1 /* align.c - handle alignment exceptions for the Power PC.
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
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.
20 #include <linux/kernel.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 #include <asm/cpu_has_feature.h>
37 #define INVALID { 0, 0 }
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 */
54 /* DSISR bits reported for a DCBZ instruction: */
55 #define DCBZ 0x5f /* 8xx/82xx dcbz faults when cache not enabled */
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.
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 */
195 * The dcbz (data cache block zero) instruction
196 * gives an alignment fault if used on non-cacheable
197 * memory. We handle the fault mainly for the
198 * case when we are running with the cache disabled
201 static int emulate_dcbz(struct pt_regs *regs, unsigned char __user *addr)
207 size = ppc64_caches.dline_size;
209 size = L1_CACHE_BYTES;
211 p = (long __user *) (regs->dar & -size);
212 if (user_mode(regs) && !access_ok(VERIFY_WRITE, p, size))
214 for (i = 0; i < size / sizeof(long); ++i)
215 if (__put_user_inatomic(0, p+i))
221 * Emulate load & store multiple instructions
222 * On 64-bit machines, these instructions only affect/use the
223 * bottom 4 bytes of each register, and the loads clear the
224 * top 4 bytes of the affected register.
226 #ifdef __BIG_ENDIAN__
228 #define REG_BYTE(rp, i) *((u8 *)((rp) + ((i) >> 2)) + ((i) & 3) + 4)
230 #define REG_BYTE(rp, i) *((u8 *)(rp) + (i))
233 #define REG_BYTE(rp, i) (*(((u8 *)((rp) + ((i)>>2)) + ((i)&3))))
236 #define SWIZ_PTR(p) ((unsigned char __user *)((p) ^ swiz))
238 #define __get_user_or_set_dar(_regs, _dest, _addr) \
241 typeof(_addr) __addr = (_addr); \
242 if (__get_user_inatomic(_dest, __addr)) { \
243 _regs->dar = (unsigned long)__addr; \
249 #define __put_user_or_set_dar(_regs, _src, _addr) \
252 typeof(_addr) __addr = (_addr); \
253 if (__put_user_inatomic(_src, __addr)) { \
254 _regs->dar = (unsigned long)__addr; \
260 static int emulate_multiple(struct pt_regs *regs, unsigned char __user *addr,
261 unsigned int reg, unsigned int nb,
262 unsigned int flags, unsigned int instr,
266 unsigned int nb0, i, bswiz;
270 * We do not try to emulate 8 bytes multiple as they aren't really
271 * available in our operating environments and we don't try to
272 * emulate multiples operations in kernel land as they should never
273 * be used/generated there at least not on unaligned boundaries
275 if (unlikely((nb > 4) || !user_mode(regs)))
278 /* lmw, stmw, lswi/x, stswi/x */
282 nb = regs->xer & 127;
286 unsigned long pc = regs->nip ^ (swiz & 4);
288 if (__get_user_or_set_dar(regs, instr,
289 (unsigned int __user *)pc))
292 if (swiz == 0 && (flags & SW))
293 instr = cpu_to_le32(instr);
294 nb = (instr >> 11) & 0x1f;
298 if (nb + reg * 4 > 128) {
299 nb0 = nb + reg * 4 - 128;
302 #ifdef __LITTLE_ENDIAN__
304 * String instructions are endian neutral but the code
305 * below is not. Force byte swapping on so that the
306 * effects of swizzling are undone in the load/store
316 if (!access_ok((flags & ST ? VERIFY_WRITE: VERIFY_READ), addr, nb+nb0))
317 return -EFAULT; /* bad address */
319 rptr = ®s->gpr[reg];
320 p = (unsigned long) addr;
321 bswiz = (flags & SW)? 3: 0;
325 * This zeroes the top 4 bytes of the affected registers
326 * in 64-bit mode, and also zeroes out any remaining
327 * bytes of the last register for lsw*.
329 memset(rptr, 0, ((nb + 3) / 4) * sizeof(unsigned long));
331 memset(®s->gpr[0], 0,
332 ((nb0 + 3) / 4) * sizeof(unsigned long));
334 for (i = 0; i < nb; ++i, ++p)
335 if (__get_user_or_set_dar(regs, REG_BYTE(rptr, i ^ bswiz),
339 rptr = ®s->gpr[0];
341 for (i = 0; i < nb0; ++i, ++p)
342 if (__get_user_or_set_dar(regs,
343 REG_BYTE(rptr, i ^ bswiz),
349 for (i = 0; i < nb; ++i, ++p)
350 if (__put_user_or_set_dar(regs, REG_BYTE(rptr, i ^ bswiz),
354 rptr = ®s->gpr[0];
356 for (i = 0; i < nb0; ++i, ++p)
357 if (__put_user_or_set_dar(regs,
358 REG_BYTE(rptr, i ^ bswiz),
367 * Emulate floating-point pair loads and stores.
368 * Only POWER6 has these instructions, and it does true little-endian,
369 * so we don't need the address swizzling.
371 static int emulate_fp_pair(struct pt_regs *regs, unsigned char __user *addr,
372 unsigned int reg, unsigned int flags)
374 char *ptr0 = (char *) ¤t->thread.TS_FPR(reg);
375 char *ptr1 = (char *) ¤t->thread.TS_FPR(reg+1);
379 return 0; /* invalid form: FRS/FRT must be even */
383 for (i = 0; i < 8; ++i) {
385 if (__get_user_or_set_dar(regs, ptr0[i^sw], addr + i))
387 if (__get_user_or_set_dar(regs, ptr1[i^sw], addr + i + 8))
390 if (__put_user_or_set_dar(regs, ptr0[i^sw], addr + i))
392 if (__put_user_or_set_dar(regs, ptr1[i^sw], addr + i + 8))
397 return 1; /* exception handled and fixed up */
401 static int emulate_lq_stq(struct pt_regs *regs, unsigned char __user *addr,
402 unsigned int reg, unsigned int flags)
404 char *ptr0 = (char *)®s->gpr[reg];
405 char *ptr1 = (char *)®s->gpr[reg+1];
409 return 0; /* invalid form: GPR must be even */
413 for (i = 0; i < 8; ++i) {
415 if (__get_user_or_set_dar(regs, ptr0[i^sw], addr + i))
417 if (__get_user_or_set_dar(regs, ptr1[i^sw], addr + i + 8))
420 if (__put_user_or_set_dar(regs, ptr0[i^sw], addr + i))
422 if (__put_user_or_set_dar(regs, ptr1[i^sw], addr + i + 8))
427 return 1; /* exception handled and fixed up */
429 #endif /* CONFIG_PPC64 */
433 static struct aligninfo spe_aligninfo[32] = {
434 { 8, LD+E8 }, /* 0 00 00: evldd[x] */
435 { 8, LD+E4 }, /* 0 00 01: evldw[x] */
436 { 8, LD }, /* 0 00 10: evldh[x] */
437 INVALID, /* 0 00 11 */
438 { 2, LD }, /* 0 01 00: evlhhesplat[x] */
439 INVALID, /* 0 01 01 */
440 { 2, LD }, /* 0 01 10: evlhhousplat[x] */
441 { 2, LD+SE }, /* 0 01 11: evlhhossplat[x] */
442 { 4, LD }, /* 0 10 00: evlwhe[x] */
443 INVALID, /* 0 10 01 */
444 { 4, LD }, /* 0 10 10: evlwhou[x] */
445 { 4, LD+SE }, /* 0 10 11: evlwhos[x] */
446 { 4, LD+E4 }, /* 0 11 00: evlwwsplat[x] */
447 INVALID, /* 0 11 01 */
448 { 4, LD }, /* 0 11 10: evlwhsplat[x] */
449 INVALID, /* 0 11 11 */
451 { 8, ST+E8 }, /* 1 00 00: evstdd[x] */
452 { 8, ST+E4 }, /* 1 00 01: evstdw[x] */
453 { 8, ST }, /* 1 00 10: evstdh[x] */
454 INVALID, /* 1 00 11 */
455 INVALID, /* 1 01 00 */
456 INVALID, /* 1 01 01 */
457 INVALID, /* 1 01 10 */
458 INVALID, /* 1 01 11 */
459 { 4, ST }, /* 1 10 00: evstwhe[x] */
460 INVALID, /* 1 10 01 */
461 { 4, ST }, /* 1 10 10: evstwho[x] */
462 INVALID, /* 1 10 11 */
463 { 4, ST+E4 }, /* 1 11 00: evstwwe[x] */
464 INVALID, /* 1 11 01 */
465 { 4, ST+E4 }, /* 1 11 10: evstwwo[x] */
466 INVALID, /* 1 11 11 */
472 #define EVLHHESPLAT 0x04
473 #define EVLHHOUSPLAT 0x06
474 #define EVLHHOSSPLAT 0x07
478 #define EVLWWSPLAT 0x0C
479 #define EVLWHSPLAT 0x0E
489 * Emulate SPE loads and stores.
490 * Only Book-E has these instructions, and it does true little-endian,
491 * so we don't need the address swizzling.
493 static int emulate_spe(struct pt_regs *regs, unsigned int reg,
503 unsigned char __user *p, *addr;
504 unsigned long *evr = ¤t->thread.evr[reg];
505 unsigned int nb, flags;
507 instr = (instr >> 1) & 0x1f;
509 /* DAR has the operand effective address */
510 addr = (unsigned char __user *)regs->dar;
512 nb = spe_aligninfo[instr].len;
513 flags = spe_aligninfo[instr].flags;
515 /* Verify the address of the operand */
516 if (unlikely(user_mode(regs) &&
517 !access_ok((flags & ST ? VERIFY_WRITE : VERIFY_READ),
522 if (unlikely(!user_mode(regs)))
525 flush_spe_to_thread(current);
527 /* If we are loading, get the data from user space, else
528 * get it from register values
537 data.w[1] = regs->gpr[reg];
540 data.h[2] = *evr >> 16;
541 data.h[3] = regs->gpr[reg] >> 16;
544 data.h[2] = *evr & 0xffff;
545 data.h[3] = regs->gpr[reg] & 0xffff;
551 data.w[1] = regs->gpr[reg];
557 temp.ll = data.ll = 0;
563 ret |= __get_user_inatomic(temp.v[0], p++);
564 ret |= __get_user_inatomic(temp.v[1], p++);
565 ret |= __get_user_inatomic(temp.v[2], p++);
566 ret |= __get_user_inatomic(temp.v[3], p++);
568 ret |= __get_user_inatomic(temp.v[4], p++);
569 ret |= __get_user_inatomic(temp.v[5], p++);
571 ret |= __get_user_inatomic(temp.v[6], p++);
572 ret |= __get_user_inatomic(temp.v[7], p++);
584 data.h[0] = temp.h[3];
585 data.h[2] = temp.h[3];
589 data.h[1] = temp.h[3];
590 data.h[3] = temp.h[3];
593 data.h[0] = temp.h[2];
594 data.h[2] = temp.h[3];
598 data.h[1] = temp.h[2];
599 data.h[3] = temp.h[3];
602 data.w[0] = temp.w[1];
603 data.w[1] = temp.w[1];
606 data.h[0] = temp.h[2];
607 data.h[1] = temp.h[2];
608 data.h[2] = temp.h[3];
609 data.h[3] = temp.h[3];
617 switch (flags & 0xf0) {
619 data.ll = swab64(data.ll);
622 data.w[0] = swab32(data.w[0]);
623 data.w[1] = swab32(data.w[1]);
625 /* Its half word endian */
627 data.h[0] = swab16(data.h[0]);
628 data.h[1] = swab16(data.h[1]);
629 data.h[2] = swab16(data.h[2]);
630 data.h[3] = swab16(data.h[3]);
636 data.w[0] = (s16)data.h[1];
637 data.w[1] = (s16)data.h[3];
640 /* Store result to memory or update registers */
646 ret |= __put_user_inatomic(data.v[0], p++);
647 ret |= __put_user_inatomic(data.v[1], p++);
648 ret |= __put_user_inatomic(data.v[2], p++);
649 ret |= __put_user_inatomic(data.v[3], p++);
651 ret |= __put_user_inatomic(data.v[4], p++);
652 ret |= __put_user_inatomic(data.v[5], p++);
654 ret |= __put_user_inatomic(data.v[6], p++);
655 ret |= __put_user_inatomic(data.v[7], p++);
661 regs->gpr[reg] = data.w[1];
666 #endif /* CONFIG_SPE */
670 * Emulate VSX instructions...
672 static int emulate_vsx(unsigned char __user *addr, unsigned int reg,
673 unsigned int areg, struct pt_regs *regs,
674 unsigned int flags, unsigned int length,
684 if (unlikely(!user_mode(regs)))
687 flush_vsx_to_thread(current);
690 ptr = (char *) ¤t->thread.fp_state.fpr[reg][0];
692 ptr = (char *) ¤t->thread.vr_state.vr[reg - 32];
694 lptr = (unsigned long *) ptr;
696 #ifdef __LITTLE_ENDIAN__
702 * The elements are BE ordered, even in LE mode, so process
703 * them in reverse order.
705 addr += length - elsize;
707 /* 8 byte memory accesses go in the top 8 bytes of the VR */
716 for (j = 0; j < length; j += elsize) {
717 for (i = 0; i < elsize; ++i) {
719 ret = __put_user_or_set_dar(regs, ptr[i^sw],
722 ret = __get_user_or_set_dar(regs, ptr[i^sw],
729 #ifdef __LITTLE_ENDIAN__
736 #ifdef __BIG_ENDIAN__
746 regs->gpr[areg] = regs->dar;
748 /* Splat load copies the same data to top and bottom 8 bytes */
750 lptr[VSX_LO] = lptr[VSX_HI];
751 /* For 8 byte loads, zero the low 8 bytes */
752 else if (!(flags & ST) && (8 == length))
762 * Called on alignment exception. Attempts to fixup
764 * Return 1 on success
765 * Return 0 if unable to handle the interrupt
766 * Return -EFAULT if data address is bad
769 int fix_alignment(struct pt_regs *regs)
771 unsigned int instr, nb, flags, instruction = 0;
772 unsigned int reg, areg;
774 unsigned char __user *addr;
775 unsigned long p, swiz;
782 #ifdef __LITTLE_ENDIAN__
791 #ifdef __LITTLE_ENDIAN__
793 unsigned char hi48[6];
795 unsigned char hi48[6];
802 * We require a complete register set, if not, then our assembly
805 CHECK_FULL_REGS(regs);
809 /* Some processors don't provide us with a DSISR we can use here,
810 * let's make one up from the instruction
812 if (cpu_has_feature(CPU_FTR_NODSISRALIGN)) {
813 unsigned long pc = regs->nip;
815 if (cpu_has_feature(CPU_FTR_PPC_LE) && (regs->msr & MSR_LE))
817 if (unlikely(__get_user_inatomic(instr,
818 (unsigned int __user *)pc)))
820 if (cpu_has_feature(CPU_FTR_REAL_LE) && (regs->msr & MSR_LE))
821 instr = cpu_to_le32(instr);
822 dsisr = make_dsisr(instr);
826 /* extract the operation and registers from the dsisr */
827 reg = (dsisr >> 5) & 0x1f; /* source/dest register */
828 areg = dsisr & 0x1f; /* register to update */
831 if ((instr >> 26) == 0x4) {
832 PPC_WARN_ALIGNMENT(spe, regs);
833 return emulate_spe(regs, reg, instr);
837 instr = (dsisr >> 10) & 0x7f;
838 instr |= (dsisr >> 13) & 0x60;
840 /* Lookup the operation in our table */
841 nb = aligninfo[instr].len;
842 flags = aligninfo[instr].flags;
845 * Handle some cases which give overlaps in the DSISR values.
847 if (IS_XFORM(instruction)) {
848 switch (get_xop(instruction)) {
849 case 532: /* ldbrx */
853 case 660: /* stdbrx */
859 case 116: /* lharx */
860 case 276: /* lqarx */
861 return 0; /* not emulated ever */
865 /* Byteswap little endian loads and stores */
867 if ((regs->msr & MSR_LE) != (MSR_KERNEL & MSR_LE)) {
869 #ifdef __BIG_ENDIAN__
871 * So-called "PowerPC little endian" mode works by
872 * swizzling addresses rather than by actually doing
873 * any byte-swapping. To emulate this, we XOR each
874 * byte address with 7. We also byte-swap, because
875 * the processor's address swizzling depends on the
876 * operand size (it xors the address with 7 for bytes,
877 * 6 for halfwords, 4 for words, 0 for doublewords) but
878 * we will xor with 7 and load/store each byte separately.
880 if (cpu_has_feature(CPU_FTR_PPC_LE))
885 /* DAR has the operand effective address */
886 addr = (unsigned char __user *)regs->dar;
889 if ((instruction & 0xfc00003e) == 0x7c000018) {
892 /* Additional register addressing bit (64 VSX vs 32 FPR/GPR) */
893 reg |= (instruction & 0x1) << 5;
894 /* Simple inline decoder instead of a table */
895 /* VSX has only 8 and 16 byte memory accesses */
897 if (instruction & 0x200)
900 /* Vector stores in little-endian mode swap individual
901 elements, so process them separately */
903 if (instruction & 0x80)
907 if ((regs->msr & MSR_LE) != (MSR_KERNEL & MSR_LE))
909 if (instruction & 0x100)
911 if (instruction & 0x040)
913 /* splat load needs a special decoder */
914 if ((instruction & 0x400) == 0){
918 PPC_WARN_ALIGNMENT(vsx, regs);
919 return emulate_vsx(addr, reg, areg, regs, flags, nb, elsize);
924 * ISA 3.0 (such as P9) copy, copy_first, paste and paste_last alignment
927 * Send a SIGBUS to the process that caused the fault.
929 * We do not emulate these because paste may contain additional metadata
930 * when pasting to a co-processor. Furthermore, paste_last is the
931 * synchronisation point for preceding copy/paste sequences.
933 if ((instruction & 0xfc0006fe) == PPC_INST_COPY)
936 /* A size of 0 indicates an instruction we don't support, with
937 * the exception of DCBZ which is handled as a special case here
940 PPC_WARN_ALIGNMENT(dcbz, regs);
941 return emulate_dcbz(regs, addr);
943 if (unlikely(nb == 0))
946 /* Load/Store Multiple instructions are handled in their own
950 PPC_WARN_ALIGNMENT(multiple, regs);
951 return emulate_multiple(regs, addr, reg, nb,
955 /* Verify the address of the operand */
956 if (unlikely(user_mode(regs) &&
957 !access_ok((flags & ST ? VERIFY_WRITE : VERIFY_READ),
961 /* Force the fprs into the save area so we can reference them */
964 if (unlikely(!user_mode(regs)))
966 flush_fp_to_thread(current);
971 /* Special case for 16-byte FP loads and stores */
972 PPC_WARN_ALIGNMENT(fp_pair, regs);
973 return emulate_fp_pair(regs, addr, reg, flags);
976 /* Special case for 16-byte loads and stores */
977 PPC_WARN_ALIGNMENT(lq_stq, regs);
978 return emulate_lq_stq(regs, addr, reg, flags);
985 PPC_WARN_ALIGNMENT(unaligned, regs);
987 /* If we are loading, get the data from user space, else
988 * get it from register values
991 unsigned int start = 0;
995 start = offsetof(union data, x32.low32);
998 start = offsetof(union data, x16.low16);
1003 p = (unsigned long)addr;
1005 for (i = 0; i < nb; i++)
1006 if (__get_user_or_set_dar(regs, data.v[start + i],
1010 } else if (flags & F) {
1011 data.ll = current->thread.TS_FPR(reg);
1013 /* Single-precision FP store requires conversion... */
1014 #ifdef CONFIG_PPC_FPU
1017 cvt_df(&data.dd, (float *)&data.x32.low32);
1018 disable_kernel_fp();
1025 data.ll = regs->gpr[reg];
1030 data.ll = swab64(data.ll);
1033 data.x32.low32 = swab32(data.x32.low32);
1036 data.x16.low16 = swab16(data.x16.low16);
1041 /* Perform other misc operations like sign extension
1042 * or floating point single precision conversion
1044 switch (flags & ~(U|SW)) {
1045 case LD+SE: /* sign extending integer loads */
1046 case LD+F+SE: /* sign extend for lfiwax */
1048 data.ll = data.x16.low16;
1049 else /* nb must be 4 */
1050 data.ll = data.x32.low32;
1053 /* Single-precision FP load requires conversion... */
1055 #ifdef CONFIG_PPC_FPU
1058 cvt_fd((float *)&data.x32.low32, &data.dd);
1059 disable_kernel_fp();
1067 /* Store result to memory or update registers */
1069 unsigned int start = 0;
1073 start = offsetof(union data, x32.low32);
1076 start = offsetof(union data, x16.low16);
1080 p = (unsigned long)addr;
1082 for (i = 0; i < nb; i++)
1083 if (__put_user_or_set_dar(regs, data.v[start + i],
1087 } else if (flags & F)
1088 current->thread.TS_FPR(reg) = data.ll;
1090 regs->gpr[reg] = data.ll;
1092 /* Update RA as needed */
1094 regs->gpr[areg] = regs->dar;