crypto: inside-secure - Base RD fetchcount on actual RD FIFO size

This patch derives the result descriptor fetch count from the actual
FIFO size advertised by the hardware. Fetching result descriptors
one at a time is a performance bottleneck for small blocks, especially
on hardware with multiple pipes. Even moreso if the HW has few rings.

Signed-off-by: Pascal van Leeuwen <pvanleeuwen@verimatrix.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

2 files changed, 40 insertions, 12 deletions

diff --git a/drivers/crypto/inside-secure/safexcel.c b/drivers/crypto/inside-secure/safexcel.c
index 32366f703332..acf26dfa7e94 100644
--- a/drivers/crypto/inside-secure/safexcel.c
+++ b/drivers/crypto/inside-secure/safexcel.c
@@ -357,13 +357,22 @@ static int safexcel_hw_setup_cdesc_rings(struct safexcel_crypto_priv *priv)
 static int safexcel_hw_setup_rdesc_rings(struct safexcel_crypto_priv *priv)
 {
 	u32 hdw, rd_size_rnd, val;
-	int i;
-
-	hdw = readl(EIP197_HIA_AIC_G(priv) + EIP197_HIA_OPTIONS);
-	hdw &= GENMASK(27, 25);
-	hdw >>= 25;
+	int i, rd_fetch_cnt;
 
-	rd_size_rnd = (priv->config.rd_size + (BIT(hdw) - 1)) >> hdw;
+	/* determine number of RD's we can fetch into the FIFO as one block */
+	rd_size_rnd = (EIP197_RD64_FETCH_SIZE +
+		      BIT(priv->hwconfig.hwdataw) - 1) >>
+		      priv->hwconfig.hwdataw;
+	if (priv->flags & SAFEXCEL_HW_EIP197) {
+		/* EIP197: try to fetch enough in 1 go to keep all pipes busy */
+		rd_fetch_cnt = (1 << priv->hwconfig.hwrfsize) / rd_size_rnd;
+		rd_fetch_cnt = min_t(uint, rd_fetch_cnt,
+				     (priv->config.pes * EIP197_FETCH_DEPTH));
+	} else {
+		/* for the EIP97, just fetch all that fits minus 1 */
+		rd_fetch_cnt = ((1 << priv->hwconfig.hwrfsize) /
+			       rd_size_rnd) - 1;
+	}
 
 	for (i = 0; i < priv->config.rings; i++) {
 		/* ring base address */
@@ -376,8 +385,8 @@ static int safexcel_hw_setup_rdesc_rings(struct safexcel_crypto_priv *priv)
 		       priv->config.rd_size,
 		       EIP197_HIA_RDR(priv, i) + EIP197_HIA_xDR_DESC_SIZE);
 
-		writel(((EIP197_FETCH_COUNT * (rd_size_rnd << hdw)) << 16) |
-		       (EIP197_FETCH_COUNT * priv->config.rd_offset),
+		writel(((rd_fetch_cnt * (rd_size_rnd << hdw)) << 16) |
+		       (rd_fetch_cnt * priv->config.rd_offset),
 		       EIP197_HIA_RDR(priv, i) + EIP197_HIA_xDR_CFG);
 
 		/* Configure DMA tx control */
@@ -1244,12 +1253,17 @@ static int safexcel_probe_generic(void *pdev,
 		priv->hwconfig.hwcfsize = ((hiaopt >> EIP197_CFSIZE_OFFSET) &
 					   EIP197_CFSIZE_MASK) +
 					  EIP197_CFSIZE_ADJUST;
+		priv->hwconfig.hwrfsize = ((hiaopt >> EIP197_RFSIZE_OFFSET) &
+					   EIP197_RFSIZE_MASK) +
+					  EIP197_RFSIZE_ADJUST;
 	} else {
 		/* EIP97 */
 		priv->hwconfig.hwdataw  = (hiaopt >> EIP197_HWDATAW_OFFSET) &
 					  EIP97_HWDATAW_MASK;
 		priv->hwconfig.hwcfsize = (hiaopt >> EIP97_CFSIZE_OFFSET) &
 					  EIP97_CFSIZE_MASK;
+		priv->hwconfig.hwrfsize = (hiaopt >> EIP97_RFSIZE_OFFSET) &
+					  EIP97_RFSIZE_MASK;
 	}
 
 	/* Get supported algorithms from EIP96 transform engine */
@@ -1257,10 +1271,11 @@ static int safexcel_probe_generic(void *pdev,
 				    EIP197_PE_EIP96_OPTIONS(0));
 
 	/* Print single info line describing what we just detected */
-	dev_info(priv->dev, "EIP%d:%x(%d)-HIA:%x(%d,%d),PE:%x,alg:%08x\n", peid,
-		 priv->hwconfig.hwver, hwctg, priv->hwconfig.hiaver,
+	dev_info(priv->dev, "EIP%d:%x(%d)-HIA:%x(%d,%d,%d),PE:%x,alg:%08x\n",
+		 peid, priv->hwconfig.hwver, hwctg, priv->hwconfig.hiaver,
 		 priv->hwconfig.hwdataw, priv->hwconfig.hwcfsize,
-		 priv->hwconfig.pever, priv->hwconfig.algo_flags);
+		 priv->hwconfig.hwrfsize, priv->hwconfig.pever,
+		 priv->hwconfig.algo_flags);
 
 	safexcel_configure(priv);
 
diff --git a/drivers/crypto/inside-secure/safexcel.h b/drivers/crypto/inside-secure/safexcel.h
index 2da757d5b404..7d740b31c056 100644
--- a/drivers/crypto/inside-secure/safexcel.h
+++ b/drivers/crypto/inside-secure/safexcel.h
@@ -30,7 +30,6 @@
 #define EIP197_DEFAULT_RING_SIZE		400
 #define EIP197_MAX_TOKENS			18
 #define EIP197_MAX_RINGS			4
-#define EIP197_FETCH_COUNT			1
 #define EIP197_FETCH_DEPTH			2
 #define EIP197_MAX_BATCH_SZ			64
 
@@ -234,6 +233,11 @@
 #define EIP97_CFSIZE_OFFSET			8
 #define EIP197_CFSIZE_MASK			GENMASK(3, 0)
 #define EIP97_CFSIZE_MASK			GENMASK(4, 0)
+#define EIP197_RFSIZE_OFFSET			12
+#define EIP197_RFSIZE_ADJUST			4
+#define EIP97_RFSIZE_OFFSET			12
+#define EIP197_RFSIZE_MASK			GENMASK(3, 0)
+#define EIP97_RFSIZE_MASK			GENMASK(4, 0)
 
 /* EIP197_HIA_AIC_R_ENABLE_CTRL */
 #define EIP197_CDR_IRQ(n)			BIT((n) * 2)
@@ -462,6 +466,14 @@ struct safexcel_result_desc {
 	struct result_data_desc result_data;
 } __packed;
 
+/*
+ * The EIP(1)97 only needs to fetch the descriptor part of
+ * the result descriptor, not the result token part!
+ */
+#define EIP197_RD64_FETCH_SIZE		((sizeof(struct safexcel_result_desc) -\
+					  sizeof(struct result_data_desc)) /\
+					 sizeof(u32))
+
 struct safexcel_token {
 	u32 packet_length:17;
 	u8 stat:2;
@@ -691,6 +703,7 @@ struct safexcel_hwconfig {
 	int pever;
 	int hwdataw;
 	int hwcfsize;
+	int hwrfsize;
 };
 
 struct safexcel_crypto_priv {