crypto: inside-secure - move cache result dma mapping to request

In heavy traffic the DMA mapping is overwritten by multiple requests as
the DMA address is stored in a global context. This patch moves this
information to the per-hash request context so that it can't be
overwritten.

Fixes: 1b44c5a60c ("crypto: inside-secure - add SafeXcel EIP197 crypto engine driver")
Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

drivers/crypto/inside-secure/safexcel.c

@@ -537,20 +537,6 @@ void safexcel_dequeue(struct safexcel_crypto_priv *priv, int ring)
 	       EIP197_HIA_CDR(priv, ring) + EIP197_HIA_xDR_PREP_COUNT);
 }
 
-void safexcel_free_context(struct safexcel_crypto_priv *priv,
-			   struct crypto_async_request *req)
-{
-	struct safexcel_context *ctx = crypto_tfm_ctx(req->tfm);
-
-	if (ctx->cache) {
-		dma_unmap_single(priv->dev, ctx->cache_dma, ctx->cache_sz,
-				 DMA_TO_DEVICE);
-		kfree(ctx->cache);
-		ctx->cache = NULL;
-		ctx->cache_sz = 0;
-	}
-}
-
 void safexcel_complete(struct safexcel_crypto_priv *priv, int ring)
 {
 	struct safexcel_command_desc *cdesc;

drivers/crypto/inside-secure/safexcel.h

@@ -578,11 +578,6 @@ struct safexcel_context {
 	int ring;
 	bool needs_inv;
 	bool exit_inv;
-
-	/* Used for ahash requests */
-	void *cache;
-	dma_addr_t cache_dma;
-	unsigned int cache_sz;
 };
 
 /*
@@ -606,8 +601,6 @@ struct safexcel_inv_result {
 
 void safexcel_dequeue(struct safexcel_crypto_priv *priv, int ring);
 void safexcel_complete(struct safexcel_crypto_priv *priv, int ring);
-void safexcel_free_context(struct safexcel_crypto_priv *priv,
-				  struct crypto_async_request *req);
 int safexcel_invalidate_cache(struct crypto_async_request *async,
 			      struct safexcel_crypto_priv *priv,
 			      dma_addr_t ctxr_dma, int ring,

drivers/crypto/inside-secure/safexcel_hash.c

@@ -43,6 +43,9 @@ struct safexcel_ahash_req {
 	u64 processed;
 
 	u8 cache[SHA256_BLOCK_SIZE] __aligned(sizeof(u32));
+	dma_addr_t cache_dma;
+	unsigned int cache_sz;
+
 	u8 cache_next[SHA256_BLOCK_SIZE] __aligned(sizeof(u32));
 };
 
@@ -165,7 +168,11 @@ static int safexcel_handle_req_result(struct safexcel_crypto_priv *priv, int rin
 		sreq->result_dma = 0;
 	}
 
-	safexcel_free_context(priv, async);
+	if (sreq->cache_dma) {
+		dma_unmap_single(priv->dev, sreq->cache_dma, sreq->cache_sz,
+				 DMA_TO_DEVICE);
+		sreq->cache_dma = 0;
+	}
 
 	cache_len = sreq->len - sreq->processed;
 	if (cache_len)
@@ -227,24 +234,15 @@ static int safexcel_ahash_send_req(struct crypto_async_request *async, int ring,
 
 	/* Add a command descriptor for the cached data, if any */
 	if (cache_len) {
-		ctx->base.cache = kzalloc(cache_len, EIP197_GFP_FLAGS(*async));
+		req->cache_dma = dma_map_single(priv->dev, req->cache,
-		if (!ctx->base.cache) {
-			ret = -ENOMEM;
-			goto unlock;
-		}
-		memcpy(ctx->base.cache, req->cache, cache_len);
-		ctx->base.cache_dma = dma_map_single(priv->dev, ctx->base.cache,
 						cache_len, DMA_TO_DEVICE);
-		if (dma_mapping_error(priv->dev, ctx->base.cache_dma)) {
+		if (dma_mapping_error(priv->dev, req->cache_dma))
-			ret = -EINVAL;
+			return -EINVAL;
-			goto free_cache;
-		}
 
-		ctx->base.cache_sz = cache_len;
+		req->cache_sz = cache_len;
 		first_cdesc = safexcel_add_cdesc(priv, ring, 1,
 						 (cache_len == len),
-						 ctx->base.cache_dma,
+						 req->cache_dma, cache_len, len,
-						 cache_len, len,
 						 ctx->base.ctxr_dma);
 		if (IS_ERR(first_cdesc)) {
 			ret = PTR_ERR(first_cdesc);
@@ -328,16 +326,12 @@ static int safexcel_ahash_send_req(struct crypto_async_request *async, int ring,
 	for (i = 0; i < n_cdesc; i++)
 		safexcel_ring_rollback_wptr(priv, &priv->ring[ring].cdr);
 unmap_cache:
-	if (ctx->base.cache_dma) {
+	if (req->cache_dma) {
-		dma_unmap_single(priv->dev, ctx->base.cache_dma,
+		dma_unmap_single(priv->dev, req->cache_dma, req->cache_sz,
-				 ctx->base.cache_sz, DMA_TO_DEVICE);
+				 DMA_TO_DEVICE);
-		ctx->base.cache_sz = 0;
+		req->cache_sz = 0;
 	}
-free_cache:
-	kfree(ctx->base.cache);
-	ctx->base.cache = NULL;
 
-unlock:
 	spin_unlock_bh(&priv->ring[ring].egress_lock);
 	return ret;
 }