Linux Audio

Check our new training course

Loading...
v6.13.7
  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/raid/pq.h>
 15#include <asm/fpu.h>
 
 
 16
 17#define NSIZE 16
 18
 19static __always_inline void LOAD_CONST(void)
 20{
 21	fpu_vrepib(24, 0x07);
 22	fpu_vrepib(25, 0x1d);
 23}
 24
 25/*
 26 * The SHLBYTE() operation shifts each of the 16 bytes in
 27 * vector register y left by 1 bit and stores the result in
 28 * vector register x.
 29 */
 30#define SHLBYTE(x, y)		fpu_vab(x, y, y)
 
 
 
 31
 32/*
 33 * For each of the 16 bytes in the vector register y the MASK()
 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 stored in vector
 36 * register x.
 37 */
 38#define MASK(x, y)		fpu_vesravb(x, y, 24)
 
 
 
 
 
 
 
 
 
 
 
 
 
 39
 40#define AND(x, y, z)		fpu_vn(x, y, z)
 41#define XOR(x, y, z)		fpu_vx(x, y, z)
 42#define LOAD_DATA(x, ptr)	fpu_vlm(x, x + $# - 1, ptr)
 43#define STORE_DATA(x, ptr)	fpu_vstm(x, x + $# - 1, ptr)
 44#define COPY_VEC(x, y)		fpu_vlr(x, y)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 45
 46static void raid6_s390vx$#_gen_syndrome(int disks, size_t bytes, void **ptrs)
 47{
 48	DECLARE_KERNEL_FPU_ONSTACK32(vxstate);
 49	u8 **dptr, *p, *q;
 50	int d, z, z0;
 51
 52	kernel_fpu_begin(&vxstate, KERNEL_VXR);
 53	LOAD_CONST();
 54
 55	dptr = (u8 **) ptrs;
 56	z0 = disks - 3;		/* Highest data disk */
 57	p = dptr[z0 + 1];	/* XOR parity */
 58	q = dptr[z0 + 2];	/* RS syndrome */
 59
 60	for (d = 0; d < bytes; d += $#*NSIZE) {
 61		LOAD_DATA(0,&dptr[z0][d]);
 62		COPY_VEC(8+$$,0+$$);
 63		for (z = z0 - 1; z >= 0; z--) {
 64			MASK(16+$$,8+$$);
 65			AND(16+$$,16+$$,25);
 66			SHLBYTE(8+$$,8+$$);
 67			XOR(8+$$,8+$$,16+$$);
 68			LOAD_DATA(16,&dptr[z][d]);
 69			XOR(0+$$,0+$$,16+$$);
 70			XOR(8+$$,8+$$,16+$$);
 71		}
 72		STORE_DATA(0,&p[d]);
 73		STORE_DATA(8,&q[d]);
 74	}
 75	kernel_fpu_end(&vxstate, KERNEL_VXR);
 76}
 77
 78static void raid6_s390vx$#_xor_syndrome(int disks, int start, int stop,
 79					size_t bytes, void **ptrs)
 80{
 81	DECLARE_KERNEL_FPU_ONSTACK32(vxstate);
 82	u8 **dptr, *p, *q;
 83	int d, z, z0;
 84
 85	dptr = (u8 **) ptrs;
 86	z0 = stop;		/* P/Q right side optimization */
 87	p = dptr[disks - 2];	/* XOR parity */
 88	q = dptr[disks - 1];	/* RS syndrome */
 89
 90	kernel_fpu_begin(&vxstate, KERNEL_VXR);
 91	LOAD_CONST();
 92
 93	for (d = 0; d < bytes; d += $#*NSIZE) {
 94		/* P/Q data pages */
 95		LOAD_DATA(0,&dptr[z0][d]);
 96		COPY_VEC(8+$$,0+$$);
 97		for (z = z0 - 1; z >= start; z--) {
 98			MASK(16+$$,8+$$);
 99			AND(16+$$,16+$$,25);
100			SHLBYTE(8+$$,8+$$);
101			XOR(8+$$,8+$$,16+$$);
102			LOAD_DATA(16,&dptr[z][d]);
103			XOR(0+$$,0+$$,16+$$);
104			XOR(8+$$,8+$$,16+$$);
105		}
106		/* P/Q left side optimization */
107		for (z = start - 1; z >= 0; z--) {
108			MASK(16+$$,8+$$);
109			AND(16+$$,16+$$,25);
110			SHLBYTE(8+$$,8+$$);
111			XOR(8+$$,8+$$,16+$$);
112		}
113		LOAD_DATA(16,&p[d]);
114		XOR(16+$$,16+$$,0+$$);
115		STORE_DATA(16,&p[d]);
116		LOAD_DATA(16,&q[d]);
117		XOR(16+$$,16+$$,8+$$);
118		STORE_DATA(16,&q[d]);
119	}
120	kernel_fpu_end(&vxstate, KERNEL_VXR);
121}
122
123static int raid6_s390vx$#_valid(void)
124{
125	return cpu_has_vx();
126}
127
128const struct raid6_calls raid6_s390vx$# = {
129	raid6_s390vx$#_gen_syndrome,
130	raid6_s390vx$#_xor_syndrome,
131	raid6_s390vx$#_valid,
132	"vx128x$#",
133	1
134};
v5.4
  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/raid/pq.h>
 15#include <asm/fpu/api.h>
 16
 17asm(".include \"asm/vx-insn.h\"\n");
 18
 19#define NSIZE 16
 20
 21static inline void LOAD_CONST(void)
 22{
 23	asm volatile("VREPIB %v24,7");
 24	asm volatile("VREPIB %v25,0x1d");
 25}
 26
 27/*
 28 * The SHLBYTE() operation shifts each of the 16 bytes in
 29 * vector register y left by 1 bit and stores the result in
 30 * vector register x.
 31 */
 32static inline void SHLBYTE(int x, int y)
 33{
 34	asm volatile ("VAB %0,%1,%1" : : "i" (x), "i" (y));
 35}
 36
 37/*
 38 * For each of the 16 bytes in the vector register y the MASK()
 39 * operation returns 0xFF if the high bit of the byte is 1,
 40 * or 0x00 if the high bit is 0. The result is stored in vector
 41 * register x.
 42 */
 43static inline void MASK(int x, int y)
 44{
 45	asm volatile ("VESRAVB	%0,%1,24" : : "i" (x), "i" (y));
 46}
 47
 48static inline void AND(int x, int y, int z)
 49{
 50	asm volatile ("VN %0,%1,%2" : : "i" (x), "i" (y), "i" (z));
 51}
 52
 53static inline void XOR(int x, int y, int z)
 54{
 55	asm volatile ("VX %0,%1,%2" : : "i" (x), "i" (y), "i" (z));
 56}
 57
 58static inline void LOAD_DATA(int x, u8 *ptr)
 59{
 60	typedef struct { u8 _[16 * $#]; } addrtype;
 61	register addrtype *__ptr asm("1") = (addrtype *) ptr;
 62
 63	asm volatile ("VLM %2,%3,0,%1"
 64		      : : "m" (*__ptr), "a" (__ptr), "i" (x),
 65			  "i" (x + $# - 1));
 66}
 67
 68static inline void STORE_DATA(int x, u8 *ptr)
 69{
 70	typedef struct { u8 _[16 * $#]; } addrtype;
 71	register addrtype *__ptr asm("1") = (addrtype *) ptr;
 72
 73	asm volatile ("VSTM %2,%3,0,1"
 74		      : "=m" (*__ptr) : "a" (__ptr), "i" (x),
 75			"i" (x + $# - 1));
 76}
 77
 78static inline void COPY_VEC(int x, int y)
 79{
 80	asm volatile ("VLR %0,%1" : : "i" (x), "i" (y));
 81}
 82
 83static void raid6_s390vx$#_gen_syndrome(int disks, size_t bytes, void **ptrs)
 84{
 85	struct kernel_fpu vxstate;
 86	u8 **dptr, *p, *q;
 87	int d, z, z0;
 88
 89	kernel_fpu_begin(&vxstate, KERNEL_VXR);
 90	LOAD_CONST();
 91
 92	dptr = (u8 **) ptrs;
 93	z0 = disks - 3;		/* Highest data disk */
 94	p = dptr[z0 + 1];	/* XOR parity */
 95	q = dptr[z0 + 2];	/* RS syndrome */
 96
 97	for (d = 0; d < bytes; d += $#*NSIZE) {
 98		LOAD_DATA(0,&dptr[z0][d]);
 99		COPY_VEC(8+$$,0+$$);
100		for (z = z0 - 1; z >= 0; z--) {
101			MASK(16+$$,8+$$);
102			AND(16+$$,16+$$,25);
103			SHLBYTE(8+$$,8+$$);
104			XOR(8+$$,8+$$,16+$$);
105			LOAD_DATA(16,&dptr[z][d]);
106			XOR(0+$$,0+$$,16+$$);
107			XOR(8+$$,8+$$,16+$$);
108		}
109		STORE_DATA(0,&p[d]);
110		STORE_DATA(8,&q[d]);
111	}
112	kernel_fpu_end(&vxstate, KERNEL_VXR);
113}
114
115static void raid6_s390vx$#_xor_syndrome(int disks, int start, int stop,
116					size_t bytes, void **ptrs)
117{
118	struct kernel_fpu vxstate;
119	u8 **dptr, *p, *q;
120	int d, z, z0;
121
122	dptr = (u8 **) ptrs;
123	z0 = stop;		/* P/Q right side optimization */
124	p = dptr[disks - 2];	/* XOR parity */
125	q = dptr[disks - 1];	/* RS syndrome */
126
127	kernel_fpu_begin(&vxstate, KERNEL_VXR);
128	LOAD_CONST();
129
130	for (d = 0; d < bytes; d += $#*NSIZE) {
131		/* P/Q data pages */
132		LOAD_DATA(0,&dptr[z0][d]);
133		COPY_VEC(8+$$,0+$$);
134		for (z = z0 - 1; z >= start; z--) {
135			MASK(16+$$,8+$$);
136			AND(16+$$,16+$$,25);
137			SHLBYTE(8+$$,8+$$);
138			XOR(8+$$,8+$$,16+$$);
139			LOAD_DATA(16,&dptr[z][d]);
140			XOR(0+$$,0+$$,16+$$);
141			XOR(8+$$,8+$$,16+$$);
142		}
143		/* P/Q left side optimization */
144		for (z = start - 1; z >= 0; z--) {
145			MASK(16+$$,8+$$);
146			AND(16+$$,16+$$,25);
147			SHLBYTE(8+$$,8+$$);
148			XOR(8+$$,8+$$,16+$$);
149		}
150		LOAD_DATA(16,&p[d]);
151		XOR(16+$$,16+$$,0+$$);
152		STORE_DATA(16,&p[d]);
153		LOAD_DATA(16,&q[d]);
154		XOR(16+$$,16+$$,8+$$);
155		STORE_DATA(16,&q[d]);
156	}
157	kernel_fpu_end(&vxstate, KERNEL_VXR);
158}
159
160static int raid6_s390vx$#_valid(void)
161{
162	return MACHINE_HAS_VX;
163}
164
165const struct raid6_calls raid6_s390vx$# = {
166	raid6_s390vx$#_gen_syndrome,
167	raid6_s390vx$#_xor_syndrome,
168	raid6_s390vx$#_valid,
169	"vx128x$#",
170	1
171};