Correct a performance bug from Bill Paul's original FreeBSD bge(4) driver:

Each call to the FreeBSD bge_start() routine the transmit producer
pointer index from the chip mailbox register BGE_MBX_TX_HOST_PROD0_LO.
The local copy of that value is then updated by bge_encap() as
bge_encap() encapsulates packets in the Tx ring. If bge_encap()
succeds in encpuslating one or more packets, bge_start() tells the
chip to start sending the newly-encinitiates writes the new value back
to the chip mailbox register.

However, comparison of the Linux drivers (Broadcom-supplied and
open-source tg3.c) and to the OpenSolaris driver confirms that
register BGE_MBX_TX_HOST_PROD0_LO is write-only to software.
Thus, we can just keep a copy in the softc, and eliminate the
(expensive) PCI register write on each call to bge_start().

From jonathan NetBSD

tested by krw@, sturm@ and I on a few different bge NICs.

2 files changed, 13 insertions, 6 deletions

diff --git a/sys/dev/pci/if_bge.c b/sys/dev/pci/if_bge.c
index de2904e4ef7..2172301074d 100644
--- a/sys/dev/pci/if_bge.c
+++ b/sys/dev/pci/if_bge.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: if_bge.c,v 1.92 2005/11/14 13:11:40 mickey Exp $	*/
+/*	$OpenBSD: if_bge.c,v 1.93 2005/11/19 23:04:48 brad Exp $	*/
 
 /*
  * Copyright (c) 2001 Wind River Systems
@@ -1058,10 +1058,14 @@ bge_init_tx_ring(struct bge_softc *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);
 	if (sc->bge_quirks & BGE_QUIRK_PRODUCER_BUG)
-		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);
 	if (sc->bge_quirks & BGE_QUIRK_PRODUCER_BUG)
 		CSR_WRITE_4(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
@@ -2805,7 +2809,7 @@ bge_start(struct ifnet *ifp)
 {
 	struct bge_softc *sc;
 	struct mbuf *m_head = NULL;
-	u_int32_t prodidx = 0;
+	u_int32_t prodidx;
 	int pkts = 0;
 
 	sc = ifp->if_softc;
@@ -2813,7 +2817,7 @@ bge_start(struct ifnet *ifp)
 	if (!sc->bge_link && ifp->if_snd.ifq_len < 10)
 		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_POLL(&ifp->if_snd, m_head);
@@ -2870,6 +2874,8 @@ bge_start(struct ifnet *ifp)
 	if (sc->bge_quirks & BGE_QUIRK_PRODUCER_BUG)
 		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/pci/if_bgereg.h b/sys/dev/pci/if_bgereg.h
index d0962d8a1a0..8ab2b8f0344 100644
--- a/sys/dev/pci/if_bgereg.h
+++ b/sys/dev/pci/if_bgereg.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: if_bgereg.h,v 1.30 2005/10/09 23:41:55 brad Exp $	*/
+/*	$OpenBSD: if_bgereg.h,v 1.31 2005/11/19 23:04:48 brad Exp $	*/
 
 /*
  * Copyright (c) 2001 Wind River Systems
@@ -2328,6 +2328,7 @@ struct bge_softc {
 	u_int16_t		bge_rx_saved_considx;
 	u_int16_t		bge_ev_saved_considx;
 	u_int16_t		bge_return_ring_cnt;
+	u_int32_t		bge_tx_prodidx;
 	u_int16_t		bge_std;	/* current std ring head */
 	u_int16_t		bge_jumbo;	/* current jumo ring head */
 	SLIST_HEAD(__bge_jfreehead, bge_jpool_entry)	bge_jfree_listhead;