From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jeff Garzik Subject: [PATCH] tg3 shutdown sequence update Date: Mon, 25 Nov 2002 01:34:16 -0500 Sender: netdev-bounce@oss.sgi.com Message-ID: <3DE1C468.1060806@pobox.com> Mime-Version: 1.0 Content-Type: multipart/mixed; boundary="------------040304080808010802090102" Cc: "David S. Miller" Return-path: To: netdev@oss.sgi.com Errors-to: netdev-bounce@oss.sgi.com List-Id: netdev.vger.kernel.org This is a multi-part message in MIME format. --------------040304080808010802090102 Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit This patch is only for testing and comment, _not_ for applying. (testers requested) The attached patch updates the tg3 net driver shutdown sequence to be a bit more correct WRT the documented sequence. Detailed changes: * bail out of tg3_stop_block ASAP if block is already disabled * use standard tg3_stop_block to disable RX MAC mode. this adds polling of the the enable bit to the standard code. * just in case, shut down DMA completion between send data completion shutdown and send DB completion shutdown * use standard tg3_stop_block to disable TX MAC mode. * don't bother to disable MAC_MODE_TDE_ENABLE bit manually, TX MAC mode disable does it for us. * add PCI posting flush for flow-through queues Does this look ok WRT errata and hardware seen in the field? One potential concern is that the error handling if tg3_stop_block fails runs through all the blocks unconditionally, and then returns an error. It does not bail early if some of the stop-block calls fail. Comments? Jeff --------------040304080808010802090102 Content-Type: text/plain; name="patch" Content-Transfer-Encoding: 7bit Content-Disposition: inline; filename="patch" --- drivers/net/tg3.c.~1~ Tue Nov 19 15:30:41 2002 +++ drivers/net/tg3.c Tue Nov 19 15:31:34 2002 @@ -2367,6 +2367,7 @@ unsigned int i; u32 len, entry, base_flags, mss; int would_hit_hwbug; + unsigned long flags; len = (skb->len - skb->data_len); @@ -2389,12 +2390,12 @@ * So we really do need to disable interrupts when taking * tx_lock here. */ - spin_lock_irq(&tp->tx_lock); + spin_lock_irqsave(&tp->tx_lock, flags); /* This is a hard error, log it. */ if (unlikely(TX_BUFFS_AVAIL(tp) <= (skb_shinfo(skb)->nr_frags + 1))) { netif_stop_queue(dev); - spin_unlock_irq(&tp->tx_lock); + spin_unlock_irqrestore(&tp->tx_lock, flags); printk(KERN_ERR PFX "%s: BUG! Tx Ring full when queue awake!\n", dev->name); return 1; @@ -2535,7 +2536,7 @@ netif_stop_queue(dev); out_unlock: - spin_unlock_irq(&tp->tx_lock); + spin_unlock_irqrestore(&tp->tx_lock, flags); dev->trans_start = jiffies; @@ -2547,6 +2548,7 @@ struct tg3 *tp = dev->priv; dma_addr_t mapping; u32 len, entry, base_flags, mss; + unsigned long flags; len = (skb->len - skb->data_len); @@ -2569,12 +2571,12 @@ * So we really do need to disable interrupts when taking * tx_lock here. */ - spin_lock_irq(&tp->tx_lock); + spin_lock_irqsave(&tp->tx_lock, flags); /* This is a hard error, log it. */ if (unlikely(TX_BUFFS_AVAIL(tp) <= (skb_shinfo(skb)->nr_frags + 1))) { netif_stop_queue(dev); - spin_unlock_irq(&tp->tx_lock); + spin_unlock_irqrestore(&tp->tx_lock, flags); printk(KERN_ERR PFX "%s: BUG! Tx Ring full when queue awake!\n", dev->name); return 1; @@ -2665,7 +2667,7 @@ if (TX_BUFFS_AVAIL(tp) <= (MAX_SKB_FRAGS + 1)) netif_stop_queue(dev); - spin_unlock_irq(&tp->tx_lock); + spin_unlock_irqrestore(&tp->tx_lock, flags); dev->trans_start = jiffies; --------------040304080808010802090102--