4 * $#-way unrolled RAID6 gen/xor functions for s390
5 * based on the vector facility
7 * Copyright IBM Corp. 2016
8 * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
10 * This file is postprocessed using unroll.awk.
13 #include <linux/raid/pq.h>
14 #include <asm/fpu/api.h>
16 asm(".include \"asm/vx-insn.h\"\n");
20 static inline void LOAD_CONST(void)
22 asm volatile("VREPIB %v24,7");
23 asm volatile("VREPIB %v25,0x1d");
27 * The SHLBYTE() operation shifts each of the 16 bytes in
28 * vector register y left by 1 bit and stores the result in
31 static inline void SHLBYTE(int x, int y)
33 asm volatile ("VAB %0,%1,%1" : : "i" (x), "i" (y));
37 * For each of the 16 bytes in the vector register y the MASK()
38 * operation returns 0xFF if the high bit of the byte is 1,
39 * or 0x00 if the high bit is 0. The result is stored in vector
42 static inline void MASK(int x, int y)
44 asm volatile ("VESRAVB %0,%1,24" : : "i" (x), "i" (y));
47 static inline void AND(int x, int y, int z)
49 asm volatile ("VN %0,%1,%2" : : "i" (x), "i" (y), "i" (z));
52 static inline void XOR(int x, int y, int z)
54 asm volatile ("VX %0,%1,%2" : : "i" (x), "i" (y), "i" (z));
57 static inline void LOAD_DATA(int x, int n, u8 *ptr)
59 typedef struct { u8 _[16*n]; } addrtype;
60 register addrtype *__ptr asm("1") = (addrtype *) ptr;
62 asm volatile ("VLM %2,%3,0,%r1"
63 : : "m" (*__ptr), "a" (__ptr), "i" (x), "i" (x + n - 1));
66 static inline void STORE_DATA(int x, int n, u8 *ptr)
68 typedef struct { u8 _[16*n]; } addrtype;
69 register addrtype *__ptr asm("1") = (addrtype *) ptr;
71 asm volatile ("VSTM %2,%3,0,1"
72 : "=m" (*__ptr) : "a" (__ptr), "i" (x), "i" (x + n - 1));
75 static inline void COPY_VEC(int x, int y)
77 asm volatile ("VLR %0,%1" : : "i" (x), "i" (y));
80 static void raid6_s390vx$#_gen_syndrome(int disks, size_t bytes, void **ptrs)
82 struct kernel_fpu vxstate;
86 kernel_fpu_begin(&vxstate, KERNEL_VXR);
90 z0 = disks - 3; /* Highest data disk */
91 p = dptr[z0 + 1]; /* XOR parity */
92 q = dptr[z0 + 2]; /* RS syndrome */
94 for (d = 0; d < bytes; d += $#*NSIZE) {
95 LOAD_DATA(0,$#,&dptr[z0][d]);
97 for (z = z0 - 1; z >= 0; z--) {
101 XOR(8+$$,8+$$,16+$$);
102 LOAD_DATA(16,$#,&dptr[z][d]);
103 XOR(0+$$,0+$$,16+$$);
104 XOR(8+$$,8+$$,16+$$);
106 STORE_DATA(0,$#,&p[d]);
107 STORE_DATA(8,$#,&q[d]);
109 kernel_fpu_end(&vxstate, KERNEL_VXR);
112 static void raid6_s390vx$#_xor_syndrome(int disks, int start, int stop,
113 size_t bytes, void **ptrs)
115 struct kernel_fpu vxstate;
120 z0 = stop; /* P/Q right side optimization */
121 p = dptr[disks - 2]; /* XOR parity */
122 q = dptr[disks - 1]; /* RS syndrome */
124 kernel_fpu_begin(&vxstate, KERNEL_VXR);
127 for (d = 0; d < bytes; d += $#*NSIZE) {
129 LOAD_DATA(0,$#,&dptr[z0][d]);
131 for (z = z0 - 1; z >= start; z--) {
135 XOR(8+$$,8+$$,16+$$);
136 LOAD_DATA(16,$#,&dptr[z][d]);
137 XOR(0+$$,0+$$,16+$$);
138 XOR(8+$$,8+$$,16+$$);
140 /* P/Q left side optimization */
141 for (z = start - 1; z >= 0; z--) {
145 XOR(8+$$,8+$$,16+$$);
147 LOAD_DATA(16,$#,&p[d]);
148 XOR(16+$$,16+$$,0+$$);
149 STORE_DATA(16,$#,&p[d]);
150 LOAD_DATA(16,$#,&q[d]);
151 XOR(16+$$,16+$$,8+$$);
152 STORE_DATA(16,$#,&q[d]);
154 kernel_fpu_end(&vxstate, KERNEL_VXR);
157 static int raid6_s390vx$#_valid(void)
159 return MACHINE_HAS_VX;
162 const struct raid6_calls raid6_s390vx$# = {
163 raid6_s390vx$#_gen_syndrome,
164 raid6_s390vx$#_xor_syndrome,
165 raid6_s390vx$#_valid,