1// SPDX-License-Identifier: GPL-2.0
2/*
3 * raid6_vx$#.c
4 *
5 * $#-way unrolled RAID6 gen/xor functions for s390
6 * based on the vector facility
7 *
8 * Copyright IBM Corp. 2016
9 * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
10 *
11 * This file is postprocessed using unroll.awk.
12 */
13
14#include <linux/cpufeature.h>
15#include <linux/raid/pq.h>
16#include <asm/fpu.h>
17
18#define NSIZE 16
19
20static __always_inline void LOAD_CONST(void)
21{
22	fpu_vrepib(24, 0x07);
23	fpu_vrepib(25, 0x1d);
24}
25
26/*
27 * The SHLBYTE() operation shifts each of the 16 bytes in
28 * vector register y left by 1 bit and stores the result in
29 * vector register x.
30 */
31#define SHLBYTE(x, y)		fpu_vab(x, y, y)
32
33/*
34 * For each of the 16 bytes in the vector register y the MASK()
35 * operation returns 0xFF if the high bit of the byte is 1,
36 * or 0x00 if the high bit is 0. The result is stored in vector
37 * register x.
38 */
39#define MASK(x, y)		fpu_vesravb(x, y, 24)
40
41#define AND(x, y, z)		fpu_vn(x, y, z)
42#define XOR(x, y, z)		fpu_vx(x, y, z)
43#define LOAD_DATA(x, ptr)	fpu_vlm(x, x + $# - 1, ptr)
44#define STORE_DATA(x, ptr)	fpu_vstm(x, x + $# - 1, ptr)
45#define COPY_VEC(x, y)		fpu_vlr(x, y)
46
47static void raid6_s390vx$#_gen_syndrome(int disks, size_t bytes, void **ptrs)
48{
49	DECLARE_KERNEL_FPU_ONSTACK32(vxstate);
50	u8 **dptr, *p, *q;
51	int d, z, z0;
52
53	kernel_fpu_begin(&vxstate, KERNEL_VXR);
54	LOAD_CONST();
55
56	dptr = (u8 **) ptrs;
57	z0 = disks - 3;		/* Highest data disk */
58	p = dptr[z0 + 1];	/* XOR parity */
59	q = dptr[z0 + 2];	/* RS syndrome */
60
61	for (d = 0; d < bytes; d += $#*NSIZE) {
62		LOAD_DATA(0,&dptr[z0][d]);
63		COPY_VEC(8+$$,0+$$);
64		for (z = z0 - 1; z >= 0; z--) {
65			MASK(16+$$,8+$$);
66			AND(16+$$,16+$$,25);
67			SHLBYTE(8+$$,8+$$);
68			XOR(8+$$,8+$$,16+$$);
69			LOAD_DATA(16,&dptr[z][d]);
70			XOR(0+$$,0+$$,16+$$);
71			XOR(8+$$,8+$$,16+$$);
72		}
73		STORE_DATA(0,&p[d]);
74		STORE_DATA(8,&q[d]);
75	}
76	kernel_fpu_end(&vxstate, KERNEL_VXR);
77}
78
79static void raid6_s390vx$#_xor_syndrome(int disks, int start, int stop,
80					size_t bytes, void **ptrs)
81{
82	DECLARE_KERNEL_FPU_ONSTACK32(vxstate);
83	u8 **dptr, *p, *q;
84	int d, z, z0;
85
86	dptr = (u8 **) ptrs;
87	z0 = stop;		/* P/Q right side optimization */
88	p = dptr[disks - 2];	/* XOR parity */
89	q = dptr[disks - 1];	/* RS syndrome */
90
91	kernel_fpu_begin(&vxstate, KERNEL_VXR);
92	LOAD_CONST();
93
94	for (d = 0; d < bytes; d += $#*NSIZE) {
95		/* P/Q data pages */
96		LOAD_DATA(0,&dptr[z0][d]);
97		COPY_VEC(8+$$,0+$$);
98		for (z = z0 - 1; z >= start; z--) {
99			MASK(16+$$,8+$$);
100			AND(16+$$,16+$$,25);
101			SHLBYTE(8+$$,8+$$);
102			XOR(8+$$,8+$$,16+$$);
103			LOAD_DATA(16,&dptr[z][d]);
104			XOR(0+$$,0+$$,16+$$);
105			XOR(8+$$,8+$$,16+$$);
106		}
107		/* P/Q left side optimization */
108		for (z = start - 1; z >= 0; z--) {
109			MASK(16+$$,8+$$);
110			AND(16+$$,16+$$,25);
111			SHLBYTE(8+$$,8+$$);
112			XOR(8+$$,8+$$,16+$$);
113		}
114		LOAD_DATA(16,&p[d]);
115		XOR(16+$$,16+$$,0+$$);
116		STORE_DATA(16,&p[d]);
117		LOAD_DATA(16,&q[d]);
118		XOR(16+$$,16+$$,8+$$);
119		STORE_DATA(16,&q[d]);
120	}
121	kernel_fpu_end(&vxstate, KERNEL_VXR);
122}
123
124static int raid6_s390vx$#_valid(void)
125{
126	return cpu_has_vx();
127}
128
129const struct raid6_calls raid6_s390vx$# = {
130	raid6_s390vx$#_gen_syndrome,
131	raid6_s390vx$#_xor_syndrome,
132	raid6_s390vx$#_valid,
133	"vx128x$#",
134	1
135};
136