1 // SPDX-License-Identifier: GPL-2.0 1 // SPDX-License-Identifier: GPL-2.0
2 /* 2 /*
3 * raid6_vx$#.c 3 * raid6_vx$#.c
4 * 4 *
5 * $#-way unrolled RAID6 gen/xor functions for 5 * $#-way unrolled RAID6 gen/xor functions for s390
6 * based on the vector facility 6 * based on the vector facility
7 * 7 *
8 * Copyright IBM Corp. 2016 8 * Copyright IBM Corp. 2016
9 * Author(s): Martin Schwidefsky <schwidefsky@d 9 * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
10 * 10 *
11 * This file is postprocessed using unroll.awk 11 * This file is postprocessed using unroll.awk.
12 */ 12 */
13 13
14 #include <linux/raid/pq.h> 14 #include <linux/raid/pq.h>
15 #include <asm/fpu.h> !! 15 #include <asm/fpu/api.h>
>> 16 #include <asm/vx-insn.h>
16 17
17 #define NSIZE 16 18 #define NSIZE 16
18 19
19 static __always_inline void LOAD_CONST(void) !! 20 static inline void LOAD_CONST(void)
20 { 21 {
21 fpu_vrepib(24, 0x07); !! 22 asm volatile("VREPIB %v24,7");
22 fpu_vrepib(25, 0x1d); !! 23 asm volatile("VREPIB %v25,0x1d");
23 } 24 }
24 25
25 /* 26 /*
26 * The SHLBYTE() operation shifts each of the 27 * The SHLBYTE() operation shifts each of the 16 bytes in
27 * vector register y left by 1 bit and stores 28 * vector register y left by 1 bit and stores the result in
28 * vector register x. 29 * vector register x.
29 */ 30 */
30 #define SHLBYTE(x, y) fpu_vab(x, y, !! 31 static inline void SHLBYTE(int x, int y)
>> 32 {
>> 33 asm volatile ("VAB %0,%1,%1" : : "i" (x), "i" (y));
>> 34 }
31 35
32 /* 36 /*
33 * For each of the 16 bytes in the vector regi 37 * For each of the 16 bytes in the vector register y the MASK()
34 * operation returns 0xFF if the high bit of t 38 * operation returns 0xFF if the high bit of the byte is 1,
35 * or 0x00 if the high bit is 0. The result is 39 * or 0x00 if the high bit is 0. The result is stored in vector
36 * register x. 40 * register x.
37 */ 41 */
38 #define MASK(x, y) fpu_vesravb(x, !! 42 static inline void MASK(int x, int y)
>> 43 {
>> 44 asm volatile ("VESRAVB %0,%1,24" : : "i" (x), "i" (y));
>> 45 }
>> 46
>> 47 static inline void AND(int x, int y, int z)
>> 48 {
>> 49 asm volatile ("VN %0,%1,%2" : : "i" (x), "i" (y), "i" (z));
>> 50 }
>> 51
>> 52 static inline void XOR(int x, int y, int z)
>> 53 {
>> 54 asm volatile ("VX %0,%1,%2" : : "i" (x), "i" (y), "i" (z));
>> 55 }
39 56
40 #define AND(x, y, z) fpu_vn(x, y, z !! 57 static inline void LOAD_DATA(int x, u8 *ptr)
41 #define XOR(x, y, z) fpu_vx(x, y, z !! 58 {
42 #define LOAD_DATA(x, ptr) fpu_vlm(x, x + !! 59 typedef struct { u8 _[16 * $#]; } addrtype;
43 #define STORE_DATA(x, ptr) fpu_vstm(x, x !! 60 register addrtype *__ptr asm("1") = (addrtype *) ptr;
44 #define COPY_VEC(x, y) fpu_vlr(x, y) !! 61
>> 62 asm volatile ("VLM %2,%3,0,%1"
>> 63 : : "m" (*__ptr), "a" (__ptr), "i" (x),
>> 64 "i" (x + $# - 1));
>> 65 }
>> 66
>> 67 static inline void STORE_DATA(int x, u8 *ptr)
>> 68 {
>> 69 typedef struct { u8 _[16 * $#]; } addrtype;
>> 70 register addrtype *__ptr asm("1") = (addrtype *) ptr;
>> 71
>> 72 asm volatile ("VSTM %2,%3,0,1"
>> 73 : "=m" (*__ptr) : "a" (__ptr), "i" (x),
>> 74 "i" (x + $# - 1));
>> 75 }
>> 76
>> 77 static inline void COPY_VEC(int x, int y)
>> 78 {
>> 79 asm volatile ("VLR %0,%1" : : "i" (x), "i" (y));
>> 80 }
45 81
46 static void raid6_s390vx$#_gen_syndrome(int di 82 static void raid6_s390vx$#_gen_syndrome(int disks, size_t bytes, void **ptrs)
47 { 83 {
48 DECLARE_KERNEL_FPU_ONSTACK32(vxstate); !! 84 struct kernel_fpu vxstate;
49 u8 **dptr, *p, *q; 85 u8 **dptr, *p, *q;
50 int d, z, z0; 86 int d, z, z0;
51 87
52 kernel_fpu_begin(&vxstate, KERNEL_VXR) 88 kernel_fpu_begin(&vxstate, KERNEL_VXR);
53 LOAD_CONST(); 89 LOAD_CONST();
54 90
55 dptr = (u8 **) ptrs; 91 dptr = (u8 **) ptrs;
56 z0 = disks - 3; /* Highest dat 92 z0 = disks - 3; /* Highest data disk */
57 p = dptr[z0 + 1]; /* XOR parity 93 p = dptr[z0 + 1]; /* XOR parity */
58 q = dptr[z0 + 2]; /* RS syndrome 94 q = dptr[z0 + 2]; /* RS syndrome */
59 95
60 for (d = 0; d < bytes; d += $#*NSIZE) 96 for (d = 0; d < bytes; d += $#*NSIZE) {
61 LOAD_DATA(0,&dptr[z0][d]); 97 LOAD_DATA(0,&dptr[z0][d]);
62 COPY_VEC(8+$$,0+$$); 98 COPY_VEC(8+$$,0+$$);
63 for (z = z0 - 1; z >= 0; z--) 99 for (z = z0 - 1; z >= 0; z--) {
64 MASK(16+$$,8+$$); 100 MASK(16+$$,8+$$);
65 AND(16+$$,16+$$,25); 101 AND(16+$$,16+$$,25);
66 SHLBYTE(8+$$,8+$$); 102 SHLBYTE(8+$$,8+$$);
67 XOR(8+$$,8+$$,16+$$); 103 XOR(8+$$,8+$$,16+$$);
68 LOAD_DATA(16,&dptr[z][ 104 LOAD_DATA(16,&dptr[z][d]);
69 XOR(0+$$,0+$$,16+$$); 105 XOR(0+$$,0+$$,16+$$);
70 XOR(8+$$,8+$$,16+$$); 106 XOR(8+$$,8+$$,16+$$);
71 } 107 }
72 STORE_DATA(0,&p[d]); 108 STORE_DATA(0,&p[d]);
73 STORE_DATA(8,&q[d]); 109 STORE_DATA(8,&q[d]);
74 } 110 }
75 kernel_fpu_end(&vxstate, KERNEL_VXR); 111 kernel_fpu_end(&vxstate, KERNEL_VXR);
76 } 112 }
77 113
78 static void raid6_s390vx$#_xor_syndrome(int di 114 static void raid6_s390vx$#_xor_syndrome(int disks, int start, int stop,
79 size_t 115 size_t bytes, void **ptrs)
80 { 116 {
81 DECLARE_KERNEL_FPU_ONSTACK32(vxstate); !! 117 struct kernel_fpu vxstate;
82 u8 **dptr, *p, *q; 118 u8 **dptr, *p, *q;
83 int d, z, z0; 119 int d, z, z0;
84 120
85 dptr = (u8 **) ptrs; 121 dptr = (u8 **) ptrs;
86 z0 = stop; /* P/Q right s 122 z0 = stop; /* P/Q right side optimization */
87 p = dptr[disks - 2]; /* XOR parity 123 p = dptr[disks - 2]; /* XOR parity */
88 q = dptr[disks - 1]; /* RS syndrome 124 q = dptr[disks - 1]; /* RS syndrome */
89 125
90 kernel_fpu_begin(&vxstate, KERNEL_VXR) 126 kernel_fpu_begin(&vxstate, KERNEL_VXR);
91 LOAD_CONST(); 127 LOAD_CONST();
92 128
93 for (d = 0; d < bytes; d += $#*NSIZE) 129 for (d = 0; d < bytes; d += $#*NSIZE) {
94 /* P/Q data pages */ 130 /* P/Q data pages */
95 LOAD_DATA(0,&dptr[z0][d]); 131 LOAD_DATA(0,&dptr[z0][d]);
96 COPY_VEC(8+$$,0+$$); 132 COPY_VEC(8+$$,0+$$);
97 for (z = z0 - 1; z >= start; z 133 for (z = z0 - 1; z >= start; z--) {
98 MASK(16+$$,8+$$); 134 MASK(16+$$,8+$$);
99 AND(16+$$,16+$$,25); 135 AND(16+$$,16+$$,25);
100 SHLBYTE(8+$$,8+$$); 136 SHLBYTE(8+$$,8+$$);
101 XOR(8+$$,8+$$,16+$$); 137 XOR(8+$$,8+$$,16+$$);
102 LOAD_DATA(16,&dptr[z][ 138 LOAD_DATA(16,&dptr[z][d]);
103 XOR(0+$$,0+$$,16+$$); 139 XOR(0+$$,0+$$,16+$$);
104 XOR(8+$$,8+$$,16+$$); 140 XOR(8+$$,8+$$,16+$$);
105 } 141 }
106 /* P/Q left side optimization 142 /* P/Q left side optimization */
107 for (z = start - 1; z >= 0; z- 143 for (z = start - 1; z >= 0; z--) {
108 MASK(16+$$,8+$$); 144 MASK(16+$$,8+$$);
109 AND(16+$$,16+$$,25); 145 AND(16+$$,16+$$,25);
110 SHLBYTE(8+$$,8+$$); 146 SHLBYTE(8+$$,8+$$);
111 XOR(8+$$,8+$$,16+$$); 147 XOR(8+$$,8+$$,16+$$);
112 } 148 }
113 LOAD_DATA(16,&p[d]); 149 LOAD_DATA(16,&p[d]);
114 XOR(16+$$,16+$$,0+$$); 150 XOR(16+$$,16+$$,0+$$);
115 STORE_DATA(16,&p[d]); 151 STORE_DATA(16,&p[d]);
116 LOAD_DATA(16,&q[d]); 152 LOAD_DATA(16,&q[d]);
117 XOR(16+$$,16+$$,8+$$); 153 XOR(16+$$,16+$$,8+$$);
118 STORE_DATA(16,&q[d]); 154 STORE_DATA(16,&q[d]);
119 } 155 }
120 kernel_fpu_end(&vxstate, KERNEL_VXR); 156 kernel_fpu_end(&vxstate, KERNEL_VXR);
121 } 157 }
122 158
123 static int raid6_s390vx$#_valid(void) 159 static int raid6_s390vx$#_valid(void)
124 { 160 {
125 return cpu_has_vx(); 161 return cpu_has_vx();
126 } 162 }
127 163
128 const struct raid6_calls raid6_s390vx$# = { 164 const struct raid6_calls raid6_s390vx$# = {
129 raid6_s390vx$#_gen_syndrome, 165 raid6_s390vx$#_gen_syndrome,
130 raid6_s390vx$#_xor_syndrome, 166 raid6_s390vx$#_xor_syndrome,
131 raid6_s390vx$#_valid, 167 raid6_s390vx$#_valid,
132 "vx128x$#", 168 "vx128x$#",
133 1 169 1
134 }; 170 };
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.