Since BGE_MBX_TX_HOST_PROD0_LO register is write-only to software,

we can cache its value in the softc. Eliminates one PCI register
write per call to bge_start().

A 1.8% speedup for UDP_RR test on my old box.

Obtained from:	NetBSD(jonathan) via delphij

2 files changed, 11 insertions, 4 deletions

diff --git a/sys/dev/bge/if_bge.c b/sys/dev/bge/if_bge.c
index 16fc559..e380e26 100644
--- a/sys/dev/bge/if_bge.c
+++ b/sys/dev/bge/if_bge.c
@@ -947,11 +947,15 @@ bge_init_tx_ring(sc)
 	sc->bge_txcnt = 0;
 	sc->bge_tx_saved_considx = 0;
 
-	CSR_WRITE_4(sc, BGE_MBX_TX_HOST_PROD0_LO, 0);
+	/* Initialize transmit producer index for host-memory send ring. */
+	sc->bge_tx_prodidx = 0;
+	CSR_WRITE_4(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
+
 	/* 5700 b2 errata */
 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
-		CSR_WRITE_4(sc, BGE_MBX_TX_HOST_PROD0_LO, 0);
+		CSR_WRITE_4(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx);
 
+	/* NIC-memory send ring not used; initialize to zero. */
 	CSR_WRITE_4(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
 	/* 5700 b2 errata */
 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
@@ -3040,7 +3044,7 @@ bge_start_locked(ifp)
 {
 	struct bge_softc *sc;
 	struct mbuf *m_head = NULL;
-	u_int32_t prodidx = 0;
+	uint32_t prodidx;
 	int count = 0;
 
 	sc = ifp->if_softc;
@@ -3048,7 +3052,7 @@ bge_start_locked(ifp)
 	if (!sc->bge_link || IFQ_DRV_IS_EMPTY(&ifp->if_snd))
 		return;
 
-	prodidx = CSR_READ_4(sc, BGE_MBX_TX_HOST_PROD0_LO);
+	prodidx = sc->bge_tx_prodidx;
 
 	while(sc->bge_cdata.bge_tx_chain[prodidx] == NULL) {
 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
@@ -3108,6 +3112,8 @@ bge_start_locked(ifp)
 	if (sc->bge_chiprev == BGE_CHIPREV_5700_BX)
 		CSR_WRITE_4(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
 
+	sc->bge_tx_prodidx = prodidx;
+
 	/*
 	 * Set a timeout in case the chip goes out to lunch.
 	 */
diff --git a/sys/dev/bge/if_bgereg.h b/sys/dev/bge/if_bgereg.h
index e894e63..9deb959 100644
--- a/sys/dev/bge/if_bgereg.h
+++ b/sys/dev/bge/if_bgereg.h
@@ -2416,6 +2416,7 @@ struct bge_softc {
 	u_int32_t		bge_stat_ticks;
 	u_int32_t		bge_rx_coal_ticks;
 	u_int32_t		bge_tx_coal_ticks;
+	u_int32_t		bge_tx_prodidx;
 	u_int32_t		bge_rx_max_coal_bds;
 	u_int32_t		bge_tx_max_coal_bds;
 	u_int32_t		bge_tx_buf_ratio;