Linux Audio

Check our new training course

Loading...
v4.10.11
 
  1/*
  2 * raid6_vx$#.c
  3 *
  4 * $#-way unrolled RAID6 gen/xor functions for s390
  5 * based on the vector facility
  6 *
  7 * Copyright IBM Corp. 2016
  8 * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
  9 *
 10 * This file is postprocessed using unroll.awk.
 11 */
 12
 13#include <linux/raid/pq.h>
 14#include <asm/fpu/api.h>
 15
 16asm(".include \"asm/vx-insn.h\"\n");
 17
 18#define NSIZE 16
 19
 20static inline void LOAD_CONST(void)
 21{
 22	asm volatile("VREPIB %v24,7");
 23	asm volatile("VREPIB %v25,0x1d");
 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 */
 31static inline void SHLBYTE(int x, int y)
 32{
 33	asm volatile ("VAB %0,%1,%1" : : "i" (x), "i" (y));
 34}
 35
 36/*
 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
 40 * register x.
 41 */
 42static inline void MASK(int x, int y)
 43{
 44	asm volatile ("VESRAVB	%0,%1,24" : : "i" (x), "i" (y));
 45}
 46
 47static 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
 52static inline void XOR(int x, int y, int z)
 53{
 54	asm volatile ("VX %0,%1,%2" : : "i" (x), "i" (y), "i" (z));
 55}
 56
 57static inline void LOAD_DATA(int x, int n, u8 *ptr)
 58{
 59	typedef struct { u8 _[16*n]; } addrtype;
 60	register addrtype *__ptr asm("1") = (addrtype *) ptr;
 61
 62	asm volatile ("VLM %2,%3,0,%r1"
 63		      : : "m" (*__ptr), "a" (__ptr), "i" (x), "i" (x + n - 1));
 64}
 65
 66static inline void STORE_DATA(int x, int n, u8 *ptr)
 67{
 68	typedef struct { u8 _[16*n]; } addrtype;
 69	register addrtype *__ptr asm("1") = (addrtype *) ptr;
 70
 71	asm volatile ("VSTM %2,%3,0,1"
 72		      : "=m" (*__ptr) : "a" (__ptr), "i" (x), "i" (x + n - 1));
 73}
 74
 75static inline void COPY_VEC(int x, int y)
 76{
 77	asm volatile ("VLR %0,%1" : : "i" (x), "i" (y));
 78}
 79
 80static void raid6_s390vx$#_gen_syndrome(int disks, size_t bytes, void **ptrs)
 81{
 82	struct kernel_fpu vxstate;
 83	u8 **dptr, *p, *q;
 84	int d, z, z0;
 85
 86	kernel_fpu_begin(&vxstate, KERNEL_VXR);
 87	LOAD_CONST();
 88
 89	dptr = (u8 **) ptrs;
 90	z0 = disks - 3;		/* Highest data disk */
 91	p = dptr[z0 + 1];	/* XOR parity */
 92	q = dptr[z0 + 2];	/* RS syndrome */
 93
 94	for (d = 0; d < bytes; d += $#*NSIZE) {
 95		LOAD_DATA(0,$#,&dptr[z0][d]);
 96		COPY_VEC(8+$$,0+$$);
 97		for (z = z0 - 1; z >= 0; 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+$$);
105		}
106		STORE_DATA(0,$#,&p[d]);
107		STORE_DATA(8,$#,&q[d]);
108	}
109	kernel_fpu_end(&vxstate, KERNEL_VXR);
110}
111
112static void raid6_s390vx$#_xor_syndrome(int disks, int start, int stop,
113					size_t bytes, void **ptrs)
114{
115	struct kernel_fpu vxstate;
116	u8 **dptr, *p, *q;
117	int d, z, z0;
118
119	dptr = (u8 **) ptrs;
120	z0 = stop;		/* P/Q right side optimization */
121	p = dptr[disks - 2];	/* XOR parity */
122	q = dptr[disks - 1];	/* RS syndrome */
123
124	kernel_fpu_begin(&vxstate, KERNEL_VXR);
125	LOAD_CONST();
126
127	for (d = 0; d < bytes; d += $#*NSIZE) {
128		/* P/Q data pages */
129		LOAD_DATA(0,$#,&dptr[z0][d]);
130		COPY_VEC(8+$$,0+$$);
131		for (z = z0 - 1; z >= start; z--) {
132			MASK(16+$$,8+$$);
133			AND(16+$$,16+$$,25);
134			SHLBYTE(8+$$,8+$$);
135			XOR(8+$$,8+$$,16+$$);
136			LOAD_DATA(16,$#,&dptr[z][d]);
137			XOR(0+$$,0+$$,16+$$);
138			XOR(8+$$,8+$$,16+$$);
139		}
140		/* P/Q left side optimization */
141		for (z = start - 1; z >= 0; z--) {
142			MASK(16+$$,8+$$);
143			AND(16+$$,16+$$,25);
144			SHLBYTE(8+$$,8+$$);
145			XOR(8+$$,8+$$,16+$$);
146		}
147		LOAD_DATA(16,$#,&p[d]);
148		XOR(16+$$,16+$$,0+$$);
149		STORE_DATA(16,$#,&p[d]);
150		LOAD_DATA(16,$#,&q[d]);
151		XOR(16+$$,16+$$,8+$$);
152		STORE_DATA(16,$#,&q[d]);
153	}
154	kernel_fpu_end(&vxstate, KERNEL_VXR);
155}
156
157static int raid6_s390vx$#_valid(void)
158{
159	return MACHINE_HAS_VX;
160}
161
162const struct raid6_calls raid6_s390vx$# = {
163	raid6_s390vx$#_gen_syndrome,
164	raid6_s390vx$#_xor_syndrome,
165	raid6_s390vx$#_valid,
166	"vx128x$#",
167	1
168};
v6.13.7
  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 * raid6_vx$#.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};