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.h>
16 16
17 #define NSIZE 16 17 #define NSIZE 16
18 18
19 static __always_inline void LOAD_CONST(void) 19 static __always_inline void LOAD_CONST(void)
20 { 20 {
21 fpu_vrepib(24, 0x07); 21 fpu_vrepib(24, 0x07);
22 fpu_vrepib(25, 0x1d); 22 fpu_vrepib(25, 0x1d);
23 } 23 }
24 24
25 /* 25 /*
26 * The SHLBYTE() operation shifts each of the 26 * The SHLBYTE() operation shifts each of the 16 bytes in
27 * vector register y left by 1 bit and stores 27 * vector register y left by 1 bit and stores the result in
28 * vector register x. 28 * vector register x.
29 */ 29 */
30 #define SHLBYTE(x, y) fpu_vab(x, y, 30 #define SHLBYTE(x, y) fpu_vab(x, y, y)
31 31
32 /* 32 /*
33 * For each of the 16 bytes in the vector regi 33 * For each of the 16 bytes in the vector register y the MASK()
34 * operation returns 0xFF if the high bit of t 34 * operation returns 0xFF if the high bit of the byte is 1,
35 * or 0x00 if the high bit is 0. The result is 35 * or 0x00 if the high bit is 0. The result is stored in vector
36 * register x. 36 * register x.
37 */ 37 */
38 #define MASK(x, y) fpu_vesravb(x, 38 #define MASK(x, y) fpu_vesravb(x, y, 24)
39 39
40 #define AND(x, y, z) fpu_vn(x, y, z 40 #define AND(x, y, z) fpu_vn(x, y, z)
41 #define XOR(x, y, z) fpu_vx(x, y, z 41 #define XOR(x, y, z) fpu_vx(x, y, z)
42 #define LOAD_DATA(x, ptr) fpu_vlm(x, x + 42 #define LOAD_DATA(x, ptr) fpu_vlm(x, x + $# - 1, ptr)
43 #define STORE_DATA(x, ptr) fpu_vstm(x, x 43 #define STORE_DATA(x, ptr) fpu_vstm(x, x + $# - 1, ptr)
44 #define COPY_VEC(x, y) fpu_vlr(x, y) 44 #define COPY_VEC(x, y) fpu_vlr(x, y)
45 45
46 static void raid6_s390vx$#_gen_syndrome(int di 46 static void raid6_s390vx$#_gen_syndrome(int disks, size_t bytes, void **ptrs)
47 { 47 {
48 DECLARE_KERNEL_FPU_ONSTACK32(vxstate); 48 DECLARE_KERNEL_FPU_ONSTACK32(vxstate);
49 u8 **dptr, *p, *q; 49 u8 **dptr, *p, *q;
50 int d, z, z0; 50 int d, z, z0;
51 51
52 kernel_fpu_begin(&vxstate, KERNEL_VXR) 52 kernel_fpu_begin(&vxstate, KERNEL_VXR);
53 LOAD_CONST(); 53 LOAD_CONST();
54 54
55 dptr = (u8 **) ptrs; 55 dptr = (u8 **) ptrs;
56 z0 = disks - 3; /* Highest dat 56 z0 = disks - 3; /* Highest data disk */
57 p = dptr[z0 + 1]; /* XOR parity 57 p = dptr[z0 + 1]; /* XOR parity */
58 q = dptr[z0 + 2]; /* RS syndrome 58 q = dptr[z0 + 2]; /* RS syndrome */
59 59
60 for (d = 0; d < bytes; d += $#*NSIZE) 60 for (d = 0; d < bytes; d += $#*NSIZE) {
61 LOAD_DATA(0,&dptr[z0][d]); 61 LOAD_DATA(0,&dptr[z0][d]);
62 COPY_VEC(8+$$,0+$$); 62 COPY_VEC(8+$$,0+$$);
63 for (z = z0 - 1; z >= 0; z--) 63 for (z = z0 - 1; z >= 0; z--) {
64 MASK(16+$$,8+$$); 64 MASK(16+$$,8+$$);
65 AND(16+$$,16+$$,25); 65 AND(16+$$,16+$$,25);
66 SHLBYTE(8+$$,8+$$); 66 SHLBYTE(8+$$,8+$$);
67 XOR(8+$$,8+$$,16+$$); 67 XOR(8+$$,8+$$,16+$$);
68 LOAD_DATA(16,&dptr[z][ 68 LOAD_DATA(16,&dptr[z][d]);
69 XOR(0+$$,0+$$,16+$$); 69 XOR(0+$$,0+$$,16+$$);
70 XOR(8+$$,8+$$,16+$$); 70 XOR(8+$$,8+$$,16+$$);
71 } 71 }
72 STORE_DATA(0,&p[d]); 72 STORE_DATA(0,&p[d]);
73 STORE_DATA(8,&q[d]); 73 STORE_DATA(8,&q[d]);
74 } 74 }
75 kernel_fpu_end(&vxstate, KERNEL_VXR); 75 kernel_fpu_end(&vxstate, KERNEL_VXR);
76 } 76 }
77 77
78 static void raid6_s390vx$#_xor_syndrome(int di 78 static void raid6_s390vx$#_xor_syndrome(int disks, int start, int stop,
79 size_t 79 size_t bytes, void **ptrs)
80 { 80 {
81 DECLARE_KERNEL_FPU_ONSTACK32(vxstate); 81 DECLARE_KERNEL_FPU_ONSTACK32(vxstate);
82 u8 **dptr, *p, *q; 82 u8 **dptr, *p, *q;
83 int d, z, z0; 83 int d, z, z0;
84 84
85 dptr = (u8 **) ptrs; 85 dptr = (u8 **) ptrs;
86 z0 = stop; /* P/Q right s 86 z0 = stop; /* P/Q right side optimization */
87 p = dptr[disks - 2]; /* XOR parity 87 p = dptr[disks - 2]; /* XOR parity */
88 q = dptr[disks - 1]; /* RS syndrome 88 q = dptr[disks - 1]; /* RS syndrome */
89 89
90 kernel_fpu_begin(&vxstate, KERNEL_VXR) 90 kernel_fpu_begin(&vxstate, KERNEL_VXR);
91 LOAD_CONST(); 91 LOAD_CONST();
92 92
93 for (d = 0; d < bytes; d += $#*NSIZE) 93 for (d = 0; d < bytes; d += $#*NSIZE) {
94 /* P/Q data pages */ 94 /* P/Q data pages */
95 LOAD_DATA(0,&dptr[z0][d]); 95 LOAD_DATA(0,&dptr[z0][d]);
96 COPY_VEC(8+$$,0+$$); 96 COPY_VEC(8+$$,0+$$);
97 for (z = z0 - 1; z >= start; z 97 for (z = z0 - 1; z >= start; z--) {
98 MASK(16+$$,8+$$); 98 MASK(16+$$,8+$$);
99 AND(16+$$,16+$$,25); 99 AND(16+$$,16+$$,25);
100 SHLBYTE(8+$$,8+$$); 100 SHLBYTE(8+$$,8+$$);
101 XOR(8+$$,8+$$,16+$$); 101 XOR(8+$$,8+$$,16+$$);
102 LOAD_DATA(16,&dptr[z][ 102 LOAD_DATA(16,&dptr[z][d]);
103 XOR(0+$$,0+$$,16+$$); 103 XOR(0+$$,0+$$,16+$$);
104 XOR(8+$$,8+$$,16+$$); 104 XOR(8+$$,8+$$,16+$$);
105 } 105 }
106 /* P/Q left side optimization 106 /* P/Q left side optimization */
107 for (z = start - 1; z >= 0; z- 107 for (z = start - 1; z >= 0; z--) {
108 MASK(16+$$,8+$$); 108 MASK(16+$$,8+$$);
109 AND(16+$$,16+$$,25); 109 AND(16+$$,16+$$,25);
110 SHLBYTE(8+$$,8+$$); 110 SHLBYTE(8+$$,8+$$);
111 XOR(8+$$,8+$$,16+$$); 111 XOR(8+$$,8+$$,16+$$);
112 } 112 }
113 LOAD_DATA(16,&p[d]); 113 LOAD_DATA(16,&p[d]);
114 XOR(16+$$,16+$$,0+$$); 114 XOR(16+$$,16+$$,0+$$);
115 STORE_DATA(16,&p[d]); 115 STORE_DATA(16,&p[d]);
116 LOAD_DATA(16,&q[d]); 116 LOAD_DATA(16,&q[d]);
117 XOR(16+$$,16+$$,8+$$); 117 XOR(16+$$,16+$$,8+$$);
118 STORE_DATA(16,&q[d]); 118 STORE_DATA(16,&q[d]);
119 } 119 }
120 kernel_fpu_end(&vxstate, KERNEL_VXR); 120 kernel_fpu_end(&vxstate, KERNEL_VXR);
121 } 121 }
122 122
123 static int raid6_s390vx$#_valid(void) 123 static int raid6_s390vx$#_valid(void)
124 { 124 {
125 return cpu_has_vx(); 125 return cpu_has_vx();
126 } 126 }
127 127
128 const struct raid6_calls raid6_s390vx$# = { 128 const struct raid6_calls raid6_s390vx$# = {
129 raid6_s390vx$#_gen_syndrome, 129 raid6_s390vx$#_gen_syndrome,
130 raid6_s390vx$#_xor_syndrome, 130 raid6_s390vx$#_xor_syndrome,
131 raid6_s390vx$#_valid, 131 raid6_s390vx$#_valid,
132 "vx128x$#", 132 "vx128x$#",
133 1 133 1
134 }; 134 };
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.