[deliverable/linux.git] / arch / powerpc / crypto / aes-spe-glue.c

/*
 * Glue code for AES implementation for SPE instructions (PPC)
 *
 * Based on generic implementation. The assembler module takes care
 * about the SPE registers so it can run from interrupt context.
 *
 * Copyright (c) 2015 Markus Stockhausen <stockhausen@collogia.de>
 *
 * 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 <crypto/aes.h>
#include <linux/module.h>
#include <linux/init.h>
#include <linux/types.h>
#include <linux/errno.h>
#include <linux/crypto.h>
#include <asm/byteorder.h>
#include <asm/switch_to.h>
#include <crypto/algapi.h>

/*
 * MAX_BYTES defines the number of bytes that are allowed to be processed
 * between preempt_disable() and preempt_enable(). e500 cores can issue two
 * instructions per clock cycle using one 32/64 bit unit (SU1) and one 32
 * bit unit (SU2). One of these can be a memory access that is executed via
 * a single load and store unit (LSU). XTS-AES-256 takes ~780 operations per
 * 16 byte block block or 25 cycles per byte. Thus 768 bytes of input data
 * will need an estimated maximum of 20,000 cycles. Headroom for cache misses
 * included. Even with the low end model clocked at 667 MHz this equals to a
 * critical time window of less than 30us. The value has been choosen to
 * process a 512 byte disk block in one or a large 1400 bytes IPsec network
 * packet in two runs.
 *
 */
#define MAX_BYTES 768

struct ppc_aes_ctx {
	u32 key_enc[AES_MAX_KEYLENGTH_U32];
	u32 key_dec[AES_MAX_KEYLENGTH_U32];
	u32 rounds;
};

struct ppc_xts_ctx {
	u32 key_enc[AES_MAX_KEYLENGTH_U32];
	u32 key_dec[AES_MAX_KEYLENGTH_U32];
	u32 key_twk[AES_MAX_KEYLENGTH_U32];
	u32 rounds;
};

extern void ppc_encrypt_aes(u8 *out, const u8 *in, u32 *key_enc, u32 rounds);
extern void ppc_decrypt_aes(u8 *out, const u8 *in, u32 *key_dec, u32 rounds);
extern void ppc_encrypt_ecb(u8 *out, const u8 *in, u32 *key_enc, u32 rounds,
			    u32 bytes);
extern void ppc_decrypt_ecb(u8 *out, const u8 *in, u32 *key_dec, u32 rounds,
			    u32 bytes);
extern void ppc_encrypt_cbc(u8 *out, const u8 *in, u32 *key_enc, u32 rounds,
			    u32 bytes, u8 *iv);
extern void ppc_decrypt_cbc(u8 *out, const u8 *in, u32 *key_dec, u32 rounds,
			    u32 bytes, u8 *iv);
extern void ppc_crypt_ctr  (u8 *out, const u8 *in, u32 *key_enc, u32 rounds,
			    u32 bytes, u8 *iv);
extern void ppc_encrypt_xts(u8 *out, const u8 *in, u32 *key_enc, u32 rounds,
			    u32 bytes, u8 *iv, u32 *key_twk);
extern void ppc_decrypt_xts(u8 *out, const u8 *in, u32 *key_dec, u32 rounds,
			    u32 bytes, u8 *iv, u32 *key_twk);

extern void ppc_expand_key_128(u32 *key_enc, const u8 *key);
extern void ppc_expand_key_192(u32 *key_enc, const u8 *key);
extern void ppc_expand_key_256(u32 *key_enc, const u8 *key);

extern void ppc_generate_decrypt_key(u32 *key_dec,u32 *key_enc,
				     unsigned int key_len);

static void spe_begin(void)
{
	/* disable preemption and save users SPE registers if required */
	preempt_disable();
	enable_kernel_spe();
}

static void spe_end(void)
{
	disable_kernel_spe();
	/* reenable preemption */
	preempt_enable();
}

static int ppc_aes_setkey(struct crypto_tfm *tfm, const u8 *in_key,
		unsigned int key_len)
{
	struct ppc_aes_ctx *ctx = crypto_tfm_ctx(tfm);

	if (key_len != AES_KEYSIZE_128 &&
	    key_len != AES_KEYSIZE_192 &&
	    key_len != AES_KEYSIZE_256) {
		tfm->crt_flags |= CRYPTO_TFM_RES_BAD_KEY_LEN;
		return -EINVAL;
	}

	switch (key_len) {
	case AES_KEYSIZE_128:
		ctx->rounds = 4;
		ppc_expand_key_128(ctx->key_enc, in_key);
		break;
	case AES_KEYSIZE_192:
		ctx->rounds = 5;
		ppc_expand_key_192(ctx->key_enc, in_key);
		break;
	case AES_KEYSIZE_256:
		ctx->rounds = 6;
		ppc_expand_key_256(ctx->key_enc, in_key);
		break;
	}

	ppc_generate_decrypt_key(ctx->key_dec, ctx->key_enc, key_len);

	return 0;
}

static int ppc_xts_setkey(struct crypto_tfm *tfm, const u8 *in_key,
		   unsigned int key_len)
{
	struct ppc_xts_ctx *ctx = crypto_tfm_ctx(tfm);
	int err;

	err = xts_check_key(tfm, in_key, key_len);
	if (err)
		return err;

	key_len >>= 1;

	if (key_len != AES_KEYSIZE_128 &&
	    key_len != AES_KEYSIZE_192 &&
	    key_len != AES_KEYSIZE_256) {
		tfm->crt_flags |= CRYPTO_TFM_RES_BAD_KEY_LEN;
		return -EINVAL;
	}

	switch (key_len) {
	case AES_KEYSIZE_128:
		ctx->rounds = 4;
		ppc_expand_key_128(ctx->key_enc, in_key);
		ppc_expand_key_128(ctx->key_twk, in_key + AES_KEYSIZE_128);
		break;
	case AES_KEYSIZE_192:
		ctx->rounds = 5;
		ppc_expand_key_192(ctx->key_enc, in_key);
		ppc_expand_key_192(ctx->key_twk, in_key + AES_KEYSIZE_192);
		break;
	case AES_KEYSIZE_256:
		ctx->rounds = 6;
		ppc_expand_key_256(ctx->key_enc, in_key);
		ppc_expand_key_256(ctx->key_twk, in_key + AES_KEYSIZE_256);
		break;
	}

	ppc_generate_decrypt_key(ctx->key_dec, ctx->key_enc, key_len);

	return 0;
}

static void ppc_aes_encrypt(struct crypto_tfm *tfm, u8 *out, const u8 *in)
{
	struct ppc_aes_ctx *ctx = crypto_tfm_ctx(tfm);

	spe_begin();
	ppc_encrypt_aes(out, in, ctx->key_enc, ctx->rounds);
	spe_end();
}

static void ppc_aes_decrypt(struct crypto_tfm *tfm, u8 *out, const u8 *in)
{
	struct ppc_aes_ctx *ctx = crypto_tfm_ctx(tfm);

	spe_begin();
	ppc_decrypt_aes(out, in, ctx->key_dec, ctx->rounds);
	spe_end();
}

static int ppc_ecb_encrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
			   struct scatterlist *src, unsigned int nbytes)
{
	struct ppc_aes_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
	struct blkcipher_walk walk;
	unsigned int ubytes;
	int err;

	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
	blkcipher_walk_init(&walk, dst, src, nbytes);
	err = blkcipher_walk_virt(desc, &walk);

	while ((nbytes = walk.nbytes)) {
		ubytes = nbytes > MAX_BYTES ?
			 nbytes - MAX_BYTES : nbytes & (AES_BLOCK_SIZE - 1);
		nbytes -= ubytes;

		spe_begin();
		ppc_encrypt_ecb(walk.dst.virt.addr, walk.src.virt.addr,
				ctx->key_enc, ctx->rounds, nbytes);
		spe_end();

		err = blkcipher_walk_done(desc, &walk, ubytes);
	}

	return err;
}

static int ppc_ecb_decrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
			   struct scatterlist *src, unsigned int nbytes)
{
	struct ppc_aes_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
	struct blkcipher_walk walk;
	unsigned int ubytes;
	int err;

	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
	blkcipher_walk_init(&walk, dst, src, nbytes);
	err = blkcipher_walk_virt(desc, &walk);

	while ((nbytes = walk.nbytes)) {
		ubytes = nbytes > MAX_BYTES ?
			 nbytes - MAX_BYTES : nbytes & (AES_BLOCK_SIZE - 1);
		nbytes -= ubytes;

		spe_begin();
		ppc_decrypt_ecb(walk.dst.virt.addr, walk.src.virt.addr,
				ctx->key_dec, ctx->rounds, nbytes);
		spe_end();

		err = blkcipher_walk_done(desc, &walk, ubytes);
	}

	return err;
}

static int ppc_cbc_encrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
			   struct scatterlist *src, unsigned int nbytes)
{
	struct ppc_aes_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
	struct blkcipher_walk walk;
	unsigned int ubytes;
	int err;

	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
	blkcipher_walk_init(&walk, dst, src, nbytes);
	err = blkcipher_walk_virt(desc, &walk);

	while ((nbytes = walk.nbytes)) {
		ubytes = nbytes > MAX_BYTES ?
			 nbytes - MAX_BYTES : nbytes & (AES_BLOCK_SIZE - 1);
		nbytes -= ubytes;

		spe_begin();
		ppc_encrypt_cbc(walk.dst.virt.addr, walk.src.virt.addr,
				ctx->key_enc, ctx->rounds, nbytes, walk.iv);
		spe_end();

		err = blkcipher_walk_done(desc, &walk, ubytes);
	}

	return err;
}

static int ppc_cbc_decrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
			   struct scatterlist *src, unsigned int nbytes)
{
	struct ppc_aes_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
	struct blkcipher_walk walk;
	unsigned int ubytes;
	int err;

	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
	blkcipher_walk_init(&walk, dst, src, nbytes);
	err = blkcipher_walk_virt(desc, &walk);

	while ((nbytes = walk.nbytes)) {
		ubytes = nbytes > MAX_BYTES ?
			 nbytes - MAX_BYTES : nbytes & (AES_BLOCK_SIZE - 1);
		nbytes -= ubytes;

		spe_begin();
		ppc_decrypt_cbc(walk.dst.virt.addr, walk.src.virt.addr,
				ctx->key_dec, ctx->rounds, nbytes, walk.iv);
		spe_end();

		err = blkcipher_walk_done(desc, &walk, ubytes);
	}

	return err;
}

static int ppc_ctr_crypt(struct blkcipher_desc *desc, struct scatterlist *dst,
			 struct scatterlist *src, unsigned int nbytes)
{
	struct ppc_aes_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
	struct blkcipher_walk walk;
	unsigned int pbytes, ubytes;
	int err;

	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
	blkcipher_walk_init(&walk, dst, src, nbytes);
	err = blkcipher_walk_virt_block(desc, &walk, AES_BLOCK_SIZE);

	while ((pbytes = walk.nbytes)) {
		pbytes = pbytes > MAX_BYTES ? MAX_BYTES : pbytes;
		pbytes = pbytes == nbytes ?
			 nbytes : pbytes & ~(AES_BLOCK_SIZE - 1);
		ubytes = walk.nbytes - pbytes;

		spe_begin();
		ppc_crypt_ctr(walk.dst.virt.addr, walk.src.virt.addr,
			      ctx->key_enc, ctx->rounds, pbytes , walk.iv);
		spe_end();

		nbytes -= pbytes;
		err = blkcipher_walk_done(desc, &walk, ubytes);
	}

	return err;
}

static int ppc_xts_encrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
			   struct scatterlist *src, unsigned int nbytes)
{
	struct ppc_xts_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
	struct blkcipher_walk walk;
	unsigned int ubytes;
	int err;
	u32 *twk;

	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
	blkcipher_walk_init(&walk, dst, src, nbytes);
	err = blkcipher_walk_virt(desc, &walk);
	twk = ctx->key_twk;

	while ((nbytes = walk.nbytes)) {
		ubytes = nbytes > MAX_BYTES ?
			 nbytes - MAX_BYTES : nbytes & (AES_BLOCK_SIZE - 1);
		nbytes -= ubytes;

		spe_begin();
		ppc_encrypt_xts(walk.dst.virt.addr, walk.src.virt.addr,
				ctx->key_enc, ctx->rounds, nbytes, walk.iv, twk);
		spe_end();

		twk = NULL;
		err = blkcipher_walk_done(desc, &walk, ubytes);
	}

	return err;
}

static int ppc_xts_decrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
			   struct scatterlist *src, unsigned int nbytes)
{
	struct ppc_xts_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
	struct blkcipher_walk walk;
	unsigned int ubytes;
	int err;
	u32 *twk;

	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
	blkcipher_walk_init(&walk, dst, src, nbytes);
	err = blkcipher_walk_virt(desc, &walk);
	twk = ctx->key_twk;

	while ((nbytes = walk.nbytes)) {
		ubytes = nbytes > MAX_BYTES ?
			 nbytes - MAX_BYTES : nbytes & (AES_BLOCK_SIZE - 1);
		nbytes -= ubytes;

		spe_begin();
		ppc_decrypt_xts(walk.dst.virt.addr, walk.src.virt.addr,
				ctx->key_dec, ctx->rounds, nbytes, walk.iv, twk);
		spe_end();

		twk = NULL;
		err = blkcipher_walk_done(desc, &walk, ubytes);
	}

	return err;
}

/*
 * Algorithm definitions. Disabling alignment (cra_alignmask=0) was chosen
 * because the e500 platform can handle unaligned reads/writes very efficently.
 * This improves IPsec thoughput by another few percent. Additionally we assume
 * that AES context is always aligned to at least 8 bytes because it is created
 * with kmalloc() in the crypto infrastructure
 *
 */
static struct crypto_alg aes_algs[] = { {
	.cra_name		=	"aes",
	.cra_driver_name	=	"aes-ppc-spe",
	.cra_priority		=	300,
	.cra_flags		=	CRYPTO_ALG_TYPE_CIPHER,
	.cra_blocksize		=	AES_BLOCK_SIZE,
	.cra_ctxsize		=	sizeof(struct ppc_aes_ctx),
	.cra_alignmask		=	0,
	.cra_module		=	THIS_MODULE,
	.cra_u			=	{
		.cipher = {
			.cia_min_keysize	=	AES_MIN_KEY_SIZE,
			.cia_max_keysize	=	AES_MAX_KEY_SIZE,
			.cia_setkey		=	ppc_aes_setkey,
			.cia_encrypt		=	ppc_aes_encrypt,
			.cia_decrypt		=	ppc_aes_decrypt
		}
	}
}, {
	.cra_name		=	"ecb(aes)",
	.cra_driver_name	=	"ecb-ppc-spe",
	.cra_priority		=	300,
	.cra_flags		=	CRYPTO_ALG_TYPE_BLKCIPHER,
	.cra_blocksize		=	AES_BLOCK_SIZE,
	.cra_ctxsize		=	sizeof(struct ppc_aes_ctx),
	.cra_alignmask		=	0,
	.cra_type		=	&crypto_blkcipher_type,
	.cra_module		=	THIS_MODULE,
	.cra_u = {
		.blkcipher = {
			.min_keysize		=	AES_MIN_KEY_SIZE,
			.max_keysize		=	AES_MAX_KEY_SIZE,
			.ivsize			=	AES_BLOCK_SIZE,
			.setkey			=	ppc_aes_setkey,
			.encrypt		=	ppc_ecb_encrypt,
			.decrypt		=	ppc_ecb_decrypt,
		}
	}
}, {
	.cra_name		=	"cbc(aes)",
	.cra_driver_name	=	"cbc-ppc-spe",
	.cra_priority		=	300,
	.cra_flags		=	CRYPTO_ALG_TYPE_BLKCIPHER,
	.cra_blocksize		=	AES_BLOCK_SIZE,
	.cra_ctxsize		=	sizeof(struct ppc_aes_ctx),
	.cra_alignmask		=	0,
	.cra_type		=	&crypto_blkcipher_type,
	.cra_module		=	THIS_MODULE,
	.cra_u = {
		.blkcipher = {
			.min_keysize		=	AES_MIN_KEY_SIZE,
			.max_keysize		=	AES_MAX_KEY_SIZE,
			.ivsize			=	AES_BLOCK_SIZE,
			.setkey			=	ppc_aes_setkey,
			.encrypt		=	ppc_cbc_encrypt,
			.decrypt		=	ppc_cbc_decrypt,
		}
	}
}, {
	.cra_name		=	"ctr(aes)",
	.cra_driver_name	=	"ctr-ppc-spe",
	.cra_priority		=	300,
	.cra_flags		=	CRYPTO_ALG_TYPE_BLKCIPHER,
	.cra_blocksize		=	1,
	.cra_ctxsize		=	sizeof(struct ppc_aes_ctx),
	.cra_alignmask		=	0,
	.cra_type		=	&crypto_blkcipher_type,
	.cra_module		=	THIS_MODULE,
	.cra_u = {
		.blkcipher = {
			.min_keysize		=	AES_MIN_KEY_SIZE,
			.max_keysize		=	AES_MAX_KEY_SIZE,
			.ivsize			=	AES_BLOCK_SIZE,
			.setkey			=	ppc_aes_setkey,
			.encrypt		=	ppc_ctr_crypt,
			.decrypt		=	ppc_ctr_crypt,
		}
	}
}, {
	.cra_name		=	"xts(aes)",
	.cra_driver_name	=	"xts-ppc-spe",
	.cra_priority		=	300,
	.cra_flags		=	CRYPTO_ALG_TYPE_BLKCIPHER,
	.cra_blocksize		=	AES_BLOCK_SIZE,
	.cra_ctxsize		=	sizeof(struct ppc_xts_ctx),
	.cra_alignmask		=	0,
	.cra_type		=	&crypto_blkcipher_type,
	.cra_module		=	THIS_MODULE,
	.cra_u = {
		.blkcipher = {
			.min_keysize		=	AES_MIN_KEY_SIZE * 2,
			.max_keysize		=	AES_MAX_KEY_SIZE * 2,
			.ivsize			=	AES_BLOCK_SIZE,
			.setkey			=	ppc_xts_setkey,
			.encrypt		=	ppc_xts_encrypt,
			.decrypt		=	ppc_xts_decrypt,
		}
	}
} };

static int __init ppc_aes_mod_init(void)
{
	return crypto_register_algs(aes_algs, ARRAY_SIZE(aes_algs));
}

static void __exit ppc_aes_mod_fini(void)
{
	crypto_unregister_algs(aes_algs, ARRAY_SIZE(aes_algs));
}

module_init(ppc_aes_mod_init);
module_exit(ppc_aes_mod_fini);

MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("AES-ECB/CBC/CTR/XTS, SPE optimized");

MODULE_ALIAS_CRYPTO("aes");
MODULE_ALIAS_CRYPTO("ecb(aes)");
MODULE_ALIAS_CRYPTO("cbc(aes)");
MODULE_ALIAS_CRYPTO("ctr(aes)");
MODULE_ALIAS_CRYPTO("xts(aes)");
MODULE_ALIAS_CRYPTO("aes-ppc-spe");
Commit	Line	Data
8a28a1a8 MS	1	/*
	2	* Glue code for AES implementation for SPE instructions (PPC)
	3	*
	4	* Based on generic implementation. The assembler module takes care
	5	* about the SPE registers so it can run from interrupt context.
	6	*
	7	* Copyright (c) 2015 Markus Stockhausen <stockhausen@collogia.de>
	8	*
	9	* This program is free software; you can redistribute it and/or modify it
	10	* under the terms of the GNU General Public License as published by the Free
	11	* Software Foundation; either version 2 of the License, or (at your option)
	12	* any later version.
	13	*
	14	*/
	15
	16	#include <crypto/aes.h>
	17	#include <linux/module.h>
	18	#include <linux/init.h>
	19	#include <linux/types.h>
	20	#include <linux/errno.h>
	21	#include <linux/crypto.h>
	22	#include <asm/byteorder.h>
	23	#include <asm/switch_to.h>
	24	#include <crypto/algapi.h>
	25
	26	/*
	27	* MAX_BYTES defines the number of bytes that are allowed to be processed
	28	* between preempt_disable() and preempt_enable(). e500 cores can issue two
	29	* instructions per clock cycle using one 32/64 bit unit (SU1) and one 32
	30	* bit unit (SU2). One of these can be a memory access that is executed via
	31	* a single load and store unit (LSU). XTS-AES-256 takes ~780 operations per
	32	* 16 byte block block or 25 cycles per byte. Thus 768 bytes of input data
	33	* will need an estimated maximum of 20,000 cycles. Headroom for cache misses
	34	* included. Even with the low end model clocked at 667 MHz this equals to a
	35	* critical time window of less than 30us. The value has been choosen to
	36	* process a 512 byte disk block in one or a large 1400 bytes IPsec network
	37	* packet in two runs.
	38	*
	39	*/
	40	#define MAX_BYTES 768
	41
	42	struct ppc_aes_ctx {
	43	u32 key_enc[AES_MAX_KEYLENGTH_U32];
	44	u32 key_dec[AES_MAX_KEYLENGTH_U32];
	45	u32 rounds;
	46	};
	47
	48	struct ppc_xts_ctx {
	49	u32 key_enc[AES_MAX_KEYLENGTH_U32];
	50	u32 key_dec[AES_MAX_KEYLENGTH_U32];
	51	u32 key_twk[AES_MAX_KEYLENGTH_U32];
	52	u32 rounds;
	53	};
	54
	55	extern void ppc_encrypt_aes(u8 out, const u8 in, u32 *key_enc, u32 rounds);
	56	extern void ppc_decrypt_aes(u8 out, const u8 in, u32 *key_dec, u32 rounds);
	57	extern void ppc_encrypt_ecb(u8 out, const u8 in, u32 *key_enc, u32 rounds,
	58	u32 bytes);
	59	extern void ppc_decrypt_ecb(u8 out, const u8 in, u32 *key_dec, u32 rounds,
	60	u32 bytes);
	61	extern void ppc_encrypt_cbc(u8 out, const u8 in, u32 *key_enc, u32 rounds,
	62	u32 bytes, u8 *iv);
	63	extern void ppc_decrypt_cbc(u8 out, const u8 in, u32 *key_dec, u32 rounds,
	64	u32 bytes, u8 *iv);
65	extern void ppc_crypt_ctr (u8 out, const u8 in, u32 *key_enc, u32 rounds,
66	u32 bytes, u8 *iv);
67	extern void ppc_encrypt_xts(u8 out, const u8 in, u32 *key_enc, u32 rounds,
68	u32 bytes, u8 iv, u32 key_twk);
69	extern void ppc_decrypt_xts(u8 out, const u8 in, u32 *key_dec, u32 rounds,
70	u32 bytes, u8 iv, u32 key_twk);
71
72	extern void ppc_expand_key_128(u32 key_enc, const u8 key);
73	extern void ppc_expand_key_192(u32 key_enc, const u8 key);
74	extern void ppc_expand_key_256(u32 key_enc, const u8 key);
75
76	extern void ppc_generate_decrypt_key(u32 key_dec,u32 key_enc,
77	unsigned int key_len);
78
79	static void spe_begin(void)
80	{
81	/* disable preemption and save users SPE registers if required */
82	preempt_disable();
83	enable_kernel_spe();
84	}
85
86	static void spe_end(void)
87	{
dc4fbba1	88	disable_kernel_spe();
8a28a1a8 MS	89	/* reenable preemption */
	90	preempt_enable();
	91	}
	92
	93	static int ppc_aes_setkey(struct crypto_tfm tfm, const u8 in_key,
	94	unsigned int key_len)
	95	{
	96	struct ppc_aes_ctx *ctx = crypto_tfm_ctx(tfm);
	97
	98	if (key_len != AES_KEYSIZE_128 &&
	99	key_len != AES_KEYSIZE_192 &&
	100	key_len != AES_KEYSIZE_256) {
	101	tfm->crt_flags \|= CRYPTO_TFM_RES_BAD_KEY_LEN;
	102	return -EINVAL;
	103	}
	104
	105	switch (key_len) {
	106	case AES_KEYSIZE_128:
	107	ctx->rounds = 4;
	108	ppc_expand_key_128(ctx->key_enc, in_key);
	109	break;
	110	case AES_KEYSIZE_192:
	111	ctx->rounds = 5;
	112	ppc_expand_key_192(ctx->key_enc, in_key);
	113	break;
	114	case AES_KEYSIZE_256:
	115	ctx->rounds = 6;
	116	ppc_expand_key_256(ctx->key_enc, in_key);
	117	break;
	118	}
	119
	120	ppc_generate_decrypt_key(ctx->key_dec, ctx->key_enc, key_len);
	121
	122	return 0;
	123	}
	124
	125	static int ppc_xts_setkey(struct crypto_tfm tfm, const u8 in_key,
	126	unsigned int key_len)
	127	{
	128	struct ppc_xts_ctx *ctx = crypto_tfm_ctx(tfm);
28856a9e SM	129	int err;
	130
	131	err = xts_check_key(tfm, in_key, key_len);
	132	if (err)
	133	return err;
8a28a1a8 MS	134
	135	key_len >>= 1;
	136
	137	if (key_len != AES_KEYSIZE_128 &&
	138	key_len != AES_KEYSIZE_192 &&
	139	key_len != AES_KEYSIZE_256) {
	140	tfm->crt_flags \|= CRYPTO_TFM_RES_BAD_KEY_LEN;
	141	return -EINVAL;
	142	}
	143
	144	switch (key_len) {
	145	case AES_KEYSIZE_128:
	146	ctx->rounds = 4;
	147	ppc_expand_key_128(ctx->key_enc, in_key);
	148	ppc_expand_key_128(ctx->key_twk, in_key + AES_KEYSIZE_128);
	149	break;
	150	case AES_KEYSIZE_192:
	151	ctx->rounds = 5;
	152	ppc_expand_key_192(ctx->key_enc, in_key);
	153	ppc_expand_key_192(ctx->key_twk, in_key + AES_KEYSIZE_192);
	154	break;
	155	case AES_KEYSIZE_256:
	156	ctx->rounds = 6;
	157	ppc_expand_key_256(ctx->key_enc, in_key);
	158	ppc_expand_key_256(ctx->key_twk, in_key + AES_KEYSIZE_256);
	159	break;
	160	}
	161
	162	ppc_generate_decrypt_key(ctx->key_dec, ctx->key_enc, key_len);
	163
	164	return 0;
	165	}
	166
	167	static void ppc_aes_encrypt(struct crypto_tfm tfm, u8 out, const u8 *in)
	168	{
	169	struct ppc_aes_ctx *ctx = crypto_tfm_ctx(tfm);
	170
	171	spe_begin();
	172	ppc_encrypt_aes(out, in, ctx->key_enc, ctx->rounds);
	173	spe_end();
	174	}
	175
	176	static void ppc_aes_decrypt(struct crypto_tfm tfm, u8 out, const u8 *in)
	177	{
	178	struct ppc_aes_ctx *ctx = crypto_tfm_ctx(tfm);
	179
	180	spe_begin();
	181	ppc_decrypt_aes(out, in, ctx->key_dec, ctx->rounds);
	182	spe_end();
	183	}
	184
	185	static int ppc_ecb_encrypt(struct blkcipher_desc desc, struct scatterlist dst,
	186	struct scatterlist *src, unsigned int nbytes)
	187	{
	188	struct ppc_aes_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
	189	struct blkcipher_walk walk;
	190	unsigned int ubytes;
	191	int err;
	192
	193	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
	194	blkcipher_walk_init(&walk, dst, src, nbytes);
	195	err = blkcipher_walk_virt(desc, &walk);
	196
	197	while ((nbytes = walk.nbytes)) {
198	ubytes = nbytes > MAX_BYTES ?
199	nbytes - MAX_BYTES : nbytes & (AES_BLOCK_SIZE - 1);
200	nbytes -= ubytes;
201
202	spe_begin();
203	ppc_encrypt_ecb(walk.dst.virt.addr, walk.src.virt.addr,
204	ctx->key_enc, ctx->rounds, nbytes);
205	spe_end();
206
207	err = blkcipher_walk_done(desc, &walk, ubytes);
208	}
209
210	return err;
211	}
212
213	static int ppc_ecb_decrypt(struct blkcipher_desc desc, struct scatterlist dst,
214	struct scatterlist *src, unsigned int nbytes)
215	{
216	struct ppc_aes_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
217	struct blkcipher_walk walk;
218	unsigned int ubytes;
219	int err;
220
221	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
222	blkcipher_walk_init(&walk, dst, src, nbytes);
223	err = blkcipher_walk_virt(desc, &walk);
224
225	while ((nbytes = walk.nbytes)) {
226	ubytes = nbytes > MAX_BYTES ?
227	nbytes - MAX_BYTES : nbytes & (AES_BLOCK_SIZE - 1);
228	nbytes -= ubytes;
229
230	spe_begin();
231	ppc_decrypt_ecb(walk.dst.virt.addr, walk.src.virt.addr,
232	ctx->key_dec, ctx->rounds, nbytes);
233	spe_end();
234
235	err = blkcipher_walk_done(desc, &walk, ubytes);
236	}
237
238	return err;
239	}
240
241	static int ppc_cbc_encrypt(struct blkcipher_desc desc, struct scatterlist dst,
242	struct scatterlist *src, unsigned int nbytes)
243	{
244	struct ppc_aes_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
245	struct blkcipher_walk walk;
246	unsigned int ubytes;
247	int err;
248
249	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
250	blkcipher_walk_init(&walk, dst, src, nbytes);
251	err = blkcipher_walk_virt(desc, &walk);
252
253	while ((nbytes = walk.nbytes)) {
254	ubytes = nbytes > MAX_BYTES ?
255	nbytes - MAX_BYTES : nbytes & (AES_BLOCK_SIZE - 1);
256	nbytes -= ubytes;
257
258	spe_begin();
259	ppc_encrypt_cbc(walk.dst.virt.addr, walk.src.virt.addr,
260	ctx->key_enc, ctx->rounds, nbytes, walk.iv);
261	spe_end();
262
263	err = blkcipher_walk_done(desc, &walk, ubytes);
264	}
265
266	return err;
267	}
268
269	static int ppc_cbc_decrypt(struct blkcipher_desc desc, struct scatterlist dst,
270	struct scatterlist *src, unsigned int nbytes)
271	{
272	struct ppc_aes_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
273	struct blkcipher_walk walk;
274	unsigned int ubytes;
275	int err;
276
277	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
278	blkcipher_walk_init(&walk, dst, src, nbytes);
279	err = blkcipher_walk_virt(desc, &walk);
280
281	while ((nbytes = walk.nbytes)) {
282	ubytes = nbytes > MAX_BYTES ?
283	nbytes - MAX_BYTES : nbytes & (AES_BLOCK_SIZE - 1);
284	nbytes -= ubytes;
285
286	spe_begin();
287	ppc_decrypt_cbc(walk.dst.virt.addr, walk.src.virt.addr,
288	ctx->key_dec, ctx->rounds, nbytes, walk.iv);
289	spe_end();
290
291	err = blkcipher_walk_done(desc, &walk, ubytes);
292	}
293
294	return err;
295	}
296
297	static int ppc_ctr_crypt(struct blkcipher_desc desc, struct scatterlist dst,
298	struct scatterlist *src, unsigned int nbytes)
299	{
300	struct ppc_aes_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
301	struct blkcipher_walk walk;
302	unsigned int pbytes, ubytes;
303	int err;
304
305	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
306	blkcipher_walk_init(&walk, dst, src, nbytes);
307	err = blkcipher_walk_virt_block(desc, &walk, AES_BLOCK_SIZE);
308
309	while ((pbytes = walk.nbytes)) {
310	pbytes = pbytes > MAX_BYTES ? MAX_BYTES : pbytes;
311	pbytes = pbytes == nbytes ?
312	nbytes : pbytes & ~(AES_BLOCK_SIZE - 1);
313	ubytes = walk.nbytes - pbytes;
314
315	spe_begin();
316	ppc_crypt_ctr(walk.dst.virt.addr, walk.src.virt.addr,
317	ctx->key_enc, ctx->rounds, pbytes , walk.iv);
318	spe_end();
319
320	nbytes -= pbytes;
321	err = blkcipher_walk_done(desc, &walk, ubytes);
322	}
323
324	return err;
325	}
326
327	static int ppc_xts_encrypt(struct blkcipher_desc desc, struct scatterlist dst,
328	struct scatterlist *src, unsigned int nbytes)
329	{
330	struct ppc_xts_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
331	struct blkcipher_walk walk;
332	unsigned int ubytes;
333	int err;
334	u32 *twk;
335
336	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
337	blkcipher_walk_init(&walk, dst, src, nbytes);
338	err = blkcipher_walk_virt(desc, &walk);
339	twk = ctx->key_twk;
340
341	while ((nbytes = walk.nbytes)) {
342	ubytes = nbytes > MAX_BYTES ?
343	nbytes - MAX_BYTES : nbytes & (AES_BLOCK_SIZE - 1);
344	nbytes -= ubytes;
345
346	spe_begin();
347	ppc_encrypt_xts(walk.dst.virt.addr, walk.src.virt.addr,
348	ctx->key_enc, ctx->rounds, nbytes, walk.iv, twk);
349	spe_end();
350
351	twk = NULL;
352	err = blkcipher_walk_done(desc, &walk, ubytes);
353	}
354
355	return err;
356	}
357
358	static int ppc_xts_decrypt(struct blkcipher_desc desc, struct scatterlist dst,
359	struct scatterlist *src, unsigned int nbytes)
360	{
361	struct ppc_xts_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
362	struct blkcipher_walk walk;
363	unsigned int ubytes;
364	int err;
365	u32 *twk;
366
367	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
368	blkcipher_walk_init(&walk, dst, src, nbytes);
369	err = blkcipher_walk_virt(desc, &walk);
370	twk = ctx->key_twk;
371
372	while ((nbytes = walk.nbytes)) {
373	ubytes = nbytes > MAX_BYTES ?
374	nbytes - MAX_BYTES : nbytes & (AES_BLOCK_SIZE - 1);
375	nbytes -= ubytes;
376
377	spe_begin();
378	ppc_decrypt_xts(walk.dst.virt.addr, walk.src.virt.addr,
379	ctx->key_dec, ctx->rounds, nbytes, walk.iv, twk);
380	spe_end();
381
382	twk = NULL;
383	err = blkcipher_walk_done(desc, &walk, ubytes);
384	}
385
386	return err;
387	}
388
389	/*
390	* Algorithm definitions. Disabling alignment (cra_alignmask=0) was chosen
391	* because the e500 platform can handle unaligned reads/writes very efficently.
392	* This improves IPsec thoughput by another few percent. Additionally we assume
393	* that AES context is always aligned to at least 8 bytes because it is created
394	* with kmalloc() in the crypto infrastructure
395	*
396	*/
397	static struct crypto_alg aes_algs[] = { {
398	.cra_name = "aes",
399	.cra_driver_name = "aes-ppc-spe",
400	.cra_priority = 300,
401	.cra_flags = CRYPTO_ALG_TYPE_CIPHER,
402	.cra_blocksize = AES_BLOCK_SIZE,
403	.cra_ctxsize = sizeof(struct ppc_aes_ctx),
404	.cra_alignmask = 0,
405	.cra_module = THIS_MODULE,
406	.cra_u = {
407	.cipher = {
408	.cia_min_keysize = AES_MIN_KEY_SIZE,
409	.cia_max_keysize = AES_MAX_KEY_SIZE,
410	.cia_setkey = ppc_aes_setkey,
411	.cia_encrypt = ppc_aes_encrypt,
412	.cia_decrypt = ppc_aes_decrypt
413	}
414	}
415	}, {
416	.cra_name = "ecb(aes)",
417	.cra_driver_name = "ecb-ppc-spe",
418	.cra_priority = 300,
419	.cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER,
420	.cra_blocksize = AES_BLOCK_SIZE,
421	.cra_ctxsize = sizeof(struct ppc_aes_ctx),
422	.cra_alignmask = 0,
423	.cra_type = &crypto_blkcipher_type,
424	.cra_module = THIS_MODULE,
425	.cra_u = {
426	.blkcipher = {
427	.min_keysize = AES_MIN_KEY_SIZE,
428	.max_keysize = AES_MAX_KEY_SIZE,
429	.ivsize = AES_BLOCK_SIZE,
430	.setkey = ppc_aes_setkey,
431	.encrypt = ppc_ecb_encrypt,
432	.decrypt = ppc_ecb_decrypt,
433	}
434	}
435	}, {
436	.cra_name = "cbc(aes)",
437	.cra_driver_name = "cbc-ppc-spe",
438	.cra_priority = 300,
439	.cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER,
440	.cra_blocksize = AES_BLOCK_SIZE,
441	.cra_ctxsize = sizeof(struct ppc_aes_ctx),
442	.cra_alignmask = 0,
443	.cra_type = &crypto_blkcipher_type,
444	.cra_module = THIS_MODULE,
445	.cra_u = {
446	.blkcipher = {
447	.min_keysize = AES_MIN_KEY_SIZE,
448	.max_keysize = AES_MAX_KEY_SIZE,
449	.ivsize = AES_BLOCK_SIZE,
450	.setkey = ppc_aes_setkey,
451	.encrypt = ppc_cbc_encrypt,
452	.decrypt = ppc_cbc_decrypt,
453	}
454	}
455	}, {
456	.cra_name = "ctr(aes)",
457	.cra_driver_name = "ctr-ppc-spe",
458	.cra_priority = 300,
459	.cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER,
460	.cra_blocksize = 1,
461	.cra_ctxsize = sizeof(struct ppc_aes_ctx),
462	.cra_alignmask = 0,
463	.cra_type = &crypto_blkcipher_type,
464	.cra_module = THIS_MODULE,
465	.cra_u = {
466	.blkcipher = {
467	.min_keysize = AES_MIN_KEY_SIZE,
468	.max_keysize = AES_MAX_KEY_SIZE,
469	.ivsize = AES_BLOCK_SIZE,
470	.setkey = ppc_aes_setkey,
471	.encrypt = ppc_ctr_crypt,
472	.decrypt = ppc_ctr_crypt,
473	}
474	}
475	}, {
476	.cra_name = "xts(aes)",
477	.cra_driver_name = "xts-ppc-spe",
478	.cra_priority = 300,
479	.cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER,
480	.cra_blocksize = AES_BLOCK_SIZE,
481	.cra_ctxsize = sizeof(struct ppc_xts_ctx),
482	.cra_alignmask = 0,
483	.cra_type = &crypto_blkcipher_type,
484	.cra_module = THIS_MODULE,
485	.cra_u = {
486	.blkcipher = {
487	.min_keysize = AES_MIN_KEY_SIZE * 2,
488	.max_keysize = AES_MAX_KEY_SIZE * 2,
489	.ivsize = AES_BLOCK_SIZE,
490	.setkey = ppc_xts_setkey,
491	.encrypt = ppc_xts_encrypt,
492	.decrypt = ppc_xts_decrypt,
493	}
494	}
495	} };
496
497	static int __init ppc_aes_mod_init(void)
498	{
499	return crypto_register_algs(aes_algs, ARRAY_SIZE(aes_algs));
500	}
501
502	static void __exit ppc_aes_mod_fini(void)
503	{
504	crypto_unregister_algs(aes_algs, ARRAY_SIZE(aes_algs));
505	}
506
507	module_init(ppc_aes_mod_init);
508	module_exit(ppc_aes_mod_fini);
509
510	MODULE_LICENSE("GPL");
511	MODULE_DESCRIPTION("AES-ECB/CBC/CTR/XTS, SPE optimized");
512
513	MODULE_ALIAS_CRYPTO("aes");
514	MODULE_ALIAS_CRYPTO("ecb(aes)");
515	MODULE_ALIAS_CRYPTO("cbc(aes)");
516	MODULE_ALIAS_CRYPTO("ctr(aes)");
517	MODULE_ALIAS_CRYPTO("xts(aes)");
518	MODULE_ALIAS_CRYPTO("aes-ppc-spe");