1 /* -*- linux-c -*- --------------------------- 1 /* -*- linux-c -*- ------------------------------------------------------- *
2 * 2 *
3 * Copyright 2002-2004 H. Peter Anvin - All 3 * Copyright 2002-2004 H. Peter Anvin - All Rights Reserved
4 * 4 *
5 * This program is free software; you can re 5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Pub 6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation, Inc., 53 Te 7 * the Free Software Foundation, Inc., 53 Temple Place Ste 330,
8 * Boston MA 02111-1307, USA; either version 8 * Boston MA 02111-1307, USA; either version 2 of the License, or
9 * (at your option) any later version; incor 9 * (at your option) any later version; incorporated herein by reference.
10 * 10 *
11 * ------------------------------------------- 11 * ----------------------------------------------------------------------- */
12 12
13 /* 13 /*
14 * raid6altivec$#.c 14 * raid6altivec$#.c
15 * 15 *
16 * $#-way unrolled portable integer math RAID- 16 * $#-way unrolled portable integer math RAID-6 instruction set
17 * 17 *
18 * This file is postprocessed using unroll.awk 18 * This file is postprocessed using unroll.awk
19 * 19 *
20 * <benh> hpa: in process, 20 * <benh> hpa: in process,
21 * you can just "steal" the vec unit with enab 21 * you can just "steal" the vec unit with enable_kernel_altivec() (but
22 * bracked this with preempt_disable/enable or 22 * bracked this with preempt_disable/enable or in a lock)
23 */ 23 */
24 24
25 #include <linux/raid/pq.h> 25 #include <linux/raid/pq.h>
26 26
27 #ifdef CONFIG_ALTIVEC 27 #ifdef CONFIG_ALTIVEC
28 28
29 #include <altivec.h> 29 #include <altivec.h>
30 #ifdef __KERNEL__ 30 #ifdef __KERNEL__
31 # include <asm/cputable.h> 31 # include <asm/cputable.h>
32 # include <asm/switch_to.h> 32 # include <asm/switch_to.h>
33 #endif /* __KERNEL__ */ 33 #endif /* __KERNEL__ */
34 34
35 /* 35 /*
36 * This is the C data type to use. We use a v 36 * This is the C data type to use. We use a vector of
37 * signed char so vec_cmpgt() will generate th 37 * signed char so vec_cmpgt() will generate the right
38 * instruction. 38 * instruction.
39 */ 39 */
40 40
41 typedef vector signed char unative_t; 41 typedef vector signed char unative_t;
42 42
43 #define NBYTES(x) ((vector signed char) {x,x,x 43 #define NBYTES(x) ((vector signed char) {x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x})
44 #define NSIZE sizeof(unative_t) 44 #define NSIZE sizeof(unative_t)
45 45
46 /* 46 /*
47 * The SHLBYTE() operation shifts each byte le 47 * The SHLBYTE() operation shifts each byte left by 1, *not*
48 * rolling over into the next byte 48 * rolling over into the next byte
49 */ 49 */
50 static inline __attribute_const__ unative_t SH 50 static inline __attribute_const__ unative_t SHLBYTE(unative_t v)
51 { 51 {
52 return vec_add(v,v); 52 return vec_add(v,v);
53 } 53 }
54 54
55 /* 55 /*
56 * The MASK() operation returns 0xFF in any by 56 * The MASK() operation returns 0xFF in any byte for which the high
57 * bit is 1, 0x00 for any byte for which the h 57 * bit is 1, 0x00 for any byte for which the high bit is 0.
58 */ 58 */
59 static inline __attribute_const__ unative_t MA 59 static inline __attribute_const__ unative_t MASK(unative_t v)
60 { 60 {
61 unative_t zv = NBYTES(0); 61 unative_t zv = NBYTES(0);
62 62
63 /* vec_cmpgt returns a vector bool cha 63 /* vec_cmpgt returns a vector bool char; thus the need for the cast */
64 return (unative_t)vec_cmpgt(zv, v); 64 return (unative_t)vec_cmpgt(zv, v);
65 } 65 }
66 66
67 67
68 /* This is noinline to make damned sure that g 68 /* This is noinline to make damned sure that gcc doesn't move any of the
69 Altivec code around the enable/disable code 69 Altivec code around the enable/disable code */
70 static void noinline 70 static void noinline
71 raid6_altivec$#_gen_syndrome_real(int disks, s 71 raid6_altivec$#_gen_syndrome_real(int disks, size_t bytes, void **ptrs)
72 { 72 {
73 u8 **dptr = (u8 **)ptrs; 73 u8 **dptr = (u8 **)ptrs;
74 u8 *p, *q; 74 u8 *p, *q;
75 int d, z, z0; 75 int d, z, z0;
76 76
77 unative_t wd$$, wq$$, wp$$, w1$$, w2$$ 77 unative_t wd$$, wq$$, wp$$, w1$$, w2$$;
78 unative_t x1d = NBYTES(0x1d); 78 unative_t x1d = NBYTES(0x1d);
79 79
80 z0 = disks - 3; /* Highest dat 80 z0 = disks - 3; /* Highest data disk */
81 p = dptr[z0+1]; /* XOR parity 81 p = dptr[z0+1]; /* XOR parity */
82 q = dptr[z0+2]; /* RS syndrome 82 q = dptr[z0+2]; /* RS syndrome */
83 83
84 for ( d = 0 ; d < bytes ; d += NSIZE*$ 84 for ( d = 0 ; d < bytes ; d += NSIZE*$# ) {
85 wq$$ = wp$$ = *(unative_t *)&d 85 wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE];
86 for ( z = z0-1 ; z >= 0 ; z-- 86 for ( z = z0-1 ; z >= 0 ; z-- ) {
87 wd$$ = *(unative_t *)& 87 wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE];
88 wp$$ = vec_xor(wp$$, w 88 wp$$ = vec_xor(wp$$, wd$$);
89 w2$$ = MASK(wq$$); 89 w2$$ = MASK(wq$$);
90 w1$$ = SHLBYTE(wq$$); 90 w1$$ = SHLBYTE(wq$$);
91 w2$$ = vec_and(w2$$, x 91 w2$$ = vec_and(w2$$, x1d);
92 w1$$ = vec_xor(w1$$, w 92 w1$$ = vec_xor(w1$$, w2$$);
93 wq$$ = vec_xor(w1$$, w 93 wq$$ = vec_xor(w1$$, wd$$);
94 } 94 }
95 *(unative_t *)&p[d+NSIZE*$$] = 95 *(unative_t *)&p[d+NSIZE*$$] = wp$$;
96 *(unative_t *)&q[d+NSIZE*$$] = 96 *(unative_t *)&q[d+NSIZE*$$] = wq$$;
97 } 97 }
98 } 98 }
99 99
100 static void raid6_altivec$#_gen_syndrome(int d 100 static void raid6_altivec$#_gen_syndrome(int disks, size_t bytes, void **ptrs)
101 { 101 {
102 preempt_disable(); 102 preempt_disable();
103 enable_kernel_altivec(); 103 enable_kernel_altivec();
104 104
105 raid6_altivec$#_gen_syndrome_real(disk 105 raid6_altivec$#_gen_syndrome_real(disks, bytes, ptrs);
106 106
107 disable_kernel_altivec(); 107 disable_kernel_altivec();
108 preempt_enable(); 108 preempt_enable();
109 } 109 }
110 110
111 int raid6_have_altivec(void); 111 int raid6_have_altivec(void);
112 #if $# == 1 112 #if $# == 1
113 int raid6_have_altivec(void) 113 int raid6_have_altivec(void)
114 { 114 {
115 /* This assumes either all CPUs have A 115 /* This assumes either all CPUs have Altivec or none does */
116 # ifdef __KERNEL__ 116 # ifdef __KERNEL__
117 return cpu_has_feature(CPU_FTR_ALTIVEC 117 return cpu_has_feature(CPU_FTR_ALTIVEC);
118 # else 118 # else
119 return 1; 119 return 1;
120 # endif 120 # endif
121 } 121 }
122 #endif 122 #endif
123 123
124 const struct raid6_calls raid6_altivec$# = { 124 const struct raid6_calls raid6_altivec$# = {
125 raid6_altivec$#_gen_syndrome, 125 raid6_altivec$#_gen_syndrome,
126 NULL, /* XOR not yet 126 NULL, /* XOR not yet implemented */
127 raid6_have_altivec, 127 raid6_have_altivec,
128 "altivecx$#", 128 "altivecx$#",
129 0 129 0
130 }; 130 };
131 131
132 #endif /* CONFIG_ALTIVEC */ 132 #endif /* CONFIG_ALTIVEC */
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.