1 /*
2  * Calculate a CRC T10-DIF with vpmsum acceleration
3  *
4  * Copyright 2017, Daniel Axtens, IBM Corporation.
5  * [based on crc32c-vpmsum_glue.c]
6  *
7  * This program is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU General Public License
9  * as published by the Free Software Foundation; either version 2
10  * of the License, or (at your option) any later version.
11  */
12 
13 #include <linux/crc-t10dif.h>
14 #include <crypto/internal/hash.h>
15 #include <crypto/internal/simd.h>
16 #include <linux/init.h>
17 #include <linux/module.h>
18 #include <linux/string.h>
19 #include <linux/kernel.h>
20 #include <linux/cpufeature.h>
21 #include <asm/simd.h>
22 #include <asm/switch_to.h>
23 
24 #define VMX_ALIGN		16
25 #define VMX_ALIGN_MASK		(VMX_ALIGN-1)
26 
27 #define VECTOR_BREAKPOINT	64
28 
29 u32 __crct10dif_vpmsum(u32 crc, unsigned char const *p, size_t len);
30 
31 static u16 crct10dif_vpmsum(u16 crci, unsigned char const *p, size_t len)
32 {
33 	unsigned int prealign;
34 	unsigned int tail;
35 	u32 crc = crci;
36 
37 	if (len < (VECTOR_BREAKPOINT + VMX_ALIGN) || !crypto_simd_usable())
38 		return crc_t10dif_generic(crc, p, len);
39 
40 	if ((unsigned long)p & VMX_ALIGN_MASK) {
41 		prealign = VMX_ALIGN - ((unsigned long)p & VMX_ALIGN_MASK);
42 		crc = crc_t10dif_generic(crc, p, prealign);
43 		len -= prealign;
44 		p += prealign;
45 	}
46 
47 	if (len & ~VMX_ALIGN_MASK) {
48 		crc <<= 16;
49 		preempt_disable();
50 		pagefault_disable();
51 		enable_kernel_altivec();
52 		crc = __crct10dif_vpmsum(crc, p, len & ~VMX_ALIGN_MASK);
53 		disable_kernel_altivec();
54 		pagefault_enable();
55 		preempt_enable();
56 		crc >>= 16;
57 	}
58 
59 	tail = len & VMX_ALIGN_MASK;
60 	if (tail) {
61 		p += len & ~VMX_ALIGN_MASK;
62 		crc = crc_t10dif_generic(crc, p, tail);
63 	}
64 
65 	return crc & 0xffff;
66 }
67 
68 static int crct10dif_vpmsum_init(struct shash_desc *desc)
69 {
70 	u16 *crc = shash_desc_ctx(desc);
71 
72 	*crc = 0;
73 	return 0;
74 }
75 
76 static int crct10dif_vpmsum_update(struct shash_desc *desc, const u8 *data,
77 			    unsigned int length)
78 {
79 	u16 *crc = shash_desc_ctx(desc);
80 
81 	*crc = crct10dif_vpmsum(*crc, data, length);
82 
83 	return 0;
84 }
85 
86 
87 static int crct10dif_vpmsum_final(struct shash_desc *desc, u8 *out)
88 {
89 	u16 *crcp = shash_desc_ctx(desc);
90 
91 	*(u16 *)out = *crcp;
92 	return 0;
93 }
94 
95 static struct shash_alg alg = {
96 	.init		= crct10dif_vpmsum_init,
97 	.update		= crct10dif_vpmsum_update,
98 	.final		= crct10dif_vpmsum_final,
99 	.descsize	= CRC_T10DIF_DIGEST_SIZE,
100 	.digestsize	= CRC_T10DIF_DIGEST_SIZE,
101 	.base		= {
102 		.cra_name		= "crct10dif",
103 		.cra_driver_name	= "crct10dif-vpmsum",
104 		.cra_priority		= 200,
105 		.cra_blocksize		= CRC_T10DIF_BLOCK_SIZE,
106 		.cra_module		= THIS_MODULE,
107 	}
108 };
109 
110 static int __init crct10dif_vpmsum_mod_init(void)
111 {
112 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
113 		return -ENODEV;
114 
115 	return crypto_register_shash(&alg);
116 }
117 
118 static void __exit crct10dif_vpmsum_mod_fini(void)
119 {
120 	crypto_unregister_shash(&alg);
121 }
122 
123 module_cpu_feature_match(PPC_MODULE_FEATURE_VEC_CRYPTO, crct10dif_vpmsum_mod_init);
124 module_exit(crct10dif_vpmsum_mod_fini);
125 
126 MODULE_AUTHOR("Daniel Axtens <dja@axtens.net>");
127 MODULE_DESCRIPTION("CRCT10DIF using vector polynomial multiply-sum instructions");
128 MODULE_LICENSE("GPL");
129 MODULE_ALIAS_CRYPTO("crct10dif");
130 MODULE_ALIAS_CRYPTO("crct10dif-vpmsum");
131