include/asm/xor.h

0001 /* SPDX-License-Identifier: GPL-2.0-only */
0002 /*
0003  *  arch/arm/include/asm/xor.h
0004  *
0005  *  Copyright (C) 2001 Russell King
0006  */
0007 #include <linux/hardirq.h>
0008 #include <asm-generic/xor.h>
0009 #include <asm/hwcap.h>
0010 #include <asm/neon.h>
0011
0012 #define __XOR(a1, a2) a1 ^= a2
0013
0014 #define GET_BLOCK_2(dst) \
0015     __asm__("ldmia  %0, {%1, %2}" \
0016         : "=r" (dst), "=r" (a1), "=r" (a2) \
0017         : "0" (dst))
0018
0019 #define GET_BLOCK_4(dst) \
0020     __asm__("ldmia  %0, {%1, %2, %3, %4}" \
0021         : "=r" (dst), "=r" (a1), "=r" (a2), "=r" (a3), "=r" (a4) \
0022         : "0" (dst))
0023
0024 #define XOR_BLOCK_2(src) \
0025     __asm__("ldmia  %0!, {%1, %2}" \
0026         : "=r" (src), "=r" (b1), "=r" (b2) \
0027         : "0" (src)); \
0028     __XOR(a1, b1); __XOR(a2, b2);
0029
0030 #define XOR_BLOCK_4(src) \
0031     __asm__("ldmia  %0!, {%1, %2, %3, %4}" \
0032         : "=r" (src), "=r" (b1), "=r" (b2), "=r" (b3), "=r" (b4) \
0033         : "0" (src)); \
0034     __XOR(a1, b1); __XOR(a2, b2); __XOR(a3, b3); __XOR(a4, b4)
0035
0036 #define PUT_BLOCK_2(dst) \
0037     __asm__ __volatile__("stmia %0!, {%2, %3}" \
0038         : "=r" (dst) \
0039         : "0" (dst), "r" (a1), "r" (a2))
0040
0041 #define PUT_BLOCK_4(dst) \
0042     __asm__ __volatile__("stmia %0!, {%2, %3, %4, %5}" \
0043         : "=r" (dst) \
0044         : "0" (dst), "r" (a1), "r" (a2), "r" (a3), "r" (a4))
0045
0046 static void
0047 xor_arm4regs_2(unsigned long bytes, unsigned long * __restrict p1,
0048            const unsigned long * __restrict p2)
0049 {
0050     unsigned int lines = bytes / sizeof(unsigned long) / 4;
0051     register unsigned int a1 __asm__("r4");
0052     register unsigned int a2 __asm__("r5");
0053     register unsigned int a3 __asm__("r6");
0054     register unsigned int a4 __asm__("r7");
0055     register unsigned int b1 __asm__("r8");
0056     register unsigned int b2 __asm__("r9");
0057     register unsigned int b3 __asm__("ip");
0058     register unsigned int b4 __asm__("lr");
0059
0060     do {
0061         GET_BLOCK_4(p1);
0062         XOR_BLOCK_4(p2);
0063         PUT_BLOCK_4(p1);
0064     } while (--lines);
0065 }
0066
0067 static void
0068 xor_arm4regs_3(unsigned long bytes, unsigned long * __restrict p1,
0069            const unsigned long * __restrict p2,
0070            const unsigned long * __restrict p3)
0071 {
0072     unsigned int lines = bytes / sizeof(unsigned long) / 4;
0073     register unsigned int a1 __asm__("r4");
0074     register unsigned int a2 __asm__("r5");
0075     register unsigned int a3 __asm__("r6");
0076     register unsigned int a4 __asm__("r7");
0077     register unsigned int b1 __asm__("r8");
0078     register unsigned int b2 __asm__("r9");
0079     register unsigned int b3 __asm__("ip");
0080     register unsigned int b4 __asm__("lr");
0081
0082     do {
0083         GET_BLOCK_4(p1);
0084         XOR_BLOCK_4(p2);
0085         XOR_BLOCK_4(p3);
0086         PUT_BLOCK_4(p1);
0087     } while (--lines);
0088 }
0089
0090 static void
0091 xor_arm4regs_4(unsigned long bytes, unsigned long * __restrict p1,
0092            const unsigned long * __restrict p2,
0093            const unsigned long * __restrict p3,
0094            const unsigned long * __restrict p4)
0095 {
0096     unsigned int lines = bytes / sizeof(unsigned long) / 2;
0097     register unsigned int a1 __asm__("r8");
0098     register unsigned int a2 __asm__("r9");
0099     register unsigned int b1 __asm__("ip");
0100     register unsigned int b2 __asm__("lr");
0101
0102     do {
0103         GET_BLOCK_2(p1);
0104         XOR_BLOCK_2(p2);
0105         XOR_BLOCK_2(p3);
0106         XOR_BLOCK_2(p4);
0107         PUT_BLOCK_2(p1);
0108     } while (--lines);
0109 }
0110
0111 static void
0112 xor_arm4regs_5(unsigned long bytes, unsigned long * __restrict p1,
0113            const unsigned long * __restrict p2,
0114            const unsigned long * __restrict p3,
0115            const unsigned long * __restrict p4,
0116            const unsigned long * __restrict p5)
0117 {
0118     unsigned int lines = bytes / sizeof(unsigned long) / 2;
0119     register unsigned int a1 __asm__("r8");
0120     register unsigned int a2 __asm__("r9");
0121     register unsigned int b1 __asm__("ip");
0122     register unsigned int b2 __asm__("lr");
0123
0124     do {
0125         GET_BLOCK_2(p1);
0126         XOR_BLOCK_2(p2);
0127         XOR_BLOCK_2(p3);
0128         XOR_BLOCK_2(p4);
0129         XOR_BLOCK_2(p5);
0130         PUT_BLOCK_2(p1);
0131     } while (--lines);
0132 }
0133
0134 static struct xor_block_template xor_block_arm4regs = {
0135     .name   = "arm4regs",
0136     .do_2   = xor_arm4regs_2,
0137     .do_3   = xor_arm4regs_3,
0138     .do_4   = xor_arm4regs_4,
0139     .do_5   = xor_arm4regs_5,
0140 };
0141
0142 #undef XOR_TRY_TEMPLATES
0143 #define XOR_TRY_TEMPLATES           \
0144     do {                    \
0145         xor_speed(&xor_block_arm4regs); \
0146         xor_speed(&xor_block_8regs);    \
0147         xor_speed(&xor_block_32regs);   \
0148         NEON_TEMPLATES;         \
0149     } while (0)
0150
0151 #ifdef CONFIG_KERNEL_MODE_NEON
0152
0153 extern struct xor_block_template const xor_block_neon_inner;
0154
0155 static void
0156 xor_neon_2(unsigned long bytes, unsigned long * __restrict p1,
0157        const unsigned long * __restrict p2)
0158 {
0159     if (in_interrupt()) {
0160         xor_arm4regs_2(bytes, p1, p2);
0161     } else {
0162         kernel_neon_begin();
0163         xor_block_neon_inner.do_2(bytes, p1, p2);
0164         kernel_neon_end();
0165     }
0166 }
0167
0168 static void
0169 xor_neon_3(unsigned long bytes, unsigned long * __restrict p1,
0170        const unsigned long * __restrict p2,
0171        const unsigned long * __restrict p3)
0172 {
0173     if (in_interrupt()) {
0174         xor_arm4regs_3(bytes, p1, p2, p3);
0175     } else {
0176         kernel_neon_begin();
0177         xor_block_neon_inner.do_3(bytes, p1, p2, p3);
0178         kernel_neon_end();
0179     }
0180 }
0181
0182 static void
0183 xor_neon_4(unsigned long bytes, unsigned long * __restrict p1,
0184        const unsigned long * __restrict p2,
0185        const unsigned long * __restrict p3,
0186        const unsigned long * __restrict p4)
0187 {
0188     if (in_interrupt()) {
0189         xor_arm4regs_4(bytes, p1, p2, p3, p4);
0190     } else {
0191         kernel_neon_begin();
0192         xor_block_neon_inner.do_4(bytes, p1, p2, p3, p4);
0193         kernel_neon_end();
0194     }
0195 }
0196
0197 static void
0198 xor_neon_5(unsigned long bytes, unsigned long * __restrict p1,
0199        const unsigned long * __restrict p2,
0200        const unsigned long * __restrict p3,
0201        const unsigned long * __restrict p4,
0202        const unsigned long * __restrict p5)
0203 {
0204     if (in_interrupt()) {
0205         xor_arm4regs_5(bytes, p1, p2, p3, p4, p5);
0206     } else {
0207         kernel_neon_begin();
0208         xor_block_neon_inner.do_5(bytes, p1, p2, p3, p4, p5);
0209         kernel_neon_end();
0210     }
0211 }
0212
0213 static struct xor_block_template xor_block_neon = {
0214     .name   = "neon",
0215     .do_2   = xor_neon_2,
0216     .do_3   = xor_neon_3,
0217     .do_4   = xor_neon_4,
0218     .do_5   = xor_neon_5
0219 };
0220
0221 #define NEON_TEMPLATES  \
0222     do { if (cpu_has_neon()) xor_speed(&xor_block_neon); } while (0)
0223 #else
0224 #define NEON_TEMPLATES
0225 #endif