[mirror_ubuntu-bionic-kernel.git] / arch / x86 / crypto / camellia_aesni_avx_glue.c

/*
 * Glue Code for x86_64/AVX/AES-NI assembler optimized version of Camellia
 *
 * Copyright © 2012 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 */

#include <linux/module.h>
#include <linux/types.h>
#include <linux/crypto.h>
#include <linux/err.h>
#include <crypto/algapi.h>
#include <crypto/ctr.h>
#include <crypto/lrw.h>
#include <crypto/xts.h>
#include <asm/xcr.h>
#include <asm/xsave.h>
#include <asm/crypto/camellia.h>
#include <asm/crypto/ablk_helper.h>
#include <asm/crypto/glue_helper.h>

#define CAMELLIA_AESNI_PARALLEL_BLOCKS 16

/* 16-way AES-NI parallel cipher functions */
asmlinkage void camellia_ecb_enc_16way(struct camellia_ctx *ctx, u8 *dst,
				       const u8 *src);
asmlinkage void camellia_ecb_dec_16way(struct camellia_ctx *ctx, u8 *dst,
				       const u8 *src);

asmlinkage void camellia_cbc_dec_16way(struct camellia_ctx *ctx, u8 *dst,
				       const u8 *src);
asmlinkage void camellia_ctr_16way(struct camellia_ctx *ctx, u8 *dst,
				   const u8 *src, le128 *iv);

static const struct common_glue_ctx camellia_enc = {
	.num_funcs = 3,
	.fpu_blocks_limit = CAMELLIA_AESNI_PARALLEL_BLOCKS,

	.funcs = { {
		.num_blocks = CAMELLIA_AESNI_PARALLEL_BLOCKS,
		.fn_u = { .ecb = GLUE_FUNC_CAST(camellia_ecb_enc_16way) }
	}, {
		.num_blocks = 2,
		.fn_u = { .ecb = GLUE_FUNC_CAST(camellia_enc_blk_2way) }
	}, {
		.num_blocks = 1,
		.fn_u = { .ecb = GLUE_FUNC_CAST(camellia_enc_blk) }
	} }
};

static const struct common_glue_ctx camellia_ctr = {
	.num_funcs = 3,
	.fpu_blocks_limit = CAMELLIA_AESNI_PARALLEL_BLOCKS,

	.funcs = { {
		.num_blocks = CAMELLIA_AESNI_PARALLEL_BLOCKS,
		.fn_u = { .ctr = GLUE_CTR_FUNC_CAST(camellia_ctr_16way) }
	}, {
		.num_blocks = 2,
		.fn_u = { .ctr = GLUE_CTR_FUNC_CAST(camellia_crypt_ctr_2way) }
	}, {
		.num_blocks = 1,
		.fn_u = { .ctr = GLUE_CTR_FUNC_CAST(camellia_crypt_ctr) }
	} }
};

static const struct common_glue_ctx camellia_dec = {
	.num_funcs = 3,
	.fpu_blocks_limit = CAMELLIA_AESNI_PARALLEL_BLOCKS,

	.funcs = { {
		.num_blocks = CAMELLIA_AESNI_PARALLEL_BLOCKS,
		.fn_u = { .ecb = GLUE_FUNC_CAST(camellia_ecb_dec_16way) }
	}, {
		.num_blocks = 2,
		.fn_u = { .ecb = GLUE_FUNC_CAST(camellia_dec_blk_2way) }
	}, {
		.num_blocks = 1,
		.fn_u = { .ecb = GLUE_FUNC_CAST(camellia_dec_blk) }
	} }
};

static const struct common_glue_ctx camellia_dec_cbc = {
	.num_funcs = 3,
	.fpu_blocks_limit = CAMELLIA_AESNI_PARALLEL_BLOCKS,

	.funcs = { {
		.num_blocks = CAMELLIA_AESNI_PARALLEL_BLOCKS,
		.fn_u = { .cbc = GLUE_CBC_FUNC_CAST(camellia_cbc_dec_16way) }
	}, {
		.num_blocks = 2,
		.fn_u = { .cbc = GLUE_CBC_FUNC_CAST(camellia_decrypt_cbc_2way) }
	}, {
		.num_blocks = 1,
		.fn_u = { .cbc = GLUE_CBC_FUNC_CAST(camellia_dec_blk) }
	} }
};

static int ecb_encrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
		       struct scatterlist *src, unsigned int nbytes)
{
	return glue_ecb_crypt_128bit(&camellia_enc, desc, dst, src, nbytes);
}

static int ecb_decrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
		       struct scatterlist *src, unsigned int nbytes)
{
	return glue_ecb_crypt_128bit(&camellia_dec, desc, dst, src, nbytes);
}

static int cbc_encrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
		       struct scatterlist *src, unsigned int nbytes)
{
	return glue_cbc_encrypt_128bit(GLUE_FUNC_CAST(camellia_enc_blk), desc,
				       dst, src, nbytes);
}

static int cbc_decrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
		       struct scatterlist *src, unsigned int nbytes)
{
	return glue_cbc_decrypt_128bit(&camellia_dec_cbc, desc, dst, src,
				       nbytes);
}

static int ctr_crypt(struct blkcipher_desc *desc, struct scatterlist *dst,
		     struct scatterlist *src, unsigned int nbytes)
{
	return glue_ctr_crypt_128bit(&camellia_ctr, desc, dst, src, nbytes);
}

static inline bool camellia_fpu_begin(bool fpu_enabled, unsigned int nbytes)
{
	return glue_fpu_begin(CAMELLIA_BLOCK_SIZE,
			      CAMELLIA_AESNI_PARALLEL_BLOCKS, NULL, fpu_enabled,
			      nbytes);
}

static inline void camellia_fpu_end(bool fpu_enabled)
{
	glue_fpu_end(fpu_enabled);
}

static int camellia_setkey(struct crypto_tfm *tfm, const u8 *in_key,
			   unsigned int key_len)
{
	return __camellia_setkey(crypto_tfm_ctx(tfm), in_key, key_len,
				 &tfm->crt_flags);
}

struct crypt_priv {
	struct camellia_ctx *ctx;
	bool fpu_enabled;
};

static void encrypt_callback(void *priv, u8 *srcdst, unsigned int nbytes)
{
	const unsigned int bsize = CAMELLIA_BLOCK_SIZE;
	struct crypt_priv *ctx = priv;
	int i;

	ctx->fpu_enabled = camellia_fpu_begin(ctx->fpu_enabled, nbytes);

	if (nbytes >= CAMELLIA_AESNI_PARALLEL_BLOCKS * bsize) {
		camellia_ecb_enc_16way(ctx->ctx, srcdst, srcdst);
		srcdst += bsize * CAMELLIA_AESNI_PARALLEL_BLOCKS;
		nbytes -= bsize * CAMELLIA_AESNI_PARALLEL_BLOCKS;
	}

	while (nbytes >= CAMELLIA_PARALLEL_BLOCKS * bsize) {
		camellia_enc_blk_2way(ctx->ctx, srcdst, srcdst);
		srcdst += bsize * CAMELLIA_PARALLEL_BLOCKS;
		nbytes -= bsize * CAMELLIA_PARALLEL_BLOCKS;
	}

	for (i = 0; i < nbytes / bsize; i++, srcdst += bsize)
		camellia_enc_blk(ctx->ctx, srcdst, srcdst);
}

static void decrypt_callback(void *priv, u8 *srcdst, unsigned int nbytes)
{
	const unsigned int bsize = CAMELLIA_BLOCK_SIZE;
	struct crypt_priv *ctx = priv;
	int i;

	ctx->fpu_enabled = camellia_fpu_begin(ctx->fpu_enabled, nbytes);

	if (nbytes >= CAMELLIA_AESNI_PARALLEL_BLOCKS * bsize) {
		camellia_ecb_dec_16way(ctx->ctx, srcdst, srcdst);
		srcdst += bsize * CAMELLIA_AESNI_PARALLEL_BLOCKS;
		nbytes -= bsize * CAMELLIA_AESNI_PARALLEL_BLOCKS;
	}

	while (nbytes >= CAMELLIA_PARALLEL_BLOCKS * bsize) {
		camellia_dec_blk_2way(ctx->ctx, srcdst, srcdst);
		srcdst += bsize * CAMELLIA_PARALLEL_BLOCKS;
		nbytes -= bsize * CAMELLIA_PARALLEL_BLOCKS;
	}

	for (i = 0; i < nbytes / bsize; i++, srcdst += bsize)
		camellia_dec_blk(ctx->ctx, srcdst, srcdst);
}

static int lrw_encrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
		       struct scatterlist *src, unsigned int nbytes)
{
	struct camellia_lrw_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
	be128 buf[CAMELLIA_AESNI_PARALLEL_BLOCKS];
	struct crypt_priv crypt_ctx = {
		.ctx = &ctx->camellia_ctx,
		.fpu_enabled = false,
	};
	struct lrw_crypt_req req = {
		.tbuf = buf,
		.tbuflen = sizeof(buf),

		.table_ctx = &ctx->lrw_table,
		.crypt_ctx = &crypt_ctx,
		.crypt_fn = encrypt_callback,
	};
	int ret;

	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
	ret = lrw_crypt(desc, dst, src, nbytes, &req);
	camellia_fpu_end(crypt_ctx.fpu_enabled);

	return ret;
}

static int lrw_decrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
		       struct scatterlist *src, unsigned int nbytes)
{
	struct camellia_lrw_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
	be128 buf[CAMELLIA_AESNI_PARALLEL_BLOCKS];
	struct crypt_priv crypt_ctx = {
		.ctx = &ctx->camellia_ctx,
		.fpu_enabled = false,
	};
	struct lrw_crypt_req req = {
		.tbuf = buf,
		.tbuflen = sizeof(buf),

		.table_ctx = &ctx->lrw_table,
		.crypt_ctx = &crypt_ctx,
		.crypt_fn = decrypt_callback,
	};
	int ret;

	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
	ret = lrw_crypt(desc, dst, src, nbytes, &req);
	camellia_fpu_end(crypt_ctx.fpu_enabled);

	return ret;
}

static int xts_encrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
		       struct scatterlist *src, unsigned int nbytes)
{
	struct camellia_xts_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
	be128 buf[CAMELLIA_AESNI_PARALLEL_BLOCKS];
	struct crypt_priv crypt_ctx = {
		.ctx = &ctx->crypt_ctx,
		.fpu_enabled = false,
	};
	struct xts_crypt_req req = {
		.tbuf = buf,
		.tbuflen = sizeof(buf),

		.tweak_ctx = &ctx->tweak_ctx,
		.tweak_fn = XTS_TWEAK_CAST(camellia_enc_blk),
		.crypt_ctx = &crypt_ctx,
		.crypt_fn = encrypt_callback,
	};
	int ret;

	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
	ret = xts_crypt(desc, dst, src, nbytes, &req);
	camellia_fpu_end(crypt_ctx.fpu_enabled);

	return ret;
}

static int xts_decrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
		       struct scatterlist *src, unsigned int nbytes)
{
	struct camellia_xts_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
	be128 buf[CAMELLIA_AESNI_PARALLEL_BLOCKS];
	struct crypt_priv crypt_ctx = {
		.ctx = &ctx->crypt_ctx,
		.fpu_enabled = false,
	};
	struct xts_crypt_req req = {
		.tbuf = buf,
		.tbuflen = sizeof(buf),

		.tweak_ctx = &ctx->tweak_ctx,
		.tweak_fn = XTS_TWEAK_CAST(camellia_enc_blk),
		.crypt_ctx = &crypt_ctx,
		.crypt_fn = decrypt_callback,
	};
	int ret;

	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
	ret = xts_crypt(desc, dst, src, nbytes, &req);
	camellia_fpu_end(crypt_ctx.fpu_enabled);

	return ret;
}

static struct crypto_alg cmll_algs[10] = { {
	.cra_name		= "__ecb-camellia-aesni",
	.cra_driver_name	= "__driver-ecb-camellia-aesni",
	.cra_priority		= 0,
	.cra_flags		= CRYPTO_ALG_TYPE_BLKCIPHER,
	.cra_blocksize		= CAMELLIA_BLOCK_SIZE,
	.cra_ctxsize		= sizeof(struct camellia_ctx),
	.cra_alignmask		= 0,
	.cra_type		= &crypto_blkcipher_type,
	.cra_module		= THIS_MODULE,
	.cra_u = {
		.blkcipher = {
			.min_keysize	= CAMELLIA_MIN_KEY_SIZE,
			.max_keysize	= CAMELLIA_MAX_KEY_SIZE,
			.setkey		= camellia_setkey,
			.encrypt	= ecb_encrypt,
			.decrypt	= ecb_decrypt,
		},
	},
}, {
	.cra_name		= "__cbc-camellia-aesni",
	.cra_driver_name	= "__driver-cbc-camellia-aesni",
	.cra_priority		= 0,
	.cra_flags		= CRYPTO_ALG_TYPE_BLKCIPHER,
	.cra_blocksize		= CAMELLIA_BLOCK_SIZE,
	.cra_ctxsize		= sizeof(struct camellia_ctx),
	.cra_alignmask		= 0,
	.cra_type		= &crypto_blkcipher_type,
	.cra_module		= THIS_MODULE,
	.cra_u = {
		.blkcipher = {
			.min_keysize	= CAMELLIA_MIN_KEY_SIZE,
			.max_keysize	= CAMELLIA_MAX_KEY_SIZE,
			.setkey		= camellia_setkey,
			.encrypt	= cbc_encrypt,
			.decrypt	= cbc_decrypt,
		},
	},
}, {
	.cra_name		= "__ctr-camellia-aesni",
	.cra_driver_name	= "__driver-ctr-camellia-aesni",
	.cra_priority		= 0,
	.cra_flags		= CRYPTO_ALG_TYPE_BLKCIPHER,
	.cra_blocksize		= 1,
	.cra_ctxsize		= sizeof(struct camellia_ctx),
	.cra_alignmask		= 0,
	.cra_type		= &crypto_blkcipher_type,
	.cra_module		= THIS_MODULE,
	.cra_u = {
		.blkcipher = {
			.min_keysize	= CAMELLIA_MIN_KEY_SIZE,
			.max_keysize	= CAMELLIA_MAX_KEY_SIZE,
			.ivsize		= CAMELLIA_BLOCK_SIZE,
			.setkey		= camellia_setkey,
			.encrypt	= ctr_crypt,
			.decrypt	= ctr_crypt,
		},
	},
}, {
	.cra_name		= "__lrw-camellia-aesni",
	.cra_driver_name	= "__driver-lrw-camellia-aesni",
	.cra_priority		= 0,
	.cra_flags		= CRYPTO_ALG_TYPE_BLKCIPHER,
	.cra_blocksize		= CAMELLIA_BLOCK_SIZE,
	.cra_ctxsize		= sizeof(struct camellia_lrw_ctx),
	.cra_alignmask		= 0,
	.cra_type		= &crypto_blkcipher_type,
	.cra_module		= THIS_MODULE,
	.cra_exit		= lrw_camellia_exit_tfm,
	.cra_u = {
		.blkcipher = {
			.min_keysize	= CAMELLIA_MIN_KEY_SIZE +
					  CAMELLIA_BLOCK_SIZE,
			.max_keysize	= CAMELLIA_MAX_KEY_SIZE +
					  CAMELLIA_BLOCK_SIZE,
			.ivsize		= CAMELLIA_BLOCK_SIZE,
			.setkey		= lrw_camellia_setkey,
			.encrypt	= lrw_encrypt,
			.decrypt	= lrw_decrypt,
		},
	},
}, {
	.cra_name		= "__xts-camellia-aesni",
	.cra_driver_name	= "__driver-xts-camellia-aesni",
	.cra_priority		= 0,
	.cra_flags		= CRYPTO_ALG_TYPE_BLKCIPHER,
	.cra_blocksize		= CAMELLIA_BLOCK_SIZE,
	.cra_ctxsize		= sizeof(struct camellia_xts_ctx),
	.cra_alignmask		= 0,
	.cra_type		= &crypto_blkcipher_type,
	.cra_module		= THIS_MODULE,
	.cra_u = {
		.blkcipher = {
			.min_keysize	= CAMELLIA_MIN_KEY_SIZE * 2,
			.max_keysize	= CAMELLIA_MAX_KEY_SIZE * 2,
			.ivsize		= CAMELLIA_BLOCK_SIZE,
			.setkey		= xts_camellia_setkey,
			.encrypt	= xts_encrypt,
			.decrypt	= xts_decrypt,
		},
	},
}, {
	.cra_name		= "ecb(camellia)",
	.cra_driver_name	= "ecb-camellia-aesni",
	.cra_priority		= 400,
	.cra_flags		= CRYPTO_ALG_TYPE_ABLKCIPHER | CRYPTO_ALG_ASYNC,
	.cra_blocksize		= CAMELLIA_BLOCK_SIZE,
	.cra_ctxsize		= sizeof(struct async_helper_ctx),
	.cra_alignmask		= 0,
	.cra_type		= &crypto_ablkcipher_type,
	.cra_module		= THIS_MODULE,
	.cra_init		= ablk_init,
	.cra_exit		= ablk_exit,
	.cra_u = {
		.ablkcipher = {
			.min_keysize	= CAMELLIA_MIN_KEY_SIZE,
			.max_keysize	= CAMELLIA_MAX_KEY_SIZE,
			.setkey		= ablk_set_key,
			.encrypt	= ablk_encrypt,
			.decrypt	= ablk_decrypt,
		},
	},
}, {
	.cra_name		= "cbc(camellia)",
	.cra_driver_name	= "cbc-camellia-aesni",
	.cra_priority		= 400,
	.cra_flags		= CRYPTO_ALG_TYPE_ABLKCIPHER | CRYPTO_ALG_ASYNC,
	.cra_blocksize		= CAMELLIA_BLOCK_SIZE,
	.cra_ctxsize		= sizeof(struct async_helper_ctx),
	.cra_alignmask		= 0,
	.cra_type		= &crypto_ablkcipher_type,
	.cra_module		= THIS_MODULE,
	.cra_init		= ablk_init,
	.cra_exit		= ablk_exit,
	.cra_u = {
		.ablkcipher = {
			.min_keysize	= CAMELLIA_MIN_KEY_SIZE,
			.max_keysize	= CAMELLIA_MAX_KEY_SIZE,
			.ivsize		= CAMELLIA_BLOCK_SIZE,
			.setkey		= ablk_set_key,
			.encrypt	= __ablk_encrypt,
			.decrypt	= ablk_decrypt,
		},
	},
}, {
	.cra_name		= "ctr(camellia)",
	.cra_driver_name	= "ctr-camellia-aesni",
	.cra_priority		= 400,
	.cra_flags		= CRYPTO_ALG_TYPE_ABLKCIPHER | CRYPTO_ALG_ASYNC,
	.cra_blocksize		= 1,
	.cra_ctxsize		= sizeof(struct async_helper_ctx),
	.cra_alignmask		= 0,
	.cra_type		= &crypto_ablkcipher_type,
	.cra_module		= THIS_MODULE,
	.cra_init		= ablk_init,
	.cra_exit		= ablk_exit,
	.cra_u = {
		.ablkcipher = {
			.min_keysize	= CAMELLIA_MIN_KEY_SIZE,
			.max_keysize	= CAMELLIA_MAX_KEY_SIZE,
			.ivsize		= CAMELLIA_BLOCK_SIZE,
			.setkey		= ablk_set_key,
			.encrypt	= ablk_encrypt,
			.decrypt	= ablk_encrypt,
			.geniv		= "chainiv",
		},
	},
}, {
	.cra_name		= "lrw(camellia)",
	.cra_driver_name	= "lrw-camellia-aesni",
	.cra_priority		= 400,
	.cra_flags		= CRYPTO_ALG_TYPE_ABLKCIPHER | CRYPTO_ALG_ASYNC,
	.cra_blocksize		= CAMELLIA_BLOCK_SIZE,
	.cra_ctxsize		= sizeof(struct async_helper_ctx),
	.cra_alignmask		= 0,
	.cra_type		= &crypto_ablkcipher_type,
	.cra_module		= THIS_MODULE,
	.cra_init		= ablk_init,
	.cra_exit		= ablk_exit,
	.cra_u = {
		.ablkcipher = {
			.min_keysize	= CAMELLIA_MIN_KEY_SIZE +
					  CAMELLIA_BLOCK_SIZE,
			.max_keysize	= CAMELLIA_MAX_KEY_SIZE +
					  CAMELLIA_BLOCK_SIZE,
			.ivsize		= CAMELLIA_BLOCK_SIZE,
			.setkey		= ablk_set_key,
			.encrypt	= ablk_encrypt,
			.decrypt	= ablk_decrypt,
		},
	},
}, {
	.cra_name		= "xts(camellia)",
	.cra_driver_name	= "xts-camellia-aesni",
	.cra_priority		= 400,
	.cra_flags		= CRYPTO_ALG_TYPE_ABLKCIPHER | CRYPTO_ALG_ASYNC,
	.cra_blocksize		= CAMELLIA_BLOCK_SIZE,
	.cra_ctxsize		= sizeof(struct async_helper_ctx),
	.cra_alignmask		= 0,
	.cra_type		= &crypto_ablkcipher_type,
	.cra_module		= THIS_MODULE,
	.cra_init		= ablk_init,
	.cra_exit		= ablk_exit,
	.cra_u = {
		.ablkcipher = {
			.min_keysize	= CAMELLIA_MIN_KEY_SIZE * 2,
			.max_keysize	= CAMELLIA_MAX_KEY_SIZE * 2,
			.ivsize		= CAMELLIA_BLOCK_SIZE,
			.setkey		= ablk_set_key,
			.encrypt	= ablk_encrypt,
			.decrypt	= ablk_decrypt,
		},
	},
} };

static int __init camellia_aesni_init(void)
{
	u64 xcr0;

	if (!cpu_has_avx || !cpu_has_aes || !cpu_has_osxsave) {
		pr_info("AVX or AES-NI instructions are not detected.\n");
		return -ENODEV;
	}

	xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
	if ((xcr0 & (XSTATE_SSE | XSTATE_YMM)) != (XSTATE_SSE | XSTATE_YMM)) {
		pr_info("AVX detected but unusable.\n");
		return -ENODEV;
	}

	return crypto_register_algs(cmll_algs, ARRAY_SIZE(cmll_algs));
}

static void __exit camellia_aesni_fini(void)
{
	crypto_unregister_algs(cmll_algs, ARRAY_SIZE(cmll_algs));
}

module_init(camellia_aesni_init);
module_exit(camellia_aesni_fini);

MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("Camellia Cipher Algorithm, AES-NI/AVX optimized");
MODULE_ALIAS("camellia");
MODULE_ALIAS("camellia-asm");
Commit	Line	Data
d9b1d2e7 JK	1	/*
	2	* Glue Code for x86_64/AVX/AES-NI assembler optimized version of Camellia
	3	*
	4	* Copyright © 2012 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
	5	*
	6	* This program is free software; you can redistribute it and/or modify
	7	* it under the terms of the GNU General Public License as published by
	8	* the Free Software Foundation; either version 2 of the License, or
	9	* (at your option) any later version.
	10	*
	11	*/
	12
	13	#include <linux/module.h>
	14	#include <linux/types.h>
	15	#include <linux/crypto.h>
	16	#include <linux/err.h>
	17	#include <crypto/algapi.h>
	18	#include <crypto/ctr.h>
	19	#include <crypto/lrw.h>
	20	#include <crypto/xts.h>
	21	#include <asm/xcr.h>
	22	#include <asm/xsave.h>
	23	#include <asm/crypto/camellia.h>
	24	#include <asm/crypto/ablk_helper.h>
	25	#include <asm/crypto/glue_helper.h>
	26
	27	#define CAMELLIA_AESNI_PARALLEL_BLOCKS 16
	28
	29	/* 16-way AES-NI parallel cipher functions */
	30	asmlinkage void camellia_ecb_enc_16way(struct camellia_ctx ctx, u8 dst,
	31	const u8 *src);
	32	asmlinkage void camellia_ecb_dec_16way(struct camellia_ctx ctx, u8 dst,
	33	const u8 *src);
	34
	35	asmlinkage void camellia_cbc_dec_16way(struct camellia_ctx ctx, u8 dst,
	36	const u8 *src);
	37	asmlinkage void camellia_ctr_16way(struct camellia_ctx ctx, u8 dst,
	38	const u8 src, le128 iv);
	39
	40	static const struct common_glue_ctx camellia_enc = {
	41	.num_funcs = 3,
	42	.fpu_blocks_limit = CAMELLIA_AESNI_PARALLEL_BLOCKS,
	43
	44	.funcs = { {
	45	.num_blocks = CAMELLIA_AESNI_PARALLEL_BLOCKS,
	46	.fn_u = { .ecb = GLUE_FUNC_CAST(camellia_ecb_enc_16way) }
	47	}, {
	48	.num_blocks = 2,
	49	.fn_u = { .ecb = GLUE_FUNC_CAST(camellia_enc_blk_2way) }
	50	}, {
	51	.num_blocks = 1,
	52	.fn_u = { .ecb = GLUE_FUNC_CAST(camellia_enc_blk) }
	53	} }
	54	};
	55
	56	static const struct common_glue_ctx camellia_ctr = {
	57	.num_funcs = 3,
	58	.fpu_blocks_limit = CAMELLIA_AESNI_PARALLEL_BLOCKS,
	59
	60	.funcs = { {
	61	.num_blocks = CAMELLIA_AESNI_PARALLEL_BLOCKS,
	62	.fn_u = { .ctr = GLUE_CTR_FUNC_CAST(camellia_ctr_16way) }
	63	}, {
	64	.num_blocks = 2,
65	.fn_u = { .ctr = GLUE_CTR_FUNC_CAST(camellia_crypt_ctr_2way) }
66	}, {
67	.num_blocks = 1,
68	.fn_u = { .ctr = GLUE_CTR_FUNC_CAST(camellia_crypt_ctr) }
69	} }
70	};
71
72	static const struct common_glue_ctx camellia_dec = {
73	.num_funcs = 3,
74	.fpu_blocks_limit = CAMELLIA_AESNI_PARALLEL_BLOCKS,
75
76	.funcs = { {
77	.num_blocks = CAMELLIA_AESNI_PARALLEL_BLOCKS,
78	.fn_u = { .ecb = GLUE_FUNC_CAST(camellia_ecb_dec_16way) }
79	}, {
80	.num_blocks = 2,
81	.fn_u = { .ecb = GLUE_FUNC_CAST(camellia_dec_blk_2way) }
82	}, {
83	.num_blocks = 1,
84	.fn_u = { .ecb = GLUE_FUNC_CAST(camellia_dec_blk) }
85	} }
86	};
87
88	static const struct common_glue_ctx camellia_dec_cbc = {
89	.num_funcs = 3,
90	.fpu_blocks_limit = CAMELLIA_AESNI_PARALLEL_BLOCKS,
91
92	.funcs = { {
93	.num_blocks = CAMELLIA_AESNI_PARALLEL_BLOCKS,
94	.fn_u = { .cbc = GLUE_CBC_FUNC_CAST(camellia_cbc_dec_16way) }
95	}, {
96	.num_blocks = 2,
97	.fn_u = { .cbc = GLUE_CBC_FUNC_CAST(camellia_decrypt_cbc_2way) }
98	}, {
99	.num_blocks = 1,
100	.fn_u = { .cbc = GLUE_CBC_FUNC_CAST(camellia_dec_blk) }
101	} }
102	};
103
104	static int ecb_encrypt(struct blkcipher_desc desc, struct scatterlist dst,
105	struct scatterlist *src, unsigned int nbytes)
106	{
107	return glue_ecb_crypt_128bit(&camellia_enc, desc, dst, src, nbytes);
108	}
109
110	static int ecb_decrypt(struct blkcipher_desc desc, struct scatterlist dst,
111	struct scatterlist *src, unsigned int nbytes)
112	{
113	return glue_ecb_crypt_128bit(&camellia_dec, desc, dst, src, nbytes);
114	}
115
116	static int cbc_encrypt(struct blkcipher_desc desc, struct scatterlist dst,
117	struct scatterlist *src, unsigned int nbytes)
118	{
119	return glue_cbc_encrypt_128bit(GLUE_FUNC_CAST(camellia_enc_blk), desc,
120	dst, src, nbytes);
121	}
122
123	static int cbc_decrypt(struct blkcipher_desc desc, struct scatterlist dst,
124	struct scatterlist *src, unsigned int nbytes)
125	{
126	return glue_cbc_decrypt_128bit(&camellia_dec_cbc, desc, dst, src,
127	nbytes);
128	}
129
130	static int ctr_crypt(struct blkcipher_desc desc, struct scatterlist dst,
131	struct scatterlist *src, unsigned int nbytes)
132	{
133	return glue_ctr_crypt_128bit(&camellia_ctr, desc, dst, src, nbytes);
134	}
135
136	static inline bool camellia_fpu_begin(bool fpu_enabled, unsigned int nbytes)
137	{
138	return glue_fpu_begin(CAMELLIA_BLOCK_SIZE,
139	CAMELLIA_AESNI_PARALLEL_BLOCKS, NULL, fpu_enabled,
140	nbytes);
141	}
142
143	static inline void camellia_fpu_end(bool fpu_enabled)
144	{
145	glue_fpu_end(fpu_enabled);
146	}
147
148	static int camellia_setkey(struct crypto_tfm tfm, const u8 in_key,
149	unsigned int key_len)
150	{
151	return __camellia_setkey(crypto_tfm_ctx(tfm), in_key, key_len,
152	&tfm->crt_flags);
153	}
154
155	struct crypt_priv {
156	struct camellia_ctx *ctx;
157	bool fpu_enabled;
158	};
159
160	static void encrypt_callback(void priv, u8 srcdst, unsigned int nbytes)
161	{
162	const unsigned int bsize = CAMELLIA_BLOCK_SIZE;
163	struct crypt_priv *ctx = priv;
164	int i;
165
166	ctx->fpu_enabled = camellia_fpu_begin(ctx->fpu_enabled, nbytes);
167
168	if (nbytes >= CAMELLIA_AESNI_PARALLEL_BLOCKS * bsize) {
169	camellia_ecb_enc_16way(ctx->ctx, srcdst, srcdst);
170	srcdst += bsize * CAMELLIA_AESNI_PARALLEL_BLOCKS;
171	nbytes -= bsize * CAMELLIA_AESNI_PARALLEL_BLOCKS;
172	}
173
174	while (nbytes >= CAMELLIA_PARALLEL_BLOCKS * bsize) {
175	camellia_enc_blk_2way(ctx->ctx, srcdst, srcdst);
176	srcdst += bsize * CAMELLIA_PARALLEL_BLOCKS;
177	nbytes -= bsize * CAMELLIA_PARALLEL_BLOCKS;
178	}
179
180	for (i = 0; i < nbytes / bsize; i++, srcdst += bsize)
181	camellia_enc_blk(ctx->ctx, srcdst, srcdst);
182	}
183
184	static void decrypt_callback(void priv, u8 srcdst, unsigned int nbytes)
185	{
186	const unsigned int bsize = CAMELLIA_BLOCK_SIZE;
187	struct crypt_priv *ctx = priv;
188	int i;
189
190	ctx->fpu_enabled = camellia_fpu_begin(ctx->fpu_enabled, nbytes);
191
192	if (nbytes >= CAMELLIA_AESNI_PARALLEL_BLOCKS * bsize) {
193	camellia_ecb_dec_16way(ctx->ctx, srcdst, srcdst);
194	srcdst += bsize * CAMELLIA_AESNI_PARALLEL_BLOCKS;
195	nbytes -= bsize * CAMELLIA_AESNI_PARALLEL_BLOCKS;
196	}
197
198	while (nbytes >= CAMELLIA_PARALLEL_BLOCKS * bsize) {
199	camellia_dec_blk_2way(ctx->ctx, srcdst, srcdst);
200	srcdst += bsize * CAMELLIA_PARALLEL_BLOCKS;
201	nbytes -= bsize * CAMELLIA_PARALLEL_BLOCKS;
202	}
203
204	for (i = 0; i < nbytes / bsize; i++, srcdst += bsize)
205	camellia_dec_blk(ctx->ctx, srcdst, srcdst);
206	}
207
208	static int lrw_encrypt(struct blkcipher_desc desc, struct scatterlist dst,
209	struct scatterlist *src, unsigned int nbytes)
210	{
211	struct camellia_lrw_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
212	be128 buf[CAMELLIA_AESNI_PARALLEL_BLOCKS];
213	struct crypt_priv crypt_ctx = {
214	.ctx = &ctx->camellia_ctx,
215	.fpu_enabled = false,
216	};
217	struct lrw_crypt_req req = {
218	.tbuf = buf,
219	.tbuflen = sizeof(buf),
220
221	.table_ctx = &ctx->lrw_table,
222	.crypt_ctx = &crypt_ctx,
223	.crypt_fn = encrypt_callback,
224	};
225	int ret;
226
227	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
228	ret = lrw_crypt(desc, dst, src, nbytes, &req);
229	camellia_fpu_end(crypt_ctx.fpu_enabled);
230
231	return ret;
232	}
233
234	static int lrw_decrypt(struct blkcipher_desc desc, struct scatterlist dst,
235	struct scatterlist *src, unsigned int nbytes)
236	{
237	struct camellia_lrw_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
238	be128 buf[CAMELLIA_AESNI_PARALLEL_BLOCKS];
239	struct crypt_priv crypt_ctx = {
240	.ctx = &ctx->camellia_ctx,
241	.fpu_enabled = false,
242	};
243	struct lrw_crypt_req req = {
244	.tbuf = buf,
245	.tbuflen = sizeof(buf),
246
247	.table_ctx = &ctx->lrw_table,
248	.crypt_ctx = &crypt_ctx,
249	.crypt_fn = decrypt_callback,
250	};
251	int ret;
252
253	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
254	ret = lrw_crypt(desc, dst, src, nbytes, &req);
255	camellia_fpu_end(crypt_ctx.fpu_enabled);
256
257	return ret;
258	}
259
260	static int xts_encrypt(struct blkcipher_desc desc, struct scatterlist dst,
261	struct scatterlist *src, unsigned int nbytes)
262	{
263	struct camellia_xts_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
264	be128 buf[CAMELLIA_AESNI_PARALLEL_BLOCKS];
265	struct crypt_priv crypt_ctx = {
266	.ctx = &ctx->crypt_ctx,
267	.fpu_enabled = false,
268	};
269	struct xts_crypt_req req = {
270	.tbuf = buf,
271	.tbuflen = sizeof(buf),
272
273	.tweak_ctx = &ctx->tweak_ctx,
274	.tweak_fn = XTS_TWEAK_CAST(camellia_enc_blk),
275	.crypt_ctx = &crypt_ctx,
276	.crypt_fn = encrypt_callback,
277	};
278	int ret;
279
280	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
281	ret = xts_crypt(desc, dst, src, nbytes, &req);
282	camellia_fpu_end(crypt_ctx.fpu_enabled);
283
284	return ret;
285	}
286
287	static int xts_decrypt(struct blkcipher_desc desc, struct scatterlist dst,
288	struct scatterlist *src, unsigned int nbytes)
289	{
290	struct camellia_xts_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
291	be128 buf[CAMELLIA_AESNI_PARALLEL_BLOCKS];
292	struct crypt_priv crypt_ctx = {
293	.ctx = &ctx->crypt_ctx,
294	.fpu_enabled = false,
295	};
296	struct xts_crypt_req req = {
297	.tbuf = buf,
298	.tbuflen = sizeof(buf),
299
300	.tweak_ctx = &ctx->tweak_ctx,
301	.tweak_fn = XTS_TWEAK_CAST(camellia_enc_blk),
302	.crypt_ctx = &crypt_ctx,
303	.crypt_fn = decrypt_callback,
304	};
305	int ret;
306
307	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
308	ret = xts_crypt(desc, dst, src, nbytes, &req);
309	camellia_fpu_end(crypt_ctx.fpu_enabled);
310
311	return ret;
312	}
313
314	static struct crypto_alg cmll_algs[10] = { {
315	.cra_name = "__ecb-camellia-aesni",
316	.cra_driver_name = "__driver-ecb-camellia-aesni",
317	.cra_priority = 0,
318	.cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER,
319	.cra_blocksize = CAMELLIA_BLOCK_SIZE,
320	.cra_ctxsize = sizeof(struct camellia_ctx),
321	.cra_alignmask = 0,
322	.cra_type = &crypto_blkcipher_type,
323	.cra_module = THIS_MODULE,
324	.cra_u = {
325	.blkcipher = {
326	.min_keysize = CAMELLIA_MIN_KEY_SIZE,
327	.max_keysize = CAMELLIA_MAX_KEY_SIZE,
328	.setkey = camellia_setkey,
329	.encrypt = ecb_encrypt,
330	.decrypt = ecb_decrypt,
331	},
332	},
333	}, {
334	.cra_name = "__cbc-camellia-aesni",
335	.cra_driver_name = "__driver-cbc-camellia-aesni",
336	.cra_priority = 0,
337	.cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER,
338	.cra_blocksize = CAMELLIA_BLOCK_SIZE,
339	.cra_ctxsize = sizeof(struct camellia_ctx),
340	.cra_alignmask = 0,
341	.cra_type = &crypto_blkcipher_type,
342	.cra_module = THIS_MODULE,
343	.cra_u = {
344	.blkcipher = {
345	.min_keysize = CAMELLIA_MIN_KEY_SIZE,
346	.max_keysize = CAMELLIA_MAX_KEY_SIZE,
347	.setkey = camellia_setkey,
348	.encrypt = cbc_encrypt,
349	.decrypt = cbc_decrypt,
350	},
351	},
352	}, {
353	.cra_name = "__ctr-camellia-aesni",
354	.cra_driver_name = "__driver-ctr-camellia-aesni",
355	.cra_priority = 0,
356	.cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER,
357	.cra_blocksize = 1,
358	.cra_ctxsize = sizeof(struct camellia_ctx),
359	.cra_alignmask = 0,
360	.cra_type = &crypto_blkcipher_type,
361	.cra_module = THIS_MODULE,
362	.cra_u = {
363	.blkcipher = {
364	.min_keysize = CAMELLIA_MIN_KEY_SIZE,
365	.max_keysize = CAMELLIA_MAX_KEY_SIZE,
366	.ivsize = CAMELLIA_BLOCK_SIZE,
367	.setkey = camellia_setkey,
368	.encrypt = ctr_crypt,
369	.decrypt = ctr_crypt,
370	},
371	},
372	}, {
373	.cra_name = "__lrw-camellia-aesni",
374	.cra_driver_name = "__driver-lrw-camellia-aesni",
375	.cra_priority = 0,
376	.cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER,
377	.cra_blocksize = CAMELLIA_BLOCK_SIZE,
378	.cra_ctxsize = sizeof(struct camellia_lrw_ctx),
379	.cra_alignmask = 0,
380	.cra_type = &crypto_blkcipher_type,
381	.cra_module = THIS_MODULE,
382	.cra_exit = lrw_camellia_exit_tfm,
383	.cra_u = {
384	.blkcipher = {
385	.min_keysize = CAMELLIA_MIN_KEY_SIZE +
386	CAMELLIA_BLOCK_SIZE,
387	.max_keysize = CAMELLIA_MAX_KEY_SIZE +
388	CAMELLIA_BLOCK_SIZE,
389	.ivsize = CAMELLIA_BLOCK_SIZE,
390	.setkey = lrw_camellia_setkey,
391	.encrypt = lrw_encrypt,
392	.decrypt = lrw_decrypt,
393	},
394	},
395	}, {
396	.cra_name = "__xts-camellia-aesni",
397	.cra_driver_name = "__driver-xts-camellia-aesni",
398	.cra_priority = 0,
399	.cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER,
400	.cra_blocksize = CAMELLIA_BLOCK_SIZE,
401	.cra_ctxsize = sizeof(struct camellia_xts_ctx),
402	.cra_alignmask = 0,
403	.cra_type = &crypto_blkcipher_type,
404	.cra_module = THIS_MODULE,
405	.cra_u = {
406	.blkcipher = {
407	.min_keysize = CAMELLIA_MIN_KEY_SIZE * 2,
408	.max_keysize = CAMELLIA_MAX_KEY_SIZE * 2,
409	.ivsize = CAMELLIA_BLOCK_SIZE,
410	.setkey = xts_camellia_setkey,
411	.encrypt = xts_encrypt,
412	.decrypt = xts_decrypt,
413	},
414	},
415	}, {
416	.cra_name = "ecb(camellia)",
417	.cra_driver_name = "ecb-camellia-aesni",
418	.cra_priority = 400,
419	.cra_flags = CRYPTO_ALG_TYPE_ABLKCIPHER \| CRYPTO_ALG_ASYNC,
420	.cra_blocksize = CAMELLIA_BLOCK_SIZE,
421	.cra_ctxsize = sizeof(struct async_helper_ctx),
422	.cra_alignmask = 0,
423	.cra_type = &crypto_ablkcipher_type,
424	.cra_module = THIS_MODULE,
425	.cra_init = ablk_init,
426	.cra_exit = ablk_exit,
427	.cra_u = {
428	.ablkcipher = {
429	.min_keysize = CAMELLIA_MIN_KEY_SIZE,
430	.max_keysize = CAMELLIA_MAX_KEY_SIZE,
431	.setkey = ablk_set_key,
432	.encrypt = ablk_encrypt,
433	.decrypt = ablk_decrypt,
434	},
435	},
436	}, {
437	.cra_name = "cbc(camellia)",
438	.cra_driver_name = "cbc-camellia-aesni",
439	.cra_priority = 400,
440	.cra_flags = CRYPTO_ALG_TYPE_ABLKCIPHER \| CRYPTO_ALG_ASYNC,
441	.cra_blocksize = CAMELLIA_BLOCK_SIZE,
442	.cra_ctxsize = sizeof(struct async_helper_ctx),
443	.cra_alignmask = 0,
444	.cra_type = &crypto_ablkcipher_type,
445	.cra_module = THIS_MODULE,
446	.cra_init = ablk_init,
447	.cra_exit = ablk_exit,
448	.cra_u = {
449	.ablkcipher = {
450	.min_keysize = CAMELLIA_MIN_KEY_SIZE,
451	.max_keysize = CAMELLIA_MAX_KEY_SIZE,
452	.ivsize = CAMELLIA_BLOCK_SIZE,
453	.setkey = ablk_set_key,
454	.encrypt = __ablk_encrypt,
455	.decrypt = ablk_decrypt,
456	},
457	},
458	}, {
459	.cra_name = "ctr(camellia)",
460	.cra_driver_name = "ctr-camellia-aesni",
461	.cra_priority = 400,
462	.cra_flags = CRYPTO_ALG_TYPE_ABLKCIPHER \| CRYPTO_ALG_ASYNC,
463	.cra_blocksize = 1,
464	.cra_ctxsize = sizeof(struct async_helper_ctx),
465	.cra_alignmask = 0,
466	.cra_type = &crypto_ablkcipher_type,
467	.cra_module = THIS_MODULE,
468	.cra_init = ablk_init,
469	.cra_exit = ablk_exit,
470	.cra_u = {
471	.ablkcipher = {
472	.min_keysize = CAMELLIA_MIN_KEY_SIZE,
473	.max_keysize = CAMELLIA_MAX_KEY_SIZE,
474	.ivsize = CAMELLIA_BLOCK_SIZE,
475	.setkey = ablk_set_key,
476	.encrypt = ablk_encrypt,
477	.decrypt = ablk_encrypt,
478	.geniv = "chainiv",
479	},
480	},
481	}, {
482	.cra_name = "lrw(camellia)",
483	.cra_driver_name = "lrw-camellia-aesni",
484	.cra_priority = 400,
485	.cra_flags = CRYPTO_ALG_TYPE_ABLKCIPHER \| CRYPTO_ALG_ASYNC,
486	.cra_blocksize = CAMELLIA_BLOCK_SIZE,
487	.cra_ctxsize = sizeof(struct async_helper_ctx),
488	.cra_alignmask = 0,
489	.cra_type = &crypto_ablkcipher_type,
490	.cra_module = THIS_MODULE,
491	.cra_init = ablk_init,
492	.cra_exit = ablk_exit,
493	.cra_u = {
494	.ablkcipher = {
495	.min_keysize = CAMELLIA_MIN_KEY_SIZE +
496	CAMELLIA_BLOCK_SIZE,
497	.max_keysize = CAMELLIA_MAX_KEY_SIZE +
498	CAMELLIA_BLOCK_SIZE,
499	.ivsize = CAMELLIA_BLOCK_SIZE,
500	.setkey = ablk_set_key,
501	.encrypt = ablk_encrypt,
502	.decrypt = ablk_decrypt,
503	},
504	},
505	}, {
506	.cra_name = "xts(camellia)",
507	.cra_driver_name = "xts-camellia-aesni",
508	.cra_priority = 400,
509	.cra_flags = CRYPTO_ALG_TYPE_ABLKCIPHER \| CRYPTO_ALG_ASYNC,
510	.cra_blocksize = CAMELLIA_BLOCK_SIZE,
511	.cra_ctxsize = sizeof(struct async_helper_ctx),
512	.cra_alignmask = 0,
513	.cra_type = &crypto_ablkcipher_type,
514	.cra_module = THIS_MODULE,
515	.cra_init = ablk_init,
516	.cra_exit = ablk_exit,
517	.cra_u = {
518	.ablkcipher = {
519	.min_keysize = CAMELLIA_MIN_KEY_SIZE * 2,
520	.max_keysize = CAMELLIA_MAX_KEY_SIZE * 2,
521	.ivsize = CAMELLIA_BLOCK_SIZE,
522	.setkey = ablk_set_key,
523	.encrypt = ablk_encrypt,
524	.decrypt = ablk_decrypt,
525	},
526	},
527	} };
528
529	static int __init camellia_aesni_init(void)
530	{
531	u64 xcr0;
532
533	if (!cpu_has_avx \|\| !cpu_has_aes \|\| !cpu_has_osxsave) {
534	pr_info("AVX or AES-NI instructions are not detected.\n");
535	return -ENODEV;
536	}
537
538	xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
539	if ((xcr0 & (XSTATE_SSE \| XSTATE_YMM)) != (XSTATE_SSE \| XSTATE_YMM)) {
540	pr_info("AVX detected but unusable.\n");
541	return -ENODEV;
542	}
543
544	return crypto_register_algs(cmll_algs, ARRAY_SIZE(cmll_algs));
545	}
546
547	static void __exit camellia_aesni_fini(void)
548	{
549	crypto_unregister_algs(cmll_algs, ARRAY_SIZE(cmll_algs));
550	}
551
552	module_init(camellia_aesni_init);
553	module_exit(camellia_aesni_fini);
554
555	MODULE_LICENSE("GPL");
556	MODULE_DESCRIPTION("Camellia Cipher Algorithm, AES-NI/AVX optimized");
557	MODULE_ALIAS("camellia");
558	MODULE_ALIAS("camellia-asm");