xref: /openbmc/linux/arch/powerpc/kernel/align.c (revision 87c2ce3b)
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 #define LD	1	/* load */
39 #define ST	2	/* store */
40 #define	SE	4	/* sign-extend value */
41 #define F	8	/* to/from fp regs */
42 #define U	0x10	/* update index register */
43 #define M	0x20	/* multiple load/store */
44 #define SW	0x40	/* byte swap int or ... */
45 #define S	0x40	/* ... single-precision fp */
46 #define SX	0x40	/* byte count in XER */
47 #define HARD	0x80	/* string, stwcx. */
48 
49 #define DCBZ	0x5f	/* 8xx/82xx dcbz faults when cache not enabled */
50 
51 #define SWAP(a, b)	(t = (a), (a) = (b), (b) = t)
52 
53 /*
54  * The PowerPC stores certain bits of the instruction that caused the
55  * alignment exception in the DSISR register.  This array maps those
56  * bits to information about the operand length and what the
57  * instruction would do.
58  */
59 static struct aligninfo aligninfo[128] = {
60 	{ 4, LD },		/* 00 0 0000: lwz / lwarx */
61 	INVALID,		/* 00 0 0001 */
62 	{ 4, ST },		/* 00 0 0010: stw */
63 	INVALID,		/* 00 0 0011 */
64 	{ 2, LD },		/* 00 0 0100: lhz */
65 	{ 2, LD+SE },		/* 00 0 0101: lha */
66 	{ 2, ST },		/* 00 0 0110: sth */
67 	{ 4, LD+M },		/* 00 0 0111: lmw */
68 	{ 4, LD+F+S },		/* 00 0 1000: lfs */
69 	{ 8, LD+F },		/* 00 0 1001: lfd */
70 	{ 4, ST+F+S },		/* 00 0 1010: stfs */
71 	{ 8, ST+F },		/* 00 0 1011: stfd */
72 	INVALID,		/* 00 0 1100 */
73 	{ 8, LD },		/* 00 0 1101: ld/ldu/lwa */
74 	INVALID,		/* 00 0 1110 */
75 	{ 8, ST },		/* 00 0 1111: std/stdu */
76 	{ 4, LD+U },		/* 00 1 0000: lwzu */
77 	INVALID,		/* 00 1 0001 */
78 	{ 4, ST+U },		/* 00 1 0010: stwu */
79 	INVALID,		/* 00 1 0011 */
80 	{ 2, LD+U },		/* 00 1 0100: lhzu */
81 	{ 2, LD+SE+U },		/* 00 1 0101: lhau */
82 	{ 2, ST+U },		/* 00 1 0110: sthu */
83 	{ 4, ST+M },		/* 00 1 0111: stmw */
84 	{ 4, LD+F+S+U },	/* 00 1 1000: lfsu */
85 	{ 8, LD+F+U },		/* 00 1 1001: lfdu */
86 	{ 4, ST+F+S+U },	/* 00 1 1010: stfsu */
87 	{ 8, ST+F+U },		/* 00 1 1011: stfdu */
88 	INVALID,		/* 00 1 1100 */
89 	INVALID,		/* 00 1 1101 */
90 	INVALID,		/* 00 1 1110 */
91 	INVALID,		/* 00 1 1111 */
92 	{ 8, LD },		/* 01 0 0000: ldx */
93 	INVALID,		/* 01 0 0001 */
94 	{ 8, ST },		/* 01 0 0010: stdx */
95 	INVALID,		/* 01 0 0011 */
96 	INVALID,		/* 01 0 0100 */
97 	{ 4, LD+SE },		/* 01 0 0101: lwax */
98 	INVALID,		/* 01 0 0110 */
99 	INVALID,		/* 01 0 0111 */
100 	{ 4, LD+M+HARD+SX },	/* 01 0 1000: lswx */
101 	{ 4, LD+M+HARD },	/* 01 0 1001: lswi */
102 	{ 4, ST+M+HARD+SX },	/* 01 0 1010: stswx */
103 	{ 4, ST+M+HARD },	/* 01 0 1011: stswi */
104 	INVALID,		/* 01 0 1100 */
105 	{ 8, LD+U },		/* 01 0 1101: ldu */
106 	INVALID,		/* 01 0 1110 */
107 	{ 8, ST+U },		/* 01 0 1111: stdu */
108 	{ 8, LD+U },		/* 01 1 0000: ldux */
109 	INVALID,		/* 01 1 0001 */
110 	{ 8, ST+U },		/* 01 1 0010: stdux */
111 	INVALID,		/* 01 1 0011 */
112 	INVALID,		/* 01 1 0100 */
113 	{ 4, LD+SE+U },		/* 01 1 0101: lwaux */
114 	INVALID,		/* 01 1 0110 */
115 	INVALID,		/* 01 1 0111 */
116 	INVALID,		/* 01 1 1000 */
117 	INVALID,		/* 01 1 1001 */
118 	INVALID,		/* 01 1 1010 */
119 	INVALID,		/* 01 1 1011 */
120 	INVALID,		/* 01 1 1100 */
121 	INVALID,		/* 01 1 1101 */
122 	INVALID,		/* 01 1 1110 */
123 	INVALID,		/* 01 1 1111 */
124 	INVALID,		/* 10 0 0000 */
125 	INVALID,		/* 10 0 0001 */
126 	INVALID,		/* 10 0 0010: stwcx. */
127 	INVALID,		/* 10 0 0011 */
128 	INVALID,		/* 10 0 0100 */
129 	INVALID,		/* 10 0 0101 */
130 	INVALID,		/* 10 0 0110 */
131 	INVALID,		/* 10 0 0111 */
132 	{ 4, LD+SW },		/* 10 0 1000: lwbrx */
133 	INVALID,		/* 10 0 1001 */
134 	{ 4, ST+SW },		/* 10 0 1010: stwbrx */
135 	INVALID,		/* 10 0 1011 */
136 	{ 2, LD+SW },		/* 10 0 1100: lhbrx */
137 	{ 4, LD+SE },		/* 10 0 1101  lwa */
138 	{ 2, ST+SW },		/* 10 0 1110: sthbrx */
139 	INVALID,		/* 10 0 1111 */
140 	INVALID,		/* 10 1 0000 */
141 	INVALID,		/* 10 1 0001 */
142 	INVALID,		/* 10 1 0010 */
143 	INVALID,		/* 10 1 0011 */
144 	INVALID,		/* 10 1 0100 */
145 	INVALID,		/* 10 1 0101 */
146 	INVALID,		/* 10 1 0110 */
147 	INVALID,		/* 10 1 0111 */
148 	INVALID,		/* 10 1 1000 */
149 	INVALID,		/* 10 1 1001 */
150 	INVALID,		/* 10 1 1010 */
151 	INVALID,		/* 10 1 1011 */
152 	INVALID,		/* 10 1 1100 */
153 	INVALID,		/* 10 1 1101 */
154 	INVALID,		/* 10 1 1110 */
155 	{ 0, ST+HARD },		/* 10 1 1111: dcbz */
156 	{ 4, LD },		/* 11 0 0000: lwzx */
157 	INVALID,		/* 11 0 0001 */
158 	{ 4, ST },		/* 11 0 0010: stwx */
159 	INVALID,		/* 11 0 0011 */
160 	{ 2, LD },		/* 11 0 0100: lhzx */
161 	{ 2, LD+SE },		/* 11 0 0101: lhax */
162 	{ 2, ST },		/* 11 0 0110: sthx */
163 	INVALID,		/* 11 0 0111 */
164 	{ 4, LD+F+S },		/* 11 0 1000: lfsx */
165 	{ 8, LD+F },		/* 11 0 1001: lfdx */
166 	{ 4, ST+F+S },		/* 11 0 1010: stfsx */
167 	{ 8, ST+F },		/* 11 0 1011: stfdx */
168 	INVALID,		/* 11 0 1100 */
169 	{ 8, LD+M },		/* 11 0 1101: lmd */
170 	INVALID,		/* 11 0 1110 */
171 	{ 8, ST+M },		/* 11 0 1111: stmd */
172 	{ 4, LD+U },		/* 11 1 0000: lwzux */
173 	INVALID,		/* 11 1 0001 */
174 	{ 4, ST+U },		/* 11 1 0010: stwux */
175 	INVALID,		/* 11 1 0011 */
176 	{ 2, LD+U },		/* 11 1 0100: lhzux */
177 	{ 2, LD+SE+U },		/* 11 1 0101: lhaux */
178 	{ 2, ST+U },		/* 11 1 0110: sthux */
179 	INVALID,		/* 11 1 0111 */
180 	{ 4, LD+F+S+U },	/* 11 1 1000: lfsux */
181 	{ 8, LD+F+U },		/* 11 1 1001: lfdux */
182 	{ 4, ST+F+S+U },	/* 11 1 1010: stfsux */
183 	{ 8, ST+F+U },		/* 11 1 1011: stfdux */
184 	INVALID,		/* 11 1 1100 */
185 	INVALID,		/* 11 1 1101 */
186 	INVALID,		/* 11 1 1110 */
187 	INVALID,		/* 11 1 1111 */
188 };
189 
190 /*
191  * Create a DSISR value from the instruction
192  */
193 static inline unsigned make_dsisr(unsigned instr)
194 {
195 	unsigned dsisr;
196 
197 
198 	/* bits  6:15 --> 22:31 */
199 	dsisr = (instr & 0x03ff0000) >> 16;
200 
201 	if (IS_XFORM(instr)) {
202 		/* bits 29:30 --> 15:16 */
203 		dsisr |= (instr & 0x00000006) << 14;
204 		/* bit     25 -->    17 */
205 		dsisr |= (instr & 0x00000040) << 8;
206 		/* bits 21:24 --> 18:21 */
207 		dsisr |= (instr & 0x00000780) << 3;
208 	} else {
209 		/* bit      5 -->    17 */
210 		dsisr |= (instr & 0x04000000) >> 12;
211 		/* bits  1: 4 --> 18:21 */
212 		dsisr |= (instr & 0x78000000) >> 17;
213 		/* bits 30:31 --> 12:13 */
214 		if (IS_DSFORM(instr))
215 			dsisr |= (instr & 0x00000003) << 18;
216 	}
217 
218 	return dsisr;
219 }
220 
221 /*
222  * The dcbz (data cache block zero) instruction
223  * gives an alignment fault if used on non-cacheable
224  * memory.  We handle the fault mainly for the
225  * case when we are running with the cache disabled
226  * for debugging.
227  */
228 static int emulate_dcbz(struct pt_regs *regs, unsigned char __user *addr)
229 {
230 	long __user *p;
231 	int i, size;
232 
233 #ifdef __powerpc64__
234 	size = ppc64_caches.dline_size;
235 #else
236 	size = L1_CACHE_BYTES;
237 #endif
238 	p = (long __user *) (regs->dar & -size);
239 	if (user_mode(regs) && !access_ok(VERIFY_WRITE, p, size))
240 		return -EFAULT;
241 	for (i = 0; i < size / sizeof(long); ++i)
242 		if (__put_user(0, p+i))
243 			return -EFAULT;
244 	return 1;
245 }
246 
247 /*
248  * Emulate load & store multiple instructions
249  * On 64-bit machines, these instructions only affect/use the
250  * bottom 4 bytes of each register, and the loads clear the
251  * top 4 bytes of the affected register.
252  */
253 #ifdef CONFIG_PPC64
254 #define REG_BYTE(rp, i)		*((u8 *)((rp) + ((i) >> 2)) + ((i) & 3) + 4)
255 #else
256 #define REG_BYTE(rp, i)		*((u8 *)(rp) + (i))
257 #endif
258 
259 static int emulate_multiple(struct pt_regs *regs, unsigned char __user *addr,
260 			    unsigned int reg, unsigned int nb,
261 			    unsigned int flags, unsigned int instr)
262 {
263 	unsigned long *rptr;
264 	unsigned int nb0, i;
265 
266 	/*
267 	 * We do not try to emulate 8 bytes multiple as they aren't really
268 	 * available in our operating environments and we don't try to
269 	 * emulate multiples operations in kernel land as they should never
270 	 * be used/generated there at least not on unaligned boundaries
271 	 */
272 	if (unlikely((nb > 4) || !user_mode(regs)))
273 		return 0;
274 
275 	/* lmw, stmw, lswi/x, stswi/x */
276 	nb0 = 0;
277 	if (flags & HARD) {
278 		if (flags & SX) {
279 			nb = regs->xer & 127;
280 			if (nb == 0)
281 				return 1;
282 		} else {
283 			if (__get_user(instr,
284 				       (unsigned int __user *)regs->nip))
285 				return -EFAULT;
286 			nb = (instr >> 11) & 0x1f;
287 			if (nb == 0)
288 				nb = 32;
289 		}
290 		if (nb + reg * 4 > 128) {
291 			nb0 = nb + reg * 4 - 128;
292 			nb = 128 - reg * 4;
293 		}
294 	} else {
295 		/* lwm, stmw */
296 		nb = (32 - reg) * 4;
297 	}
298 
299 	if (!access_ok((flags & ST ? VERIFY_WRITE: VERIFY_READ), addr, nb+nb0))
300 		return -EFAULT;	/* bad address */
301 
302 	rptr = &regs->gpr[reg];
303 	if (flags & LD) {
304 		/*
305 		 * This zeroes the top 4 bytes of the affected registers
306 		 * in 64-bit mode, and also zeroes out any remaining
307 		 * bytes of the last register for lsw*.
308 		 */
309 		memset(rptr, 0, ((nb + 3) / 4) * sizeof(unsigned long));
310 		if (nb0 > 0)
311 			memset(&regs->gpr[0], 0,
312 			       ((nb0 + 3) / 4) * sizeof(unsigned long));
313 
314 		for (i = 0; i < nb; ++i)
315 			if (__get_user(REG_BYTE(rptr, i), addr + i))
316 				return -EFAULT;
317 		if (nb0 > 0) {
318 			rptr = &regs->gpr[0];
319 			addr += nb;
320 			for (i = 0; i < nb0; ++i)
321 				if (__get_user(REG_BYTE(rptr, i), addr + i))
322 					return -EFAULT;
323 		}
324 
325 	} else {
326 		for (i = 0; i < nb; ++i)
327 			if (__put_user(REG_BYTE(rptr, i), addr + i))
328 				return -EFAULT;
329 		if (nb0 > 0) {
330 			rptr = &regs->gpr[0];
331 			addr += nb;
332 			for (i = 0; i < nb0; ++i)
333 				if (__put_user(REG_BYTE(rptr, i), addr + i))
334 					return -EFAULT;
335 		}
336 	}
337 	return 1;
338 }
339 
340 
341 /*
342  * Called on alignment exception. Attempts to fixup
343  *
344  * Return 1 on success
345  * Return 0 if unable to handle the interrupt
346  * Return -EFAULT if data address is bad
347  */
348 
349 int fix_alignment(struct pt_regs *regs)
350 {
351 	unsigned int instr, nb, flags;
352 	unsigned int reg, areg;
353 	unsigned int dsisr;
354 	unsigned char __user *addr;
355 	unsigned char __user *p;
356 	int ret, t;
357 	union {
358 		u64 ll;
359 		double dd;
360 		unsigned char v[8];
361 		struct {
362 			unsigned hi32;
363 			int	 low32;
364 		} x32;
365 		struct {
366 			unsigned char hi48[6];
367 			short	      low16;
368 		} x16;
369 	} data;
370 
371 	/*
372 	 * We require a complete register set, if not, then our assembly
373 	 * is broken
374 	 */
375 	CHECK_FULL_REGS(regs);
376 
377 	dsisr = regs->dsisr;
378 
379 	/* Some processors don't provide us with a DSISR we can use here,
380 	 * let's make one up from the instruction
381 	 */
382 	if (cpu_has_feature(CPU_FTR_NODSISRALIGN)) {
383 		unsigned int real_instr;
384 		if (unlikely(__get_user(real_instr,
385 					(unsigned int __user *)regs->nip)))
386 			return -EFAULT;
387 		dsisr = make_dsisr(real_instr);
388 	}
389 
390 	/* extract the operation and registers from the dsisr */
391 	reg = (dsisr >> 5) & 0x1f;	/* source/dest register */
392 	areg = dsisr & 0x1f;		/* register to update */
393 	instr = (dsisr >> 10) & 0x7f;
394 	instr |= (dsisr >> 13) & 0x60;
395 
396 	/* Lookup the operation in our table */
397 	nb = aligninfo[instr].len;
398 	flags = aligninfo[instr].flags;
399 
400 	/* DAR has the operand effective address */
401 	addr = (unsigned char __user *)regs->dar;
402 
403 	/* A size of 0 indicates an instruction we don't support, with
404 	 * the exception of DCBZ which is handled as a special case here
405 	 */
406 	if (instr == DCBZ)
407 		return emulate_dcbz(regs, addr);
408 	if (unlikely(nb == 0))
409 		return 0;
410 
411 	/* Load/Store Multiple instructions are handled in their own
412 	 * function
413 	 */
414 	if (flags & M)
415 		return emulate_multiple(regs, addr, reg, nb, flags, instr);
416 
417 	/* Verify the address of the operand */
418 	if (unlikely(user_mode(regs) &&
419 		     !access_ok((flags & ST ? VERIFY_WRITE : VERIFY_READ),
420 				addr, nb)))
421 		return -EFAULT;
422 
423 	/* Force the fprs into the save area so we can reference them */
424 	if (flags & F) {
425 		/* userland only */
426 		if (unlikely(!user_mode(regs)))
427 			return 0;
428 		flush_fp_to_thread(current);
429 	}
430 
431 	/* If we are loading, get the data from user space, else
432 	 * get it from register values
433 	 */
434 	if (flags & LD) {
435 		data.ll = 0;
436 		ret = 0;
437 		p = addr;
438 		switch (nb) {
439 		case 8:
440 			ret |= __get_user(data.v[0], p++);
441 			ret |= __get_user(data.v[1], p++);
442 			ret |= __get_user(data.v[2], p++);
443 			ret |= __get_user(data.v[3], p++);
444 		case 4:
445 			ret |= __get_user(data.v[4], p++);
446 			ret |= __get_user(data.v[5], p++);
447 		case 2:
448 			ret |= __get_user(data.v[6], p++);
449 			ret |= __get_user(data.v[7], p++);
450 			if (unlikely(ret))
451 				return -EFAULT;
452 		}
453 	} else if (flags & F)
454 		data.dd = current->thread.fpr[reg];
455 	else
456 		data.ll = regs->gpr[reg];
457 
458 	/* Perform other misc operations like sign extension, byteswap,
459 	 * or floating point single precision conversion
460 	 */
461 	switch (flags & ~U) {
462 	case LD+SE:	/* sign extend */
463 		if ( nb == 2 )
464 			data.ll = data.x16.low16;
465 		else	/* nb must be 4 */
466 			data.ll = data.x32.low32;
467 		break;
468 	case LD+S:	/* byte-swap */
469 	case ST+S:
470 		if (nb == 2) {
471 			SWAP(data.v[6], data.v[7]);
472 		} else {
473 			SWAP(data.v[4], data.v[7]);
474 			SWAP(data.v[5], data.v[6]);
475 		}
476 		break;
477 
478 	/* Single-precision FP load and store require conversions... */
479 	case LD+F+S:
480 #ifdef CONFIG_PPC_FPU
481 		preempt_disable();
482 		enable_kernel_fp();
483 		cvt_fd((float *)&data.v[4], &data.dd, &current->thread);
484 		preempt_enable();
485 #else
486 		return 0;
487 #endif
488 		break;
489 	case ST+F+S:
490 #ifdef CONFIG_PPC_FPU
491 		preempt_disable();
492 		enable_kernel_fp();
493 		cvt_df(&data.dd, (float *)&data.v[4], &current->thread);
494 		preempt_enable();
495 #else
496 		return 0;
497 #endif
498 		break;
499 	}
500 
501 	/* Store result to memory or update registers */
502 	if (flags & ST) {
503 		ret = 0;
504 		p = addr;
505 		switch (nb) {
506 		case 8:
507 			ret |= __put_user(data.v[0], p++);
508 			ret |= __put_user(data.v[1], p++);
509 			ret |= __put_user(data.v[2], p++);
510 			ret |= __put_user(data.v[3], p++);
511 		case 4:
512 			ret |= __put_user(data.v[4], p++);
513 			ret |= __put_user(data.v[5], p++);
514 		case 2:
515 			ret |= __put_user(data.v[6], p++);
516 			ret |= __put_user(data.v[7], p++);
517 		}
518 		if (unlikely(ret))
519 			return -EFAULT;
520 	} else if (flags & F)
521 		current->thread.fpr[reg] = data.dd;
522 	else
523 		regs->gpr[reg] = data.ll;
524 
525 	/* Update RA as needed */
526 	if (flags & U)
527 		regs->gpr[areg] = regs->dar;
528 
529 	return 1;
530 }
531