: lq */
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 	{ 16, ST },		/* 10 0 1111: stq */
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 	{ 4, LD+F },		/* 11 1 1101: lfiwzx */
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 __BIG_ENDIAN__
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 #endif
264 
265 #ifdef __LITTLE_ENDIAN__
266 #define REG_BYTE(rp, i)		(*(((u8 *)((rp) + ((i)>>2)) + ((i)&3))))
267 #endif
268 
269 #define SWIZ_PTR(p)		((unsigned char __user *)((p) ^ swiz))
270 
271 static int emulate_multiple(struct pt_regs *regs, unsigned char __user *addr,
272 			    unsigned int reg, unsigned int nb,
273 			    unsigned int flags, unsigned int instr,
274 			    unsigned long swiz)
275 {
276 	unsigned long *rptr;
277 	unsigned int nb0, i, bswiz;
278 	unsigned long p;
279 
280 	/*
281 	 * We do not try to emulate 8 bytes multiple as they aren't really
282 	 * available in our operating environments and we don't try to
283 	 * emulate multiples operations in kernel land as they should never
284 	 * be used/generated there at least not on unaligned boundaries
285 	 */
286 	if (unlikely((nb > 4) || !user_mode(regs)))
287 		return 0;
288 
289 	/* lmw, stmw, lswi/x, stswi/x */
290 	nb0 = 0;
291 	if (flags & HARD) {
292 		if (flags & SX) {
293 			nb = regs->xer & 127;
294 			if (nb == 0)
295 				return 1;
296 		} else {
297 			unsigned long pc = regs->nip ^ (swiz & 4);
298 
299 			if (__get_user_inatomic(instr,
300 						(unsigned int __user *)pc))
301 				return -EFAULT;
302 			if (swiz == 0 && (flags & SW))
303 				instr = cpu_to_le32(instr);
304 			nb = (instr >> 11) & 0x1f;
305 			if (nb == 0)
306 				nb = 32;
307 		}
308 		if (nb + reg * 4 > 128) {
309 			nb0 = nb + reg * 4 - 128;
310 			nb = 128 - reg * 4;
311 		}
312 #ifdef __LITTLE_ENDIAN__
313 		/*
314 		 *  String instructions are endian neutral but the code
315 		 *  below is not.  Force byte swapping on so that the
316 		 *  effects of swizzling are undone in the load/store
317 		 *  loops below.
318 		 */
319 		flags ^= SW;
320 #endif
321 	} else {
322 		/* lwm, stmw */
323 		nb = (32 - reg) * 4;
324 	}
325 
326 	if (!access_ok((flags & ST ? VERIFY_WRITE: VERIFY_READ), addr, nb+nb0))
327 		return -EFAULT;	/* bad address */
328 
329 	rptr = &regs->gpr[reg];
330 	p = (unsigned long) addr;
331 	bswiz = (flags & SW)? 3: 0;
332 
333 	if (!(flags & ST)) {
334 		/*
335 		 * This zeroes the top 4 bytes of the affected registers
336 		 * in 64-bit mode, and also zeroes out any remaining
337 		 * bytes of the last register for lsw*.
338 		 */
339 		memset(rptr, 0, ((nb + 3) / 4) * sizeof(unsigned long));
340 		if (nb0 > 0)
341 			memset(&regs->gpr[0], 0,
342 			       ((nb0 + 3) / 4) * sizeof(unsigned long));
343 
344 		for (i = 0; i < nb; ++i, ++p)
345 			if (__get_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 (__get_user_inatomic(REG_BYTE(rptr,
353 								 i ^ bswiz),
354 							SWIZ_PTR(p)))
355 					return -EFAULT;
356 		}
357 
358 	} else {
359 		for (i = 0; i < nb; ++i, ++p)
360 			if (__put_user_inatomic(REG_BYTE(rptr, i ^ bswiz),
361 						SWIZ_PTR(p)))
362 				return -EFAULT;
363 		if (nb0 > 0) {
364 			rptr = &regs->gpr[0];
365 			addr += nb;
366 			for (i = 0; i < nb0; ++i, ++p)
367 				if (__put_user_inatomic(REG_BYTE(rptr,
368 								 i ^ bswiz),
369 							SWIZ_PTR(p)))
370 					return -EFAULT;
371 		}
372 	}
373 	return 1;
374 }
375 
376 /*
377  * Emulate floating-point pair loads and stores.
378  * Only POWER6 has these instructions, and it does true little-endian,
379  * so we don't need the address swizzling.
380  */
381 static int emulate_fp_pair(unsigned char __user *addr, unsigned int reg,
382 			   unsigned int flags)
383 {
384 	char *ptr0 = (char *) &current->thread.TS_FPR(reg);
385 	char *ptr1 = (char *) &current->thread.TS_FPR(reg+1);
386 	int i, ret, sw = 0;
387 
388 	if (reg & 1)
389 		return 0;	/* invalid form: FRS/FRT must be even */
390 	if (flags & SW)
391 		sw = 7;
392 	ret = 0;
393 	for (i = 0; i < 8; ++i) {
394 		if (!(flags & ST)) {
395 			ret |= __get_user(ptr0[i^sw], addr + i);
396 			ret |= __get_user(ptr1[i^sw], addr + i + 8);
397 		} else {
398 			ret |= __put_user(ptr0[i^sw], addr + i);
399 			ret |= __put_user(ptr1[i^sw], addr + i + 8);
400 		}
401 	}
402 	if (ret)
403 		return -EFAULT;
404 	return 1;	/* exception handled and fixed up */
405 }
406 
407 #ifdef CONFIG_PPC64
408 static int emulate_lq_stq(struct pt_regs *regs, unsigned char __user *addr,
409 			  unsigned int reg, unsigned int flags)
410 {
411 	char *ptr0 = (char *)&regs->gpr[reg];
412 	char *ptr1 = (char *)&regs->gpr[reg+1];
413 	int i, ret, sw = 0;
414 
415 	if (reg & 1)
416 		return 0;	/* invalid form: GPR must be even */
417 	if (flags & SW)
418 		sw = 7;
419 	ret = 0;
420 	for (i = 0; i < 8; ++i) {
421 		if (!(flags & ST)) {
422 			ret |= __get_user(ptr0[i^sw], addr + i);
423 			ret |= __get_user(ptr1[i^sw], addr + i + 8);
424 		} else {
425 			ret |= __put_user(ptr0[i^sw], addr + i);
426 			ret |= __put_user(ptr1[i^sw], addr + i + 8);
427 		}
428 	}
429 	if (ret)
430 		return -EFAULT;
431 	return 1;	/* exception handled and fixed up */
432 }
433 #endif /* CONFIG_PPC64 */
434 
435 #ifdef CONFIG_SPE
436 
437 static struct aligninfo spe_aligninfo[32] = {
438 	{ 8, LD+E8 },		/* 0 00 00: evldd[x] */
439 	{ 8, LD+E4 },		/* 0 00 01: evldw[x] */
440 	{ 8, LD },		/* 0 00 10: evldh[x] */
441 	INVALID,		/* 0 00 11 */
442 	{ 2, LD },		/* 0 01 00: evlhhesplat[x] */
443 	INVALID,		/* 0 01 01 */
444 	{ 2, LD },		/* 0 01 10: evlhhousplat[x] */
445 	{ 2, LD+SE },		/* 0 01 11: evlhhossplat[x] */
446 	{ 4, LD },		/* 0 10 00: evlwhe[x] */
447 	INVALID,		/* 0 10 01 */
448 	{ 4, LD },		/* 0 10 10: evlwhou[x] */
449 	{ 4, LD+SE },		/* 0 10 11: evlwhos[x] */
450 	{ 4, LD+E4 },		/* 0 11 00: evlwwsplat[x] */
451 	INVALID,		/* 0 11 01 */
452 	{ 4, LD },		/* 0 11 10: evlwhsplat[x] */
453 	INVALID,		/* 0 11 11 */
454 
455 	{ 8, ST+E8 },		/* 1 00 00: evstdd[x] */
456 	{ 8, ST+E4 },		/* 1 00 01: evstdw[x] */
457 	{ 8, ST },		/* 1 00 10: evstdh[x] */
458 	INVALID,		/* 1 00 11 */
459 	INVALID,		/* 1 01 00 */
460 	INVALID,		/* 1 01 01 */
461 	INVALID,		/* 1 01 10 */
462 	INVALID,		/* 1 01 11 */
463 	{ 4, ST },		/* 1 10 00: evstwhe[x] */
464 	INVALID,		/* 1 10 01 */
465 	{ 4, ST },		/* 1 10 10: evstwho[x] */
466 	INVALID,		/* 1 10 11 */
467 	{ 4, ST+E4 },		/* 1 11 00: evstwwe[x] */
468 	INVALID,		/* 1 11 01 */
469 	{ 4, ST+E4 },		/* 1 11 10: evstwwo[x] */
470 	INVALID,		/* 1 11 11 */
471 };
472 
473 #define	EVLDD		0x00
474 #define	EVLDW		0x01
475 #define	EVLDH		0x02
476 #define	EVLHHESPLAT	0x04
477 #define	EVLHHOUSPLAT	0x06
478 #define	EVLHHOSSPLAT	0x07
479 #define	EVLWHE		0x08
480 #define	EVLWHOU		0x0A
481 #define	EVLWHOS		0x0B
482 #define	EVLWWSPLAT	0x0C
483 #define	EVLWHSPLAT	0x0E
484 #define	EVSTDD		0x10
485 #define	EVSTDW		0x11
486 #define	EVSTDH		0x12
487 #define	EVSTWHE		0x18
488 #define	EVSTWHO		0x1A
489 #define	EVSTWWE		0x1C
490 #define	EVSTWWO		0x1E
491 
492 /*
493  * Emulate SPE loads and stores.
494  * Only Book-E has these instructions, and it does true little-endian,
495  * so we don't need the address swizzling.
496  */
497 static int emulate_spe(struct pt_regs *regs, unsigned int reg,
498 		       unsigned int instr)
499 {
500 	int ret;
501 	union {
502 		u64 ll;
503 		u32 w[2];
504 		u16 h[4];
505 		u8 v[8];
506 	} data, temp;
507 	unsigned char __user *p, *addr;
508 	unsigned long *evr = &current->thread.evr[reg];
509 	unsigned int nb, flags;
510 
511 	instr = (instr >> 1) & 0x1f;
512 
513 	/* DAR has the operand effective address */
514 	addr = (unsigned char __user *)regs->dar;
515 
516 	nb = spe_aligninfo[instr].len;
517 	flags = spe_aligninfo[instr].flags;
518 
519 	/* Verify the address of the operand */
520 	if (unlikely(user_mode(regs) &&
521 		     !access_ok((flags & ST ? VERIFY_WRITE : VERIFY_READ),
522 				addr, nb)))
523 		return -EFAULT;
524 
525 	/* userland only */
526 	if (unlikely(!user_mode(regs)))
527 		return 0;
528 
529 	flush_spe_to_thread(current);
530 
531 	/* If we are loading, get the data from user space, else
532 	 * get it from register values
533 	 */
534 	if (flags & ST) {
535 		data.ll = 0;
536 		switch (instr) {
537 		case EVSTDD:
538 		case EVSTDW:
539 		case EVSTDH:
540 			data.w[0] = *evr;
541 			data.w[1] = regs->gpr[reg];
542 			break;
543 		case EVSTWHE:
544 			data.h[2] = *evr >> 16;
545 			data.h[3] = regs->gpr[reg] >> 16;
546 			break;
547 		case EVSTWHO:
548 			data.h[2] = *evr & 0xffff;
549 			data.h[3] = regs->gpr[reg] & 0xffff;
550 			break;
551 		case EVSTWWE:
552 			data.w[1] = *evr;
553 			break;
554 		case EVSTWWO:
555 			data.w[1] = regs->gpr[reg];
556 			break;
557 		default:
558 			return -EINVAL;
559 		}
560 	} else {
561 		temp.ll = data.ll = 0;
562 		ret = 0;
563 		p = addr;
564 
565 		switch (nb) {
566 		case 8:
567 			ret |= __get_user_inatomic(temp.v[0], p++);
568 			ret |= __get_user_inatomic(temp.v[1], p++);
569 			ret |= __get_user_inatomic(temp.v[2], p++);
570 			ret |= __get_user_inatomic(temp.v[3], p++);
571 		case 4:
572 			ret |= __get_user_inatomic(temp.v[4], p++);
573 			ret |= __get_user_inatomic(temp.v[5], p++);
574 		case 2:
575 			ret |= __get_user_inatomic(temp.v[6], p++);
576 			ret |= __get_user_inatomic(temp.v[7], p++);
577 			if (unlikely(ret))
578 				return -EFAULT;
579 		}
580 
581 		switch (instr) {
582 		case EVLDD:
583 		case EVLDW:
584 		case EVLDH:
585 			data.ll = temp.ll;
586 			break;
587 		case EVLHHESPLAT:
588 			data.h[0] = temp.h[3];
589 			data.h[2] = temp.h[3];
590 			break;
591 		case EVLHHOUSPLAT:
592 		case EVLHHOSSPLAT:
593 			data.h[1] = temp.h[3];
594 			data.h[3] = temp.h[3];
595 			break;
596 		case EVLWHE:
597 			data.h[0] = temp.h[2];
598 			data.h[2] = temp.h[3];
599 			break;
600 		case EVLWHOU:
601 		case EVLWHOS:
602 			data.h[1] = temp.h[2];
603 			data.h[3] = temp.h[3];
604 			break;
605 		case EVLWWSPLAT:
606 			data.w[0] = temp.w[1];
607 			data.w[1] = temp.w[1];
608 			break;
609 		case EVLWHSPLAT:
610 			data.h[0] = temp.h[2];
611 			data.h[1] = temp.h[2];
612 			data.h[2] = temp.h[3];
613 			data.h[3] = temp.h[3];
614 			break;
615 		default:
616 			return -EINVAL;
617 		}
618 	}
619 
620 	if (flags & SW) {
621 		switch (flags & 0xf0) {
622 		case E8:
623 			data.ll = swab64(data.ll);
624 			break;
625 		case E4:
626 			data.w[0] = swab32(data.w[0]);
627 			data.w[1] = swab32(data.w[1]);
628 			break;
629 		/* Its half word endian */
630 		default:
631 			data.h[0] = swab16(data.h[0]);
632 			data.h[1] = swab16(data.h[1]);
633 			data.h[2] = swab16(data.h[2]);
634 			data.h[3] = swab16(data.h[3]);
635 			break;
636 		}
637 	}
638 
639 	if (flags & SE) {
640 		data.w[0] = (s16)data.h[1];
641 		data.w[1] = (s16)data.h[3];
642 	}
643 
644 	/* Store result to memory or update registers */
645 	if (flags & ST) {
646 		ret = 0;
647 		p = addr;
648 		switch (nb) {
649 		case 8:
650 			ret |= __put_user_inatomic(data.v[0], p++);
651 			ret |= __put_user_inatomic(data.v[1], p++);
652 			ret |= __put_user_inatomic(data.v[2], p++);
653 			ret |= __put_user_inatomic(data.v[3], p++);
654 		case 4:
655 			ret |= __put_user_inatomic(data.v[4], p++);
656 			ret |= __put_user_inatomic(data.v[5], p++);
657 		case 2:
658 			ret |= __put_user_inatomic(data.v[6], p++);
659 			ret |= __put_user_inatomic(data.v[7], p++);
660 		}
661 		if (unlikely(ret))
662 			return -EFAULT;
663 	} else {
664 		*evr = data.w[0];
665 		regs->gpr[reg] = data.w[1];
666 	}
667 
668 	return 1;
669 }
670 #endif /* CONFIG_SPE */
671 
672 #ifdef CONFIG_VSX
673 /*
674  * Emulate VSX instructions...
675  */
676 static int emulate_vsx(unsigned char __user *addr, unsigned int reg,
677 		       unsigned int areg, struct pt_regs *regs,
678 		       unsigned int flags, unsigned int length,
679 		       unsigned int elsize)
680 {
681 	char *ptr;
682 	unsigned long *lptr;
683 	int ret = 0;
684 	int sw = 0;
685 	int i, j;
686 
687 	/* userland only */
688 	if (unlikely(!user_mode(regs)))
689 		return 0;
690 
691 	flush_vsx_to_thread(current);
692 
693 	if (reg < 32)
694 		ptr = (char *) &current->thread.fp_state.fpr[reg][0];
695 	else
696 		ptr = (char *) &current->thread.vr_state.vr[reg - 32];
697 
698 	lptr = (unsigned long *) ptr;
699 
700 #ifdef __LITTLE_ENDIAN__
701 	if (flags & SW) {
702 		elsize = length;
703 		sw = length-1;
704 	} else {
705 		/*
706 		 * The elements are BE ordered, even in LE mode, so process
707 		 * them in reverse order.
708 		 */
709 		addr += length - elsize;
710 
711 		/* 8 byte memory accesses go in the top 8 bytes of the VR */
712 		if (length == 8)
713 			ptr += 8;
714 	}
715 #else
716 	if (flags & SW)
717 		sw = elsize-1;
718 #endif
719 
720 	for (j = 0; j < length; j += elsize) {
721 		for (i = 0; i < elsize; ++i) {
722 			if (flags & ST)
723 				ret |= __put_user(ptr[i^sw], addr + i);
724 			else
725 				ret |= __get_user(ptr[i^sw], addr + i);
726 		}
727 		ptr  += elsize;
728 #ifdef __LITTLE_ENDIAN__
729 		addr -= elsize;
730 #else
731 		addr += elsize;
732 #endif
733 	}
734 
735 #ifdef __BIG_ENDIAN__
736 #define VSX_HI 0
737 #define VSX_LO 1
738 #else
739 #define VSX_HI 1
740 #define VSX_LO 0
741 #endif
742 
743 	if (!ret) {
744 		if (flags & U)
745 			regs->gpr[areg] = regs->dar;
746 
747 		/* Splat load copies the same data to top and bottom 8 bytes */
748 		if (flags & SPLT)
749 			lptr[VSX_LO] = lptr[VSX_HI];
750 		/* For 8 byte loads, zero the low 8 bytes */
751 		else if (!(flags & ST) && (8 == length))
752 			lptr[VSX_LO] = 0;
753 	} else
754 		return -EFAULT;
755 
756 	return 1;
757 }
758 #endif
759 
760 /*
761  * Called on alignment exception. Attempts to fixup
762  *
763  * Return 1 on success
764  * Return 0 if unable to handle the interrupt
765  * Return -EFAULT if data address is bad
766  */
767 
768 int fix_alignment(struct pt_regs *regs)
769 {
770 	unsigned int instr, nb, flags, instruction = 0;
771 	unsigned int reg, areg;
772 	unsigned int dsisr;
773 	unsigned char __user *addr;
774 	unsigned long p, swiz;
775 	int ret, i;
776 	union data {
777 		u64 ll;
778 		double dd;
779 		unsigned char v[8];
780 		struct {
781 #ifdef __LITTLE_ENDIAN__
782 			int	 low32;
783 			unsigned hi32;
784 #else
785 			unsigned hi32;
786 			int	 low32;
787 #endif
788 		} x32;
789 		struct {
790 #ifdef __LITTLE_ENDIAN__
791 			short	      low16;
792 			unsigned char hi48[6];
793 #else
794 			unsigned char hi48[6];
795 			short	      low16;
796 #endif
797 		} x16;
798 	} data;
799 
800 	/*
801 	 * We require a complete register set, if not, then our assembly
802 	 * is broken
803 	 */
804 	CHECK_FULL_REGS(regs);
805 
806 	dsisr = regs->dsisr;
807 
808 	/* Some processors don't provide us with a DSISR we can use here,
809 	 * let's make one up from the instruction
810 	 */
811 	if (cpu_has_feature(CPU_FTR_NODSISRALIGN)) {
812 		unsigned long pc = regs->nip;
813 
814 		if (cpu_has_feature(CPU_FTR_PPC_LE) && (regs->msr & MSR_LE))
815 			pc ^= 4;
816 		if (unlikely(__get_user_inatomic(instr,
817 						 (unsigned int __user *)pc)))
818 			return -EFAULT;
819 		if (cpu_has_feature(CPU_FTR_REAL_LE) && (regs->msr & MSR_LE))
820 			instr = cpu_to_le32(instr);
821 		dsisr = make_dsisr(instr);
822 		instruction = instr;
823 	}
824 
825 	/* extract the operation and registers from the dsisr */
826 	reg = (dsisr >> 5) & 0x1f;	/* source/dest register */
827 	areg = dsisr & 0x1f;		/* register to update */
828 
829 #ifdef CONFIG_SPE
830 	if ((instr >> 26) == 0x4) {
831 		PPC_WARN_ALIGNMENT(spe, regs);
832 		return emulate_spe(regs, reg, instr);
833 	}
834 #endif
835 
836 	instr = (dsisr >> 10) & 0x7f;
837 	instr |= (dsisr >> 13) & 0x60;
838 
839 	/* Lookup the operation in our table */
840 	nb = aligninfo[instr].len;
841 	flags = aligninfo[instr].flags;
842 
843 	/* ldbrx/stdbrx overlap lfs/stfs in the DSISR unfortunately */
844 	if (IS_XFORM(instruction) && ((instruction >> 1) & 0x3ff) == 532) {
845 		nb = 8;
846 		flags = LD+SW;
847 	} else if (IS_XFORM(instruction) &&
848 		   ((instruction >> 1) & 0x3ff) == 660) {
849 		nb = 8;
850 		flags = ST+SW;
851 	}
852 
853 	/* Byteswap little endian loads and stores */
854 	swiz = 0;
855 	if ((regs->msr & MSR_LE) != (MSR_KERNEL & MSR_LE)) {
856 		flags ^= SW;
857 #ifdef __BIG_ENDIAN__
858 		/*
859 		 * So-called "PowerPC little endian" mode works by
860 		 * swizzling addresses rather than by actually doing
861 		 * any byte-swapping.  To emulate this, we XOR each
862 		 * byte address with 7.  We also byte-swap, because
863 		 * the processor's address swizzling depends on the
864 		 * operand size (it xors the address with 7 for bytes,
865 		 * 6 for halfwords, 4 for words, 0 for doublewords) but
866 		 * we will xor with 7 and load/store each byte separately.
867 		 */
868 		if (cpu_has_feature(CPU_FTR_PPC_LE))
869 			swiz = 7;
870 #endif
871 	}
872 
873 	/* DAR has the operand effective address */
874 	addr = (unsigned char __user *)regs->dar;
875 
876 #ifdef CONFIG_VSX
877 	if ((instruction & 0xfc00003e) == 0x7c000018) {
878 		unsigned int elsize;
879 
880 		/* Additional register addressing bit (64 VSX vs 32 FPR/GPR) */
881 		reg |= (instruction & 0x1) << 5;
882 		/* Simple inline decoder instead of a table */
883 		/* VSX has only 8 and 16 byte memory accesses */
884 		nb = 8;
885 		if (instruction & 0x200)
886 			nb = 16;
887 
888 		/* Vector stores in little-endian mode swap individual
889 		   elements, so process them separately */
890 		elsize = 4;
891 		if (instruction & 0x80)
892 			elsize = 8;
893 
894 		flags = 0;
895 		if ((regs->msr & MSR_LE) != (MSR_KERNEL & MSR_LE))
896 			flags |= SW;
897 		if (instruction & 0x100)
898 			flags |= ST;
899 		if (instruction & 0x040)
900 			flags |= U;
901 		/* splat load needs a special decoder */
902 		if ((instruction & 0x400) == 0){
903 			flags |= SPLT;
904 			nb = 8;
905 		}
906 		PPC_WARN_ALIGNMENT(vsx, regs);
907 		return emulate_vsx(addr, reg, areg, regs, flags, nb, elsize);
908 	}
909 #endif
910 	/* A size of 0 indicates an instruction we don't support, with
911 	 * the exception of DCBZ which is handled as a special case here
912 	 */
913 	if (instr == DCBZ) {
914 		PPC_WARN_ALIGNMENT(dcbz, regs);
915 		return emulate_dcbz(regs, addr);
916 	}
917 	if (unlikely(nb == 0))
918 		return 0;
919 
920 	/* Load/Store Multiple instructions are handled in their own
921 	 * function
922 	 */
923 	if (flags & M) {
924 		PPC_WARN_ALIGNMENT(multiple, regs);
925 		return emulate_multiple(regs, addr, reg, nb,
926 					flags, instr, swiz);
927 	}
928 
929 	/* Verify the address of the operand */
930 	if (unlikely(user_mode(regs) &&
931 		     !access_ok((flags & ST ? VERIFY_WRITE : VERIFY_READ),
932 				addr, nb)))
933 		return -EFAULT;
934 
935 	/* Force the fprs into the save area so we can reference them */
936 	if (flags & F) {
937 		/* userland only */
938 		if (unlikely(!user_mode(regs)))
939 			return 0;
940 		flush_fp_to_thread(current);
941 	}
942 
943 	if ((nb == 16)) {
944 		if (flags & F) {
945 			/* Special case for 16-byte FP loads and stores */
946 			PPC_WARN_ALIGNMENT(fp_pair, regs);
947 			return emulate_fp_pair(addr, reg, flags);
948 		} else {
949 #ifdef CONFIG_PPC64
950 			/* Special case for 16-byte loads and stores */
951 			PPC_WARN_ALIGNMENT(lq_stq, regs);
952 			return emulate_lq_stq(regs, addr, reg, flags);
953 #else
954 			return 0;
955 #endif
956 		}
957 	}
958 
959 	PPC_WARN_ALIGNMENT(unaligned, regs);
960 
961 	/* If we are loading, get the data from user space, else
962 	 * get it from register values
963 	 */
964 	if (!(flags & ST)) {
965 		unsigned int start = 0;
966 
967 		switch (nb) {
968 		case 4:
969 			start = offsetof(union data, x32.low32);
970 			break;
971 		case 2:
972 			start = offsetof(union data, x16.low16);
973 			break;
974 		}
975 
976 		data.ll = 0;
977 		ret = 0;
978 		p = (unsigned long)addr;
979 
980 		for (i = 0; i < nb; i++)
981 			ret |= __get_user_inatomic(data.v[start + i],
982 						   SWIZ_PTR(p++));
983 
984 		if (unlikely(ret))
985 			return -EFAULT;
986 
987 	} else if (flags & F) {
988 		data.ll = current->thread.TS_FPR(reg);
989 		if (flags & S) {
990 			/* Single-precision FP store requires conversion... */
991 #ifdef CONFIG_PPC_FPU
992 			preempt_disable();
993 			enable_kernel_fp();
994 			cvt_df(&data.dd, (float *)&data.x32.low32);
995 			preempt_enable();
996 #else
997 			return 0;
998 #endif
999 		}
1000 	} else
1001 		data.ll = regs->gpr[reg];
1002 
1003 	if (flags & SW) {
1004 		switch (nb) {
1005 		case 8:
1006 			data.ll = swab64(data.ll);
1007 			break;
1008 		case 4:
1009 			data.x32.low32 = swab32(data.x32.low32);
1010 			break;
1011 		case 2:
1012 			data.x16.low16 = swab16(data.x16.low16);
1013 			break;
1014 		}
1015 	}
1016 
1017 	/* Perform other misc operations like sign extension
1018 	 * or floating point single precision conversion
1019 	 */
1020 	switch (flags & ~(U|SW)) {
1021 	case LD+SE:	/* sign extending integer loads */
1022 	case LD+F+SE:	/* sign extend for lfiwax */
1023 		if ( nb == 2 )
1024 			data.ll = data.x16.low16;
1025 		else	/* nb must be 4 */
1026 			data.ll = data.x32.low32;
1027 		break;
1028 
1029 	/* Single-precision FP load requires conversion... */
1030 	case LD+F+S:
1031 #ifdef CONFIG_PPC_FPU
1032 		preempt_disable();
1033 		enable_kernel_fp();
1034 		cvt_fd((float *)&data.x32.low32, &data.dd);
1035 		preempt_enable();
1036 #else
1037 		return 0;
1038 #endif
1039 		break;
1040 	}
1041 
1042 	/* Store result to memory or update registers */
1043 	if (flags & ST) {
1044 		unsigned int start = 0;
1045 
1046 		switch (nb) {
1047 		case 4:
1048 			start = offsetof(union data, x32.low32);
1049 			break;
1050 		case 2:
1051 			start = offsetof(union data, x16.low16);
1052 			break;
1053 		}
1054 
1055 		ret = 0;
1056 		p = (unsigned long)addr;
1057 
1058 		for (i = 0; i < nb; i++)
1059 			ret |= __put_user_inatomic(data.v[start + i],
1060 						   SWIZ_PTR(p++));
1061 
1062 		if (unlikely(ret))
1063 			return -EFAULT;
1064 	} else if (flags & F)
1065 		current->thread.TS_FPR(reg) = data.ll;
1066 	else
1067 		regs->gpr[reg] = data.ll;
1068 
1069 	/* Update RA as needed */
1070 	if (flags & U)
1071 		regs->gpr[areg] = regs->dar;
1072 
1073 	return 1;
1074 }
1075