powerpc/crypto: AES-XCBC mode routines for nx encryption
authorKent Yoder <key@linux.vnet.ibm.com>
Mon, 14 May 2012 11:06:09 +0000 (11:06 +0000)
committerBenjamin Herrenschmidt <benh@kernel.crashing.org>
Wed, 16 May 2012 05:05:43 +0000 (15:05 +1000)
These routines add support for AES in XCBC mode on the Power7+ CPU's
in-Nest accelerator driver.

Signed-off-by: Kent Yoder <key@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
drivers/crypto/nx/nx-aes-xcbc.c [new file with mode: 0644]

diff --git a/drivers/crypto/nx/nx-aes-xcbc.c b/drivers/crypto/nx/nx-aes-xcbc.c
new file mode 100644 (file)
index 0000000..93923e4
--- /dev/null
@@ -0,0 +1,236 @@
+/**
+ * AES XCBC routines supporting the Power 7+ Nest Accelerators driver
+ *
+ * Copyright (C) 2011-2012 International Business Machines Inc.
+ *
+ * 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; version 2 only.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ *
+ * Author: Kent Yoder <yoder1@us.ibm.com>
+ */
+
+#include <crypto/internal/hash.h>
+#include <crypto/aes.h>
+#include <crypto/algapi.h>
+#include <linux/module.h>
+#include <linux/types.h>
+#include <linux/crypto.h>
+#include <asm/vio.h>
+
+#include "nx_csbcpb.h"
+#include "nx.h"
+
+
+struct xcbc_state {
+       u8 state[AES_BLOCK_SIZE];
+       unsigned int count;
+       u8 buffer[AES_BLOCK_SIZE];
+};
+
+static int nx_xcbc_set_key(struct crypto_shash *desc,
+                          const u8            *in_key,
+                          unsigned int         key_len)
+{
+       struct nx_crypto_ctx *nx_ctx = crypto_shash_ctx(desc);
+
+       switch (key_len) {
+       case AES_KEYSIZE_128:
+               nx_ctx->ap = &nx_ctx->props[NX_PROPS_AES_128];
+               break;
+       default:
+               return -EINVAL;
+       }
+
+       memcpy(nx_ctx->priv.xcbc.key, in_key, key_len);
+
+       return 0;
+}
+
+static int nx_xcbc_init(struct shash_desc *desc)
+{
+       struct xcbc_state *sctx = shash_desc_ctx(desc);
+       struct nx_crypto_ctx *nx_ctx = crypto_tfm_ctx(&desc->tfm->base);
+       struct nx_csbcpb *csbcpb = nx_ctx->csbcpb;
+       struct nx_sg *out_sg;
+
+       nx_ctx_init(nx_ctx, HCOP_FC_AES);
+
+       memset(sctx, 0, sizeof *sctx);
+
+       NX_CPB_SET_KEY_SIZE(csbcpb, NX_KS_AES_128);
+       csbcpb->cpb.hdr.mode = NX_MODE_AES_XCBC_MAC;
+
+       memcpy(csbcpb->cpb.aes_xcbc.key, nx_ctx->priv.xcbc.key, AES_BLOCK_SIZE);
+       memset(nx_ctx->priv.xcbc.key, 0, sizeof *nx_ctx->priv.xcbc.key);
+
+       out_sg = nx_build_sg_list(nx_ctx->out_sg, (u8 *)sctx->state,
+                                 AES_BLOCK_SIZE, nx_ctx->ap->sglen);
+       nx_ctx->op.outlen = (nx_ctx->out_sg - out_sg) * sizeof(struct nx_sg);
+
+       return 0;
+}
+
+static int nx_xcbc_update(struct shash_desc *desc,
+                         const u8          *data,
+                         unsigned int       len)
+{
+       struct xcbc_state *sctx = shash_desc_ctx(desc);
+       struct nx_crypto_ctx *nx_ctx = crypto_tfm_ctx(&desc->tfm->base);
+       struct nx_csbcpb *csbcpb = nx_ctx->csbcpb;
+       struct nx_sg *in_sg;
+       u32 to_process, leftover;
+       int rc = 0;
+
+       if (NX_CPB_FDM(csbcpb) & NX_FDM_CONTINUATION) {
+               /* we've hit the nx chip previously and we're updating again,
+                * so copy over the partial digest */
+               memcpy(csbcpb->cpb.aes_xcbc.cv,
+                      csbcpb->cpb.aes_xcbc.out_cv_mac, AES_BLOCK_SIZE);
+       }
+
+       /* 2 cases for total data len:
+        *  1: <= AES_BLOCK_SIZE: copy into state, return 0
+        *  2: > AES_BLOCK_SIZE: process X blocks, copy in leftover
+        */
+       if (len + sctx->count <= AES_BLOCK_SIZE) {
+               memcpy(sctx->buffer + sctx->count, data, len);
+               sctx->count += len;
+               goto out;
+       }
+
+       /* to_process: the AES_BLOCK_SIZE data chunk to process in this
+        * update */
+       to_process = (sctx->count + len) & ~(AES_BLOCK_SIZE - 1);
+       leftover = (sctx->count + len) & (AES_BLOCK_SIZE - 1);
+
+       /* the hardware will not accept a 0 byte operation for this algorithm
+        * and the operation MUST be finalized to be correct. So if we happen
+        * to get an update that falls on a block sized boundary, we must
+        * save off the last block to finalize with later. */
+       if (!leftover) {
+               to_process -= AES_BLOCK_SIZE;
+               leftover = AES_BLOCK_SIZE;
+       }
+
+       if (sctx->count) {
+               in_sg = nx_build_sg_list(nx_ctx->in_sg, sctx->buffer,
+                                        sctx->count, nx_ctx->ap->sglen);
+               in_sg = nx_build_sg_list(in_sg, (u8 *)data,
+                                        to_process - sctx->count,
+                                        nx_ctx->ap->sglen);
+               nx_ctx->op.inlen = (nx_ctx->in_sg - in_sg) *
+                                       sizeof(struct nx_sg);
+       } else {
+               in_sg = nx_build_sg_list(nx_ctx->in_sg, (u8 *)data, to_process,
+                                        nx_ctx->ap->sglen);
+               nx_ctx->op.inlen = (nx_ctx->in_sg - in_sg) *
+                                       sizeof(struct nx_sg);
+       }
+
+       NX_CPB_FDM(csbcpb) |= NX_FDM_INTERMEDIATE;
+
+       if (!nx_ctx->op.inlen || !nx_ctx->op.outlen) {
+               rc = -EINVAL;
+               goto out;
+       }
+
+       rc = nx_hcall_sync(nx_ctx, &nx_ctx->op,
+                          desc->flags & CRYPTO_TFM_REQ_MAY_SLEEP);
+       if (rc)
+               goto out;
+
+       atomic_inc(&(nx_ctx->stats->aes_ops));
+
+       /* copy the leftover back into the state struct */
+       memcpy(sctx->buffer, data + len - leftover, leftover);
+       sctx->count = leftover;
+
+       /* everything after the first update is continuation */
+       NX_CPB_FDM(csbcpb) |= NX_FDM_CONTINUATION;
+out:
+       return rc;
+}
+
+static int nx_xcbc_final(struct shash_desc *desc, u8 *out)
+{
+       struct xcbc_state *sctx = shash_desc_ctx(desc);
+       struct nx_crypto_ctx *nx_ctx = crypto_tfm_ctx(&desc->tfm->base);
+       struct nx_csbcpb *csbcpb = nx_ctx->csbcpb;
+       struct nx_sg *in_sg, *out_sg;
+       int rc = 0;
+
+       if (NX_CPB_FDM(csbcpb) & NX_FDM_CONTINUATION) {
+               /* we've hit the nx chip previously, now we're finalizing,
+                * so copy over the partial digest */
+               memcpy(csbcpb->cpb.aes_xcbc.cv,
+                      csbcpb->cpb.aes_xcbc.out_cv_mac, AES_BLOCK_SIZE);
+       } else if (sctx->count == 0) {
+               /* we've never seen an update, so this is a 0 byte op. The
+                * hardware cannot handle a 0 byte op, so just copy out the
+                * known 0 byte result. This is cheaper than allocating a
+                * software context to do a 0 byte op */
+               u8 data[] = { 0x75, 0xf0, 0x25, 0x1d, 0x52, 0x8a, 0xc0, 0x1c,
+                             0x45, 0x73, 0xdf, 0xd5, 0x84, 0xd7, 0x9f, 0x29 };
+               memcpy(out, data, sizeof(data));
+               goto out;
+       }
+
+       /* final is represented by continuing the operation and indicating that
+        * this is not an intermediate operation */
+       NX_CPB_FDM(csbcpb) &= ~NX_FDM_INTERMEDIATE;
+
+       in_sg = nx_build_sg_list(nx_ctx->in_sg, (u8 *)sctx->buffer,
+                                sctx->count, nx_ctx->ap->sglen);
+       out_sg = nx_build_sg_list(nx_ctx->out_sg, out, AES_BLOCK_SIZE,
+                                 nx_ctx->ap->sglen);
+
+       nx_ctx->op.inlen = (nx_ctx->in_sg - in_sg) * sizeof(struct nx_sg);
+       nx_ctx->op.outlen = (nx_ctx->out_sg - out_sg) * sizeof(struct nx_sg);
+
+       if (!nx_ctx->op.outlen) {
+               rc = -EINVAL;
+               goto out;
+       }
+
+       rc = nx_hcall_sync(nx_ctx, &nx_ctx->op,
+                          desc->flags & CRYPTO_TFM_REQ_MAY_SLEEP);
+       if (rc)
+               goto out;
+
+       atomic_inc(&(nx_ctx->stats->aes_ops));
+
+       memcpy(out, csbcpb->cpb.aes_xcbc.out_cv_mac, AES_BLOCK_SIZE);
+out:
+       return rc;
+}
+
+struct shash_alg nx_shash_aes_xcbc_alg = {
+       .digestsize = AES_BLOCK_SIZE,
+       .init       = nx_xcbc_init,
+       .update     = nx_xcbc_update,
+       .final      = nx_xcbc_final,
+       .setkey     = nx_xcbc_set_key,
+       .descsize   = sizeof(struct xcbc_state),
+       .statesize  = sizeof(struct xcbc_state),
+       .base       = {
+               .cra_name        = "xcbc(aes)",
+               .cra_driver_name = "xcbc-aes-nx",
+               .cra_priority    = 300,
+               .cra_flags       = CRYPTO_ALG_TYPE_SHASH,
+               .cra_blocksize   = AES_BLOCK_SIZE,
+               .cra_module      = THIS_MODULE,
+               .cra_ctxsize     = sizeof(struct nx_crypto_ctx),
+               .cra_init        = nx_crypto_ctx_aes_xcbc_init,
+               .cra_exit        = nx_crypto_ctx_exit,
+       }
+};