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
28struct aligninfo {
29	unsigned char len;
30	unsigned char flags;
31};
32
33#define IS_XFORM(inst)	(((inst) >> 26) == 31)
34#define IS_DSFORM(inst)	(((inst) >> 26) >= 56)
35
36#define INVALID	{ 0, 0 }
37
38/* Bits in the flags field */
39#define LD	0	/* load */
40#define ST	1	/* store */
41#define	SE	2	/* sign-extend value */
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
50/* DSISR bits reported for a DCBZ instruction: */
51#define DCBZ	0x5f	/* 8xx/82xx dcbz faults when cache not enabled */
52
53#define SWAP(a, b)	(t = (a), (a) = (b), (b) = t)
54
55/*
56 * The PowerPC stores certain bits of the instruction that caused the
57 * alignment exception in the DSISR register.  This array maps those
58 * bits to information about the operand length and what the
59 * instruction would do.
60 */
61static struct aligninfo aligninfo[128] = {
62	{ 4, LD },		/* 00 0 0000: lwz / lwarx */
63	INVALID,		/* 00 0 0001 */
64	{ 4, ST },		/* 00 0 0010: stw */
65	INVALID,		/* 00 0 0011 */
66	{ 2, LD },		/* 00 0 0100: lhz */
67	{ 2, LD+SE },		/* 00 0 0101: lha */
68	{ 2, ST },		/* 00 0 0110: sth */
69	{ 4, LD+M },		/* 00 0 0111: lmw */
70	{ 4, LD+F+S },		/* 00 0 1000: lfs */
71	{ 8, LD+F },		/* 00 0 1001: lfd */
72	{ 4, ST+F+S },		/* 00 0 1010: stfs */
73	{ 8, ST+F },		/* 00 0 1011: stfd */
74	INVALID,		/* 00 0 1100 */
75	{ 8, LD },		/* 00 0 1101: ld/ldu/lwa */
76	INVALID,		/* 00 0 1110 */
77	{ 8, ST },		/* 00 0 1111: std/stdu */
78	{ 4, LD+U },		/* 00 1 0000: lwzu */
79	INVALID,		/* 00 1 0001 */
80	{ 4, ST+U },		/* 00 1 0010: stwu */
81	INVALID,		/* 00 1 0011 */
82	{ 2, LD+U },		/* 00 1 0100: lhzu */
83	{ 2, LD+SE+U },		/* 00 1 0101: lhau */
84	{ 2, ST+U },		/* 00 1 0110: sthu */
85	{ 4, ST+M },		/* 00 1 0111: stmw */
86	{ 4, LD+F+S+U },	/* 00 1 1000: lfsu */
87	{ 8, LD+F+U },		/* 00 1 1001: lfdu */
88	{ 4, ST+F+S+U },	/* 00 1 1010: stfsu */
89	{ 8, ST+F+U },		/* 00 1 1011: stfdu */
90	INVALID,		/* 00 1 1100 */
91	INVALID,		/* 00 1 1101 */
92	INVALID,		/* 00 1 1110 */
93	INVALID,		/* 00 1 1111 */
94	{ 8, LD },		/* 01 0 0000: ldx */
95	INVALID,		/* 01 0 0001 */
96	{ 8, ST },		/* 01 0 0010: stdx */
97	INVALID,		/* 01 0 0011 */
98	INVALID,		/* 01 0 0100 */
99	{ 4, LD+SE },		/* 01 0 0101: lwax */
100	INVALID,		/* 01 0 0110 */
101	INVALID,		/* 01 0 0111 */
102	{ 4, LD+M+HARD+SX },	/* 01 0 1000: lswx */
103	{ 4, LD+M+HARD },	/* 01 0 1001: lswi */
104	{ 4, ST+M+HARD+SX },	/* 01 0 1010: stswx */
105	{ 4, ST+M+HARD },	/* 01 0 1011: stswi */
106	INVALID,		/* 01 0 1100 */
107	{ 8, LD+U },		/* 01 0 1101: ldu */
108	INVALID,		/* 01 0 1110 */
109	{ 8, ST+U },		/* 01 0 1111: stdu */
110	{ 8, LD+U },		/* 01 1 0000: ldux */
111	INVALID,		/* 01 1 0001 */
112	{ 8, ST+U },		/* 01 1 0010: stdux */
113	INVALID,		/* 01 1 0011 */
114	INVALID,		/* 01 1 0100 */
115	{ 4, LD+SE+U },		/* 01 1 0101: lwaux */
116	INVALID,		/* 01 1 0110 */
117	INVALID,		/* 01 1 0111 */
118	INVALID,		/* 01 1 1000 */
119	INVALID,		/* 01 1 1001 */
120	INVALID,		/* 01 1 1010 */
121	INVALID,		/* 01 1 1011 */
122	INVALID,		/* 01 1 1100 */
123	INVALID,		/* 01 1 1101 */
124	INVALID,		/* 01 1 1110 */
125	INVALID,		/* 01 1 1111 */
126	INVALID,		/* 10 0 0000 */
127	INVALID,		/* 10 0 0001 */
128	INVALID,		/* 10 0 0010: stwcx. */
129	INVALID,		/* 10 0 0011 */
130	INVALID,		/* 10 0 0100 */
131	INVALID,		/* 10 0 0101 */
132	INVALID,		/* 10 0 0110 */
133	INVALID,		/* 10 0 0111 */
134	{ 4, LD+SW },		/* 10 0 1000: lwbrx */
135	INVALID,		/* 10 0 1001 */
136	{ 4, ST+SW },		/* 10 0 1010: stwbrx */
137	INVALID,		/* 10 0 1011 */
138	{ 2, LD+SW },		/* 10 0 1100: lhbrx */
139	{ 4, LD+SE },		/* 10 0 1101  lwa */
140	{ 2, ST+SW },		/* 10 0 1110: sthbrx */
141	INVALID,		/* 10 0 1111 */
142	INVALID,		/* 10 1 0000 */
143	INVALID,		/* 10 1 0001 */
144	INVALID,		/* 10 1 0010 */
145	INVALID,		/* 10 1 0011 */
146	INVALID,		/* 10 1 0100 */
147	INVALID,		/* 10 1 0101 */
148	INVALID,		/* 10 1 0110 */
149	INVALID,		/* 10 1 0111 */
150	INVALID,		/* 10 1 1000 */
151	INVALID,		/* 10 1 1001 */
152	INVALID,		/* 10 1 1010 */
153	INVALID,		/* 10 1 1011 */
154	INVALID,		/* 10 1 1100 */
155	INVALID,		/* 10 1 1101 */
156	INVALID,		/* 10 1 1110 */
157	{ 0, ST+HARD },		/* 10 1 1111: dcbz */
158	{ 4, LD },		/* 11 0 0000: lwzx */
159	INVALID,		/* 11 0 0001 */
160	{ 4, ST },		/* 11 0 0010: stwx */
161	INVALID,		/* 11 0 0011 */
162	{ 2, LD },		/* 11 0 0100: lhzx */
163	{ 2, LD+SE },		/* 11 0 0101: lhax */
164	{ 2, ST },		/* 11 0 0110: sthx */
165	INVALID,		/* 11 0 0111 */
166	{ 4, LD+F+S },		/* 11 0 1000: lfsx */
167	{ 8, LD+F },		/* 11 0 1001: lfdx */
168	{ 4, ST+F+S },		/* 11 0 1010: stfsx */
169	{ 8, ST+F },		/* 11 0 1011: stfdx */
170	INVALID,		/* 11 0 1100 */
171	{ 8, LD+M },		/* 11 0 1101: lmd */
172	INVALID,		/* 11 0 1110 */
173	{ 8, ST+M },		/* 11 0 1111: stmd */
174	{ 4, LD+U },		/* 11 1 0000: lwzux */
175	INVALID,		/* 11 1 0001 */
176	{ 4, ST+U },		/* 11 1 0010: stwux */
177	INVALID,		/* 11 1 0011 */
178	{ 2, LD+U },		/* 11 1 0100: lhzux */
179	{ 2, LD+SE+U },		/* 11 1 0101: lhaux */
180	{ 2, ST+U },		/* 11 1 0110: sthux */
181	INVALID,		/* 11 1 0111 */
182	{ 4, LD+F+S+U },	/* 11 1 1000: lfsux */
183	{ 8, LD+F+U },		/* 11 1 1001: lfdux */
184	{ 4, ST+F+S+U },	/* 11 1 1010: stfsux */
185	{ 8, ST+F+U },		/* 11 1 1011: stfdux */
186	INVALID,		/* 11 1 1100 */
187	INVALID,		/* 11 1 1101 */
188	INVALID,		/* 11 1 1110 */
189	INVALID,		/* 11 1 1111 */
190};
191
192/*
193 * Create a DSISR value from the instruction
194 */
195static inline unsigned make_dsisr(unsigned instr)
196{
197	unsigned dsisr;
198
199
200	/* bits  6:15 --> 22:31 */
201	dsisr = (instr & 0x03ff0000) >> 16;
202
203	if (IS_XFORM(instr)) {
204		/* bits 29:30 --> 15:16 */
205		dsisr |= (instr & 0x00000006) << 14;
206		/* bit     25 -->    17 */
207		dsisr |= (instr & 0x00000040) << 8;
208		/* bits 21:24 --> 18:21 */
209		dsisr |= (instr & 0x00000780) << 3;
210	} else {
211		/* bit      5 -->    17 */
212		dsisr |= (instr & 0x04000000) >> 12;
213		/* bits  1: 4 --> 18:21 */
214		dsisr |= (instr & 0x78000000) >> 17;
215		/* bits 30:31 --> 12:13 */
216		if (IS_DSFORM(instr))
217			dsisr |= (instr & 0x00000003) << 18;
218	}
219
220	return dsisr;
221}
222
223/*
224 * The dcbz (data cache block zero) instruction
225 * gives an alignment fault if used on non-cacheable
226 * memory.  We handle the fault mainly for the
227 * case when we are running with the cache disabled
228 * for debugging.
229 */
230static int emulate_dcbz(struct pt_regs *regs, unsigned char __user *addr)
231{
232	long __user *p;
233	int i, size;
234
235#ifdef __powerpc64__
236	size = ppc64_caches.dline_size;
237#else
238	size = L1_CACHE_BYTES;
239#endif
240	p = (long __user *) (regs->dar & -size);
241	if (user_mode(regs) && !access_ok(VERIFY_WRITE, p, size))
242		return -EFAULT;
243	for (i = 0; i < size / sizeof(long); ++i)
244		if (__put_user_inatomic(0, p+i))
245			return -EFAULT;
246	return 1;
247}
248
249/*
250 * Emulate load & store multiple instructions
251 * On 64-bit machines, these instructions only affect/use the
252 * bottom 4 bytes of each register, and the loads clear the
253 * top 4 bytes of the affected register.
254 */
255#ifdef CONFIG_PPC64
256#define REG_BYTE(rp, i)		*((u8 *)((rp) + ((i) >> 2)) + ((i) & 3) + 4)
257#else
258#define REG_BYTE(rp, i)		*((u8 *)(rp) + (i))
259#endif
260
261#define SWIZ_PTR(p)		((unsigned char __user *)((p) ^ swiz))
262
263static int emulate_multiple(struct pt_regs *regs, unsigned char __user *addr,
264			    unsigned int reg, unsigned int nb,
265			    unsigned int flags, unsigned int instr,
266			    unsigned long swiz)
267{
268	unsigned long *rptr;
269	unsigned int nb0, i, bswiz;
270	unsigned long p;
271
272	/*
273	 * We do not try to emulate 8 bytes multiple as they aren't really
274	 * available in our operating environments and we don't try to
275	 * emulate multiples operations in kernel land as they should never
276	 * be used/generated there at least not on unaligned boundaries
277	 */
278	if (unlikely((nb > 4) || !user_mode(regs)))
279		return 0;
280
281	/* lmw, stmw, lswi/x, stswi/x */
282	nb0 = 0;
283	if (flags & HARD) {
284		if (flags & SX) {
285			nb = regs->xer & 127;
286			if (nb == 0)
287				return 1;
288		} else {
289			unsigned long pc = regs->nip ^ (swiz & 4);
290
291			if (__get_user_inatomic(instr,
292						(unsigned int __user *)pc))
293				return -EFAULT;
294			if (swiz == 0 && (flags & SW))
295				instr = cpu_to_le32(instr);
296			nb = (instr >> 11) & 0x1f;
297			if (nb == 0)
298				nb = 32;
299		}
300		if (nb + reg * 4 > 128) {
301			nb0 = nb + reg * 4 - 128;
302			nb = 128 - reg * 4;
303		}
304	} else {
305		/* lwm, stmw */
306		nb = (32 - reg) * 4;
307	}
308
309	if (!access_ok((flags & ST ? VERIFY_WRITE: VERIFY_READ), addr, nb+nb0))
310		return -EFAULT;	/* bad address */
311
312	rptr = &regs->gpr[reg];
313	p = (unsigned long) addr;
314	bswiz = (flags & SW)? 3: 0;
315
316	if (!(flags & ST)) {
317		/*
318		 * This zeroes the top 4 bytes of the affected registers
319		 * in 64-bit mode, and also zeroes out any remaining
320		 * bytes of the last register for lsw*.
321		 */
322		memset(rptr, 0, ((nb + 3) / 4) * sizeof(unsigned long));
323		if (nb0 > 0)
324			memset(&regs->gpr[0], 0,
325			       ((nb0 + 3) / 4) * sizeof(unsigned long));
326
327		for (i = 0; i < nb; ++i, ++p)
328			if (__get_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 (__get_user_inatomic(REG_BYTE(rptr,
336								 i ^ bswiz),
337							SWIZ_PTR(p)))
338					return -EFAULT;
339		}
340
341	} else {
342		for (i = 0; i < nb; ++i, ++p)
343			if (__put_user_inatomic(REG_BYTE(rptr, i ^ bswiz),
344						SWIZ_PTR(p)))
345				return -EFAULT;
346		if (nb0 > 0) {
347			rptr = &regs->gpr[0];
348			addr += nb;
349			for (i = 0; i < nb0; ++i, ++p)
350				if (__put_user_inatomic(REG_BYTE(rptr,
351								 i ^ bswiz),
352							SWIZ_PTR(p)))
353					return -EFAULT;
354		}
355	}
356	return 1;
357}
358
359
360/*
361 * Called on alignment exception. Attempts to fixup
362 *
363 * Return 1 on success
364 * Return 0 if unable to handle the interrupt
365 * Return -EFAULT if data address is bad
366 */
367
368int fix_alignment(struct pt_regs *regs)
369{
370	unsigned int instr, nb, flags;
371	unsigned int reg, areg;
372	unsigned int dsisr;
373	unsigned char __user *addr;
374	unsigned long p, swiz;
375	int ret, t;
376	union {
377		u64 ll;
378		double dd;
379		unsigned char v[8];
380		struct {
381			unsigned hi32;
382			int	 low32;
383		} x32;
384		struct {
385			unsigned char hi48[6];
386			short	      low16;
387		} x16;
388	} data;
389
390	/*
391	 * We require a complete register set, if not, then our assembly
392	 * is broken
393	 */
394	CHECK_FULL_REGS(regs);
395
396	dsisr = regs->dsisr;
397
398	/* Some processors don't provide us with a DSISR we can use here,
399	 * let's make one up from the instruction
400	 */
401	if (cpu_has_feature(CPU_FTR_NODSISRALIGN)) {
402		unsigned long pc = regs->nip;
403
404		if (cpu_has_feature(CPU_FTR_PPC_LE) && (regs->msr & MSR_LE))
405			pc ^= 4;
406		if (unlikely(__get_user_inatomic(instr,
407						 (unsigned int __user *)pc)))
408			return -EFAULT;
409		if (cpu_has_feature(CPU_FTR_REAL_LE) && (regs->msr & MSR_LE))
410			instr = cpu_to_le32(instr);
411		dsisr = make_dsisr(instr);
412	}
413
414	/* extract the operation and registers from the dsisr */
415	reg = (dsisr >> 5) & 0x1f;	/* source/dest register */
416	areg = dsisr & 0x1f;		/* register to update */
417	instr = (dsisr >> 10) & 0x7f;
418	instr |= (dsisr >> 13) & 0x60;
419
420	/* Lookup the operation in our table */
421	nb = aligninfo[instr].len;
422	flags = aligninfo[instr].flags;
423
424	/* Byteswap little endian loads and stores */
425	swiz = 0;
426	if (regs->msr & MSR_LE) {
427		flags ^= SW;
428		/*
429		 * So-called "PowerPC little endian" mode works by
430		 * swizzling addresses rather than by actually doing
431		 * any byte-swapping.  To emulate this, we XOR each
432		 * byte address with 7.  We also byte-swap, because
433		 * the processor's address swizzling depends on the
434		 * operand size (it xors the address with 7 for bytes,
435		 * 6 for halfwords, 4 for words, 0 for doublewords) but
436		 * we will xor with 7 and load/store each byte separately.
437		 */
438		if (cpu_has_feature(CPU_FTR_PPC_LE))
439			swiz = 7;
440	}
441
442	/* DAR has the operand effective address */
443	addr = (unsigned char __user *)regs->dar;
444
445	/* A size of 0 indicates an instruction we don't support, with
446	 * the exception of DCBZ which is handled as a special case here
447	 */
448	if (instr == DCBZ)
449		return emulate_dcbz(regs, addr);
450	if (unlikely(nb == 0))
451		return 0;
452
453	/* Load/Store Multiple instructions are handled in their own
454	 * function
455	 */
456	if (flags & M)
457		return emulate_multiple(regs, addr, reg, nb,
458					flags, instr, swiz);
459
460	/* Verify the address of the operand */
461	if (unlikely(user_mode(regs) &&
462		     !access_ok((flags & ST ? VERIFY_WRITE : VERIFY_READ),
463				addr, nb)))
464		return -EFAULT;
465
466	/* Force the fprs into the save area so we can reference them */
467	if (flags & F) {
468		/* userland only */
469		if (unlikely(!user_mode(regs)))
470			return 0;
471		flush_fp_to_thread(current);
472	}
473
474	/* If we are loading, get the data from user space, else
475	 * get it from register values
476	 */
477	if (!(flags & ST)) {
478		data.ll = 0;
479		ret = 0;
480		p = (unsigned long) addr;
481		switch (nb) {
482		case 8:
483			ret |= __get_user_inatomic(data.v[0], SWIZ_PTR(p++));
484			ret |= __get_user_inatomic(data.v[1], SWIZ_PTR(p++));
485			ret |= __get_user_inatomic(data.v[2], SWIZ_PTR(p++));
486			ret |= __get_user_inatomic(data.v[3], SWIZ_PTR(p++));
487		case 4:
488			ret |= __get_user_inatomic(data.v[4], SWIZ_PTR(p++));
489			ret |= __get_user_inatomic(data.v[5], SWIZ_PTR(p++));
490		case 2:
491			ret |= __get_user_inatomic(data.v[6], SWIZ_PTR(p++));
492			ret |= __get_user_inatomic(data.v[7], SWIZ_PTR(p++));
493			if (unlikely(ret))
494				return -EFAULT;
495		}
496	} else if (flags & F) {
497		data.dd = current->thread.fpr[reg];
498		if (flags & S) {
499			/* Single-precision FP store requires conversion... */
500#ifdef CONFIG_PPC_FPU
501			preempt_disable();
502			enable_kernel_fp();
503			cvt_df(&data.dd, (float *)&data.v[4], &current->thread);
504			preempt_enable();
505#else
506			return 0;
507#endif
508		}
509	} else
510		data.ll = regs->gpr[reg];
511
512	if (flags & SW) {
513		switch (nb) {
514		case 8:
515			SWAP(data.v[0], data.v[7]);
516			SWAP(data.v[1], data.v[6]);
517			SWAP(data.v[2], data.v[5]);
518			SWAP(data.v[3], data.v[4]);
519			break;
520		case 4:
521			SWAP(data.v[4], data.v[7]);
522			SWAP(data.v[5], data.v[6]);
523			break;
524		case 2:
525			SWAP(data.v[6], data.v[7]);
526			break;
527		}
528	}
529
530	/* Perform other misc operations like sign extension
531	 * or floating point single precision conversion
532	 */
533	switch (flags & ~(U|SW)) {
534	case LD+SE:	/* sign extend */
535		if ( nb == 2 )
536			data.ll = data.x16.low16;
537		else	/* nb must be 4 */
538			data.ll = data.x32.low32;
539		break;
540
541	/* Single-precision FP load requires conversion... */
542	case LD+F+S:
543#ifdef CONFIG_PPC_FPU
544		preempt_disable();
545		enable_kernel_fp();
546		cvt_fd((float *)&data.v[4], &data.dd, &current->thread);
547		preempt_enable();
548#else
549		return 0;
550#endif
551		break;
552	}
553
554	/* Store result to memory or update registers */
555	if (flags & ST) {
556		ret = 0;
557		p = (unsigned long) addr;
558		switch (nb) {
559		case 8:
560			ret |= __put_user_inatomic(data.v[0], SWIZ_PTR(p++));
561			ret |= __put_user_inatomic(data.v[1], SWIZ_PTR(p++));
562			ret |= __put_user_inatomic(data.v[2], SWIZ_PTR(p++));
563			ret |= __put_user_inatomic(data.v[3], SWIZ_PTR(p++));
564		case 4:
565			ret |= __put_user_inatomic(data.v[4], SWIZ_PTR(p++));
566			ret |= __put_user_inatomic(data.v[5], SWIZ_PTR(p++));
567		case 2:
568			ret |= __put_user_inatomic(data.v[6], SWIZ_PTR(p++));
569			ret |= __put_user_inatomic(data.v[7], SWIZ_PTR(p++));
570		}
571		if (unlikely(ret))
572			return -EFAULT;
573	} else if (flags & F)
574		current->thread.fpr[reg] = data.dd;
575	else
576		regs->gpr[reg] = data.ll;
577
578	/* Update RA as needed */
579	if (flags & U)
580		regs->gpr[areg] = regs->dar;
581
582	return 1;
583}
584