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/system.h>
25 #include <asm/cache.h>
26 #include <asm/cputable.h>
33 #define IS_XFORM(inst) (((inst) >> 26) == 31)
34 #define IS_DSFORM(inst) (((inst) >> 26) >= 56)
36 #define INVALID { 0, 0 }
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 */
52 /* DSISR bits reported for a DCBZ instruction: */
53 #define DCBZ 0x5f /* 8xx/82xx dcbz faults when cache not enabled */
55 #define SWAP(a, b) (t = (a), (a) = (b), (b) = t)
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 INVALID, /* 00 0 1100 */
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 INVALID, /* 10 0 1111 */
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 INVALID, /* 11 1 1101 */
190 INVALID, /* 11 1 1110 */
191 INVALID, /* 11 1 1111 */
195 * Create a DSISR value from the instruction
197 static inline unsigned make_dsisr(unsigned instr)
202 /* bits 6:15 --> 22:31 */
203 dsisr = (instr & 0x03ff0000) >> 16;
205 if (IS_XFORM(instr)) {
206 /* bits 29:30 --> 15:16 */
207 dsisr |= (instr & 0x00000006) << 14;
209 dsisr |= (instr & 0x00000040) << 8;
210 /* bits 21:24 --> 18:21 */
211 dsisr |= (instr & 0x00000780) << 3;
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;
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
232 static int emulate_dcbz(struct pt_regs *regs, unsigned char __user *addr)
238 size = ppc64_caches.dline_size;
240 size = L1_CACHE_BYTES;
242 p = (long __user *) (regs->dar & -size);
243 if (user_mode(regs) && !access_ok(VERIFY_WRITE, p, size))
245 for (i = 0; i < size / sizeof(long); ++i)
246 if (__put_user_inatomic(0, p+i))
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.
258 #define REG_BYTE(rp, i) *((u8 *)((rp) + ((i) >> 2)) + ((i) & 3) + 4)
260 #define REG_BYTE(rp, i) *((u8 *)(rp) + (i))
263 #define SWIZ_PTR(p) ((unsigned char __user *)((p) ^ swiz))
265 static int emulate_multiple(struct pt_regs *regs, unsigned char __user *addr,
266 unsigned int reg, unsigned int nb,
267 unsigned int flags, unsigned int instr,
271 unsigned int nb0, i, bswiz;
275 * We do not try to emulate 8 bytes multiple as they aren't really
276 * available in our operating environments and we don't try to
277 * emulate multiples operations in kernel land as they should never
278 * be used/generated there at least not on unaligned boundaries
280 if (unlikely((nb > 4) || !user_mode(regs)))
283 /* lmw, stmw, lswi/x, stswi/x */
287 nb = regs->xer & 127;
291 unsigned long pc = regs->nip ^ (swiz & 4);
293 if (__get_user_inatomic(instr,
294 (unsigned int __user *)pc))
296 if (swiz == 0 && (flags & SW))
297 instr = cpu_to_le32(instr);
298 nb = (instr >> 11) & 0x1f;
302 if (nb + reg * 4 > 128) {
303 nb0 = nb + reg * 4 - 128;
311 if (!access_ok((flags & ST ? VERIFY_WRITE: VERIFY_READ), addr, nb+nb0))
312 return -EFAULT; /* bad address */
314 rptr = ®s->gpr[reg];
315 p = (unsigned long) addr;
316 bswiz = (flags & SW)? 3: 0;
320 * This zeroes the top 4 bytes of the affected registers
321 * in 64-bit mode, and also zeroes out any remaining
322 * bytes of the last register for lsw*.
324 memset(rptr, 0, ((nb + 3) / 4) * sizeof(unsigned long));
326 memset(®s->gpr[0], 0,
327 ((nb0 + 3) / 4) * sizeof(unsigned long));
329 for (i = 0; i < nb; ++i, ++p)
330 if (__get_user_inatomic(REG_BYTE(rptr, i ^ bswiz),
334 rptr = ®s->gpr[0];
336 for (i = 0; i < nb0; ++i, ++p)
337 if (__get_user_inatomic(REG_BYTE(rptr,
344 for (i = 0; i < nb; ++i, ++p)
345 if (__put_user_inatomic(REG_BYTE(rptr, i ^ bswiz),
349 rptr = ®s->gpr[0];
351 for (i = 0; i < nb0; ++i, ++p)
352 if (__put_user_inatomic(REG_BYTE(rptr,
362 * Emulate floating-point pair loads and stores.
363 * Only POWER6 has these instructions, and it does true little-endian,
364 * so we don't need the address swizzling.
366 static int emulate_fp_pair(struct pt_regs *regs, unsigned char __user *addr,
367 unsigned int reg, unsigned int flags)
369 char *ptr = (char *) ¤t->thread.fpr[reg];
375 return 0; /* invalid form: FRS/FRT must be even */
377 /* not byte-swapped - easy */
379 ret = __copy_from_user(ptr, addr, 16);
381 ret = __copy_to_user(addr, ptr, 16);
383 /* each FPR value is byte-swapped separately */
385 for (i = 0; i < 16; ++i) {
387 ret |= __get_user(ptr[i^7], addr + i);
389 ret |= __put_user(ptr[i^7], addr + i);
394 return 1; /* exception handled and fixed up */
399 static struct aligninfo spe_aligninfo[32] = {
400 { 8, LD+E8 }, /* 0 00 00: evldd[x] */
401 { 8, LD+E4 }, /* 0 00 01: evldw[x] */
402 { 8, LD }, /* 0 00 10: evldh[x] */
403 INVALID, /* 0 00 11 */
404 { 2, LD }, /* 0 01 00: evlhhesplat[x] */
405 INVALID, /* 0 01 01 */
406 { 2, LD }, /* 0 01 10: evlhhousplat[x] */
407 { 2, LD+SE }, /* 0 01 11: evlhhossplat[x] */
408 { 4, LD }, /* 0 10 00: evlwhe[x] */
409 INVALID, /* 0 10 01 */
410 { 4, LD }, /* 0 10 10: evlwhou[x] */
411 { 4, LD+SE }, /* 0 10 11: evlwhos[x] */
412 { 4, LD+E4 }, /* 0 11 00: evlwwsplat[x] */
413 INVALID, /* 0 11 01 */
414 { 4, LD }, /* 0 11 10: evlwhsplat[x] */
415 INVALID, /* 0 11 11 */
417 { 8, ST+E8 }, /* 1 00 00: evstdd[x] */
418 { 8, ST+E4 }, /* 1 00 01: evstdw[x] */
419 { 8, ST }, /* 1 00 10: evstdh[x] */
420 INVALID, /* 1 00 11 */
421 INVALID, /* 1 01 00 */
422 INVALID, /* 1 01 01 */
423 INVALID, /* 1 01 10 */
424 INVALID, /* 1 01 11 */
425 { 4, ST }, /* 1 10 00: evstwhe[x] */
426 INVALID, /* 1 10 01 */
427 { 4, ST }, /* 1 10 10: evstwho[x] */
428 INVALID, /* 1 10 11 */
429 { 4, ST+E4 }, /* 1 11 00: evstwwe[x] */
430 INVALID, /* 1 11 01 */
431 { 4, ST+E4 }, /* 1 11 10: evstwwo[x] */
432 INVALID, /* 1 11 11 */
438 #define EVLHHESPLAT 0x04
439 #define EVLHHOUSPLAT 0x06
440 #define EVLHHOSSPLAT 0x07
444 #define EVLWWSPLAT 0x0C
445 #define EVLWHSPLAT 0x0E
455 * Emulate SPE loads and stores.
456 * Only Book-E has these instructions, and it does true little-endian,
457 * so we don't need the address swizzling.
459 static int emulate_spe(struct pt_regs *regs, unsigned int reg,
469 unsigned char __user *p, *addr;
470 unsigned long *evr = ¤t->thread.evr[reg];
471 unsigned int nb, flags;
473 instr = (instr >> 1) & 0x1f;
475 /* DAR has the operand effective address */
476 addr = (unsigned char __user *)regs->dar;
478 nb = spe_aligninfo[instr].len;
479 flags = spe_aligninfo[instr].flags;
481 /* Verify the address of the operand */
482 if (unlikely(user_mode(regs) &&
483 !access_ok((flags & ST ? VERIFY_WRITE : VERIFY_READ),
488 if (unlikely(!user_mode(regs)))
491 flush_spe_to_thread(current);
493 /* If we are loading, get the data from user space, else
494 * get it from register values
503 data.w[1] = regs->gpr[reg];
506 data.h[2] = *evr >> 16;
507 data.h[3] = regs->gpr[reg] >> 16;
510 data.h[2] = *evr & 0xffff;
511 data.h[3] = regs->gpr[reg] & 0xffff;
517 data.w[1] = regs->gpr[reg];
523 temp.ll = data.ll = 0;
529 ret |= __get_user_inatomic(temp.v[0], p++);
530 ret |= __get_user_inatomic(temp.v[1], p++);
531 ret |= __get_user_inatomic(temp.v[2], p++);
532 ret |= __get_user_inatomic(temp.v[3], p++);
534 ret |= __get_user_inatomic(temp.v[4], p++);
535 ret |= __get_user_inatomic(temp.v[5], p++);
537 ret |= __get_user_inatomic(temp.v[6], p++);
538 ret |= __get_user_inatomic(temp.v[7], p++);
550 data.h[0] = temp.h[3];
551 data.h[2] = temp.h[3];
555 data.h[1] = temp.h[3];
556 data.h[3] = temp.h[3];
559 data.h[0] = temp.h[2];
560 data.h[2] = temp.h[3];
564 data.h[1] = temp.h[2];
565 data.h[3] = temp.h[3];
568 data.w[0] = temp.w[1];
569 data.w[1] = temp.w[1];
572 data.h[0] = temp.h[2];
573 data.h[1] = temp.h[2];
574 data.h[2] = temp.h[3];
575 data.h[3] = temp.h[3];
583 switch (flags & 0xf0) {
585 SWAP(data.v[0], data.v[7]);
586 SWAP(data.v[1], data.v[6]);
587 SWAP(data.v[2], data.v[5]);
588 SWAP(data.v[3], data.v[4]);
592 SWAP(data.v[0], data.v[3]);
593 SWAP(data.v[1], data.v[2]);
594 SWAP(data.v[4], data.v[7]);
595 SWAP(data.v[5], data.v[6]);
597 /* Its half word endian */
599 SWAP(data.v[0], data.v[1]);
600 SWAP(data.v[2], data.v[3]);
601 SWAP(data.v[4], data.v[5]);
602 SWAP(data.v[6], data.v[7]);
608 data.w[0] = (s16)data.h[1];
609 data.w[1] = (s16)data.h[3];
612 /* Store result to memory or update registers */
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++);
623 ret |= __put_user_inatomic(data.v[4], p++);
624 ret |= __put_user_inatomic(data.v[5], p++);
626 ret |= __put_user_inatomic(data.v[6], p++);
627 ret |= __put_user_inatomic(data.v[7], p++);
633 regs->gpr[reg] = data.w[1];
638 #endif /* CONFIG_SPE */
641 * Called on alignment exception. Attempts to fixup
643 * Return 1 on success
644 * Return 0 if unable to handle the interrupt
645 * Return -EFAULT if data address is bad
648 int fix_alignment(struct pt_regs *regs)
650 unsigned int instr, nb, flags;
651 unsigned int reg, areg;
653 unsigned char __user *addr;
654 unsigned long p, swiz;
665 unsigned char hi48[6];
671 * We require a complete register set, if not, then our assembly
674 CHECK_FULL_REGS(regs);
678 /* Some processors don't provide us with a DSISR we can use here,
679 * let's make one up from the instruction
681 if (cpu_has_feature(CPU_FTR_NODSISRALIGN)) {
682 unsigned long pc = regs->nip;
684 if (cpu_has_feature(CPU_FTR_PPC_LE) && (regs->msr & MSR_LE))
686 if (unlikely(__get_user_inatomic(instr,
687 (unsigned int __user *)pc)))
689 if (cpu_has_feature(CPU_FTR_REAL_LE) && (regs->msr & MSR_LE))
690 instr = cpu_to_le32(instr);
691 dsisr = make_dsisr(instr);
694 /* extract the operation and registers from the dsisr */
695 reg = (dsisr >> 5) & 0x1f; /* source/dest register */
696 areg = dsisr & 0x1f; /* register to update */
699 if ((instr >> 26) == 0x4)
700 return emulate_spe(regs, reg, instr);
703 instr = (dsisr >> 10) & 0x7f;
704 instr |= (dsisr >> 13) & 0x60;
706 /* Lookup the operation in our table */
707 nb = aligninfo[instr].len;
708 flags = aligninfo[instr].flags;
710 /* Byteswap little endian loads and stores */
712 if (regs->msr & MSR_LE) {
715 * So-called "PowerPC little endian" mode works by
716 * swizzling addresses rather than by actually doing
717 * any byte-swapping. To emulate this, we XOR each
718 * byte address with 7. We also byte-swap, because
719 * the processor's address swizzling depends on the
720 * operand size (it xors the address with 7 for bytes,
721 * 6 for halfwords, 4 for words, 0 for doublewords) but
722 * we will xor with 7 and load/store each byte separately.
724 if (cpu_has_feature(CPU_FTR_PPC_LE))
728 /* DAR has the operand effective address */
729 addr = (unsigned char __user *)regs->dar;
731 /* A size of 0 indicates an instruction we don't support, with
732 * the exception of DCBZ which is handled as a special case here
735 return emulate_dcbz(regs, addr);
736 if (unlikely(nb == 0))
739 /* Load/Store Multiple instructions are handled in their own
743 return emulate_multiple(regs, addr, reg, nb,
746 /* Verify the address of the operand */
747 if (unlikely(user_mode(regs) &&
748 !access_ok((flags & ST ? VERIFY_WRITE : VERIFY_READ),
752 /* Force the fprs into the save area so we can reference them */
755 if (unlikely(!user_mode(regs)))
757 flush_fp_to_thread(current);
760 /* Special case for 16-byte FP loads and stores */
762 return emulate_fp_pair(regs, addr, reg, flags);
764 /* If we are loading, get the data from user space, else
765 * get it from register values
770 p = (unsigned long) addr;
773 ret |= __get_user_inatomic(data.v[0], SWIZ_PTR(p++));
774 ret |= __get_user_inatomic(data.v[1], SWIZ_PTR(p++));
775 ret |= __get_user_inatomic(data.v[2], SWIZ_PTR(p++));
776 ret |= __get_user_inatomic(data.v[3], SWIZ_PTR(p++));
778 ret |= __get_user_inatomic(data.v[4], SWIZ_PTR(p++));
779 ret |= __get_user_inatomic(data.v[5], SWIZ_PTR(p++));
781 ret |= __get_user_inatomic(data.v[6], SWIZ_PTR(p++));
782 ret |= __get_user_inatomic(data.v[7], SWIZ_PTR(p++));
786 } else if (flags & F) {
787 data.dd = current->thread.fpr[reg];
789 /* Single-precision FP store requires conversion... */
790 #ifdef CONFIG_PPC_FPU
793 cvt_df(&data.dd, (float *)&data.v[4], ¤t->thread);
800 data.ll = regs->gpr[reg];
805 SWAP(data.v[0], data.v[7]);
806 SWAP(data.v[1], data.v[6]);
807 SWAP(data.v[2], data.v[5]);
808 SWAP(data.v[3], data.v[4]);
811 SWAP(data.v[4], data.v[7]);
812 SWAP(data.v[5], data.v[6]);
815 SWAP(data.v[6], data.v[7]);
820 /* Perform other misc operations like sign extension
821 * or floating point single precision conversion
823 switch (flags & ~(U|SW)) {
824 case LD+SE: /* sign extending integer loads */
825 case LD+F+SE: /* sign extend for lfiwax */
827 data.ll = data.x16.low16;
828 else /* nb must be 4 */
829 data.ll = data.x32.low32;
832 /* Single-precision FP load requires conversion... */
834 #ifdef CONFIG_PPC_FPU
837 cvt_fd((float *)&data.v[4], &data.dd, ¤t->thread);
845 /* Store result to memory or update registers */
848 p = (unsigned long) addr;
851 ret |= __put_user_inatomic(data.v[0], SWIZ_PTR(p++));
852 ret |= __put_user_inatomic(data.v[1], SWIZ_PTR(p++));
853 ret |= __put_user_inatomic(data.v[2], SWIZ_PTR(p++));
854 ret |= __put_user_inatomic(data.v[3], SWIZ_PTR(p++));
856 ret |= __put_user_inatomic(data.v[4], SWIZ_PTR(p++));
857 ret |= __put_user_inatomic(data.v[5], SWIZ_PTR(p++));
859 ret |= __put_user_inatomic(data.v[6], SWIZ_PTR(p++));
860 ret |= __put_user_inatomic(data.v[7], SWIZ_PTR(p++));
864 } else if (flags & F)
865 current->thread.fpr[reg] = data.dd;
867 regs->gpr[reg] = data.ll;
869 /* Update RA as needed */
871 regs->gpr[areg] = regs->dar;