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 <linux/init.h>
16 #include <linux/module.h>
17 #include <linux/string.h>
18 #include <linux/kernel.h>
19 #include <linux/cpufeature.h>
20 #include <asm/switch_to.h>
21 
22 #define VMX_ALIGN		16
23 #define VMX_ALIGN_MASK		(VMX_ALIGN-1)
24 
25 #define VECTOR_BREAKPOINT	64
26 
27 u32 __crct10dif_vpmsum(u32 crc, unsigned char const *p, size_t len);
28 
29 static u16 crct10dif_vpmsum(u16 crci, unsigned char const *p, size_t len)
30 {
31 	unsigned int prealign;
32 	unsigned int tail;
33 	u32 crc = crci;
34 
35 	if (len < (VECTOR_BREAKPOINT + VMX_ALIGN) || in_interrupt())
36 		return crc_t10dif_generic(crc, p, len);
37 
38 	if ((unsigned long)p & VMX_ALIGN_MASK) {
39 		prealign = VMX_ALIGN - ((unsigned long)p & VMX_ALIGN_MASK);
40 		crc = crc_t10dif_generic(crc, p, prealign);
41 		len -= prealign;
42 		p += prealign;
43 	}
44 
45 	if (len & ~VMX_ALIGN_MASK) {
46 		crc <<= 16;
47 		preempt_disable();
48 		pagefault_disable();
49 		enable_kernel_altivec();
50 		crc = __crct10dif_vpmsum(crc, p, len & ~VMX_ALIGN_MASK);
51 		disable_kernel_altivec();
52 		pagefault_enable();
53 		preempt_enable();
54 		crc >>= 16;
55 	}
56 
57 	tail = len & VMX_ALIGN_MASK;
58 	if (tail) {
59 		p += len & ~VMX_ALIGN_MASK;
60 		crc = crc_t10dif_generic(crc, p, tail);
61 	}
62 
63 	return crc & 0xffff;
64 }
65 
66 static int crct10dif_vpmsum_init(struct shash_desc *desc)
67 {
68 	u16 *crc = shash_desc_ctx(desc);
69 
70 	*crc = 0;
71 	return 0;
72 }
73 
74 static int crct10dif_vpmsum_update(struct shash_desc *desc, const u8 *data,
75 			    unsigned int length)
76 {
77 	u16 *crc = shash_desc_ctx(desc);
78 
79 	*crc = crct10dif_vpmsum(*crc, data, length);
80 
81 	return 0;
82 }
83 
84 
85 static int crct10dif_vpmsum_final(struct shash_desc *desc, u8 *out)
86 {
87 	u16 *crcp = shash_desc_ctx(desc);
88 
89 	*(u16 *)out = *crcp;
90 	return 0;
91 }
92 
93 static struct shash_alg alg = {
94 	.init		= crct10dif_vpmsum_init,
95 	.update		= crct10dif_vpmsum_update,
96 	.final		= crct10dif_vpmsum_final,
97 	.descsize	= CRC_T10DIF_DIGEST_SIZE,
98 	.digestsize	= CRC_T10DIF_DIGEST_SIZE,
99 	.base		= {
100 		.cra_name		= "crct10dif",
101 		.cra_driver_name	= "crct10dif-vpmsum",
102 		.cra_priority		= 200,
103 		.cra_blocksize		= CRC_T10DIF_BLOCK_SIZE,
104 		.cra_module		= THIS_MODULE,
105 	}
106 };
107 
108 static int __init crct10dif_vpmsum_mod_init(void)
109 {
110 	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
111 		return -ENODEV;
112 
113 	return crypto_register_shash(&alg);
114 }
115 
116 static void __exit crct10dif_vpmsum_mod_fini(void)
117 {
118 	crypto_unregister_shash(&alg);
119 }
120 
121 module_cpu_feature_match(PPC_MODULE_FEATURE_VEC_CRYPTO, crct10dif_vpmsum_mod_init);
122 module_exit(crct10dif_vpmsum_mod_fini);
123 
124 MODULE_AUTHOR("Daniel Axtens <dja@axtens.net>");
125 MODULE_DESCRIPTION("CRCT10DIF using vector polynomial multiply-sum instructions");
126 MODULE_LICENSE("GPL");
127 MODULE_ALIAS_CRYPTO("crct10dif");
128 MODULE_ALIAS_CRYPTO("crct10dif-vpmsum");
129