commit 41b976414c88016e2c9d9b2f6667ee67a998d388 Author: David Woodhouse Date: Wed Sep 23 09:45:31 2015 +0100 8139cp: Dump contents of descriptor ring on TX timeout We are seeing unexplained TX timeouts under heavy load. Let's try to get a better idea of what's going on. Signed-off-by: David Woodhouse Signed-off-by: David S. Miller commit 7f4c685633e2df9ba10d49a31dda13715745db37 Author: David Woodhouse Date: Wed Sep 23 09:45:16 2015 +0100 8139cp: Fix DMA unmapping of transmitted buffers The low 16 bits of the 'opts1' field in the TX descriptor are supposed to still contain the buffer length when the descriptor is handed back to us. In practice, at least on my hardware, they don't. So stash the original value of the opts1 field and get the length to unmap from there. There are other ways we could have worked out the length, but I actually want a stash of the opts1 field anyway so that I can dump it alongside the contents of the descriptor ring when we suffer a TX timeout. Signed-off-by: David Woodhouse Signed-off-by: David S. Miller commit 0a5aeee0b79fa99d8e04c98dd4e87d4f52aa497b Author: David Woodhouse Date: Wed Sep 23 09:44:57 2015 +0100 8139cp: Reduce duplicate csum/tso code in cp_start_xmit() We calculate the value of the opts1 descriptor field in three different places. With two different behaviours when given an invalid packet to be checksummed — none of them correct. Sort that out. Signed-off-by: David Woodhouse Signed-off-by: David S. Miller commit a3b804043f490aeec57d8ca5baccdd35e6250857 Author: David Woodhouse Date: Wed Sep 23 09:44:38 2015 +0100 8139cp: Fix TSO/scatter-gather descriptor setup When sending a TSO frame in multiple buffers, we were neglecting to set the first descriptor up in TSO mode. Signed-off-by: David Woodhouse Signed-off-by: David S. Miller commit 26b0bad6ac3a0167792dc4ffb276c29bc597d239 Author: David Woodhouse Date: Wed Sep 23 09:44:06 2015 +0100 8139cp: Fix tx_queued debug message to print correct slot numbers After a certain amount of staring at the debug output of this driver, I realised it was lying to me. Signed-off-by: David Woodhouse Signed-off-by: David S. Miller commit aaa0062ecf4877a26dea66bee1039c6eaf906c94 Author: David Woodhouse Date: Wed Sep 23 09:43:41 2015 +0100 8139cp: Do not re-enable RX interrupts in cp_tx_timeout() If an RX interrupt was already received but NAPI has not yet run when the RX timeout happens, we end up in cp_tx_timeout() with RX interrupts already disabled. Blindly re-enabling them will cause an IRQ storm. (This is made particularly horrid by the fact that cp_interrupt() always returns that it's handled the interrupt, even when it hasn't actually done anything. If it didn't do that, the core IRQ code would have detected the storm and handled it, I'd have had a clear smoking gun backtrace instead of just a spontaneously resetting router, and I'd have at *least* two days of my life back. Changing the return value of cp_interrupt() will be argued about under separate cover.) Unconditionally leave RX interrupts disabled after the reset, and schedule NAPI to check the receive ring and re-enable them. Signed-off-by: David Woodhouse Signed-off-by: David S. Miller commit 7a8a8e75d505147358b225173e890ada43a267e2 Author: David Woodhouse Date: Fri Sep 18 00:21:54 2015 +0100 8139cp: Call __cp_set_rx_mode() from cp_tx_timeout() Unless we reset the RX config, on real hardware I don't seem to receive any packets after a TX timeout. Signed-off-by: David Woodhouse Signed-off-by: David S. Miller commit fc27bd115b334e3ebdc682a42a47c3aea2566dcc Author: David Woodhouse Date: Fri Sep 18 00:19:08 2015 +0100 8139cp: Use dev_kfree_skb_any() instead of dev_kfree_skb() in cp_clean_rings() This can be called from cp_tx_timeout() with interrupts disabled. Spotted by Francois Romieu Signed-off-by: David Woodhouse Signed-off-by: David S. Miller --- a/drivers/net/ethernet/realtek/8139cp.c +++ b/drivers/net/ethernet/realtek/8139cp.c @@ -157,6 +157,7 @@ enum { NWayAdvert = 0x66, /* MII ADVERTISE */ NWayLPAR = 0x68, /* MII LPA */ NWayExpansion = 0x6A, /* MII Expansion */ + TxDmaOkLowDesc = 0x82, /* Low 16 bit address of a Tx descriptor. */ Config5 = 0xD8, /* Config5 */ TxPoll = 0xD9, /* Tell chip to check Tx descriptors for work */ RxMaxSize = 0xDA, /* Max size of an Rx packet (8169 only) */ @@ -341,6 +342,7 @@ struct cp_private { unsigned tx_tail; struct cp_desc *tx_ring; struct sk_buff *tx_skb[CP_TX_RING_SIZE]; + u32 tx_opts[CP_TX_RING_SIZE]; unsigned rx_buf_sz; unsigned wol_enabled : 1; /* Is Wake-on-LAN enabled? */ @@ -665,7 +667,7 @@ static void cp_tx (struct cp_private *cp BUG_ON(!skb); dma_unmap_single(&cp->pdev->dev, le64_to_cpu(txd->addr), - le32_to_cpu(txd->opts1) & 0xffff, + cp->tx_opts[tx_tail] & 0xffff, PCI_DMA_TODEVICE); if (status & LastFrag) { @@ -733,7 +735,7 @@ static netdev_tx_t cp_start_xmit (struct { struct cp_private *cp = netdev_priv(dev); unsigned entry; - u32 eor, flags; + u32 eor, opts1; unsigned long intr_flags; __le32 opts2; int mss = 0; @@ -753,6 +755,21 @@ static netdev_tx_t cp_start_xmit (struct mss = skb_shinfo(skb)->gso_size; opts2 = cpu_to_le32(cp_tx_vlan_tag(skb)); + opts1 = DescOwn; + if (mss) + opts1 |= LargeSend | ((mss & MSSMask) << MSSShift); + else if (skb->ip_summed == CHECKSUM_PARTIAL) { + const struct iphdr *ip = ip_hdr(skb); + if (ip->protocol == IPPROTO_TCP) + opts1 |= IPCS | TCPCS; + else if (ip->protocol == IPPROTO_UDP) + opts1 |= IPCS | UDPCS; + else { + WARN_ONCE(1, + "Net bug: asked to checksum invalid Legacy IP packet\n"); + goto out_dma_error; + } + } if (skb_shinfo(skb)->nr_frags == 0) { struct cp_desc *txd = &cp->tx_ring[entry]; @@ -768,31 +785,20 @@ static netdev_tx_t cp_start_xmit (struct txd->addr = cpu_to_le64(mapping); wmb(); - flags = eor | len | DescOwn | FirstFrag | LastFrag; - - if (mss) - flags |= LargeSend | ((mss & MSSMask) << MSSShift); - else if (skb->ip_summed == CHECKSUM_PARTIAL) { - const struct iphdr *ip = ip_hdr(skb); - if (ip->protocol == IPPROTO_TCP) - flags |= IPCS | TCPCS; - else if (ip->protocol == IPPROTO_UDP) - flags |= IPCS | UDPCS; - else - WARN_ON(1); /* we need a WARN() */ - } + opts1 |= eor | len | FirstFrag | LastFrag; - txd->opts1 = cpu_to_le32(flags); + txd->opts1 = cpu_to_le32(opts1); wmb(); cp->tx_skb[entry] = skb; - entry = NEXT_TX(entry); + cp->tx_opts[entry] = opts1; + netif_dbg(cp, tx_queued, cp->dev, "tx queued, slot %d, skblen %d\n", + entry, skb->len); } else { struct cp_desc *txd; - u32 first_len, first_eor; + u32 first_len, first_eor, ctrl; dma_addr_t first_mapping; int frag, first_entry = entry; - const struct iphdr *ip = ip_hdr(skb); /* We must give this initial chunk to the device last. * Otherwise we could race with the device. @@ -805,14 +811,14 @@ static netdev_tx_t cp_start_xmit (struct goto out_dma_error; cp->tx_skb[entry] = skb; - entry = NEXT_TX(entry); for (frag = 0; frag < skb_shinfo(skb)->nr_frags; frag++) { const skb_frag_t *this_frag = &skb_shinfo(skb)->frags[frag]; u32 len; - u32 ctrl; dma_addr_t mapping; + entry = NEXT_TX(entry); + len = skb_frag_size(this_frag); mapping = dma_map_single(&cp->pdev->dev, skb_frag_address(this_frag), @@ -824,19 +830,7 @@ static netdev_tx_t cp_start_xmit (struct eor = (entry == (CP_TX_RING_SIZE - 1)) ? RingEnd : 0; - ctrl = eor | len | DescOwn; - - if (mss) - ctrl |= LargeSend | - ((mss & MSSMask) << MSSShift); - else if (skb->ip_summed == CHECKSUM_PARTIAL) { - if (ip->protocol == IPPROTO_TCP) - ctrl |= IPCS | TCPCS; - else if (ip->protocol == IPPROTO_UDP) - ctrl |= IPCS | UDPCS; - else - BUG(); - } + ctrl = opts1 | eor | len; if (frag == skb_shinfo(skb)->nr_frags - 1) ctrl |= LastFrag; @@ -849,8 +843,8 @@ static netdev_tx_t cp_start_xmit (struct txd->opts1 = cpu_to_le32(ctrl); wmb(); + cp->tx_opts[entry] = ctrl; cp->tx_skb[entry] = skb; - entry = NEXT_TX(entry); } txd = &cp->tx_ring[first_entry]; @@ -858,27 +852,17 @@ static netdev_tx_t cp_start_xmit (struct txd->addr = cpu_to_le64(first_mapping); wmb(); - if (skb->ip_summed == CHECKSUM_PARTIAL) { - if (ip->protocol == IPPROTO_TCP) - txd->opts1 = cpu_to_le32(first_eor | first_len | - FirstFrag | DescOwn | - IPCS | TCPCS); - else if (ip->protocol == IPPROTO_UDP) - txd->opts1 = cpu_to_le32(first_eor | first_len | - FirstFrag | DescOwn | - IPCS | UDPCS); - else - BUG(); - } else - txd->opts1 = cpu_to_le32(first_eor | first_len | - FirstFrag | DescOwn); + ctrl = opts1 | first_eor | first_len | FirstFrag; + txd->opts1 = cpu_to_le32(ctrl); wmb(); + + cp->tx_opts[first_entry] = ctrl; + netif_dbg(cp, tx_queued, cp->dev, "tx queued, slots %d-%d, skblen %d\n", + first_entry, entry, skb->len); } - cp->tx_head = entry; + cp->tx_head = NEXT_TX(entry); netdev_sent_queue(dev, skb->len); - netif_dbg(cp, tx_queued, cp->dev, "tx queued, slot %d, skblen %d\n", - entry, skb->len); if (TX_BUFFS_AVAIL(cp) <= (MAX_SKB_FRAGS + 1)) netif_stop_queue(dev); @@ -1115,6 +1099,7 @@ static int cp_init_rings (struct cp_priv { memset(cp->tx_ring, 0, sizeof(struct cp_desc) * CP_TX_RING_SIZE); cp->tx_ring[CP_TX_RING_SIZE - 1].opts1 = cpu_to_le32(RingEnd); + memset(cp->tx_opts, 0, sizeof(cp->tx_opts)); cp_init_rings_index(cp); @@ -1151,7 +1136,7 @@ static void cp_clean_rings (struct cp_pr desc = cp->rx_ring + i; dma_unmap_single(&cp->pdev->dev,le64_to_cpu(desc->addr), cp->rx_buf_sz, PCI_DMA_FROMDEVICE); - dev_kfree_skb(cp->rx_skb[i]); + dev_kfree_skb_any(cp->rx_skb[i]); } } @@ -1164,7 +1149,7 @@ static void cp_clean_rings (struct cp_pr le32_to_cpu(desc->opts1) & 0xffff, PCI_DMA_TODEVICE); if (le32_to_cpu(desc->opts1) & LastFrag) - dev_kfree_skb(skb); + dev_kfree_skb_any(skb); cp->dev->stats.tx_dropped++; } } @@ -1172,6 +1157,7 @@ static void cp_clean_rings (struct cp_pr memset(cp->rx_ring, 0, sizeof(struct cp_desc) * CP_RX_RING_SIZE); memset(cp->tx_ring, 0, sizeof(struct cp_desc) * CP_TX_RING_SIZE); + memset(cp->tx_opts, 0, sizeof(cp->tx_opts)); memset(cp->rx_skb, 0, sizeof(struct sk_buff *) * CP_RX_RING_SIZE); memset(cp->tx_skb, 0, sizeof(struct sk_buff *) * CP_TX_RING_SIZE); @@ -1249,7 +1235,7 @@ static void cp_tx_timeout(struct net_dev { struct cp_private *cp = netdev_priv(dev); unsigned long flags; - int rc; + int rc, i; netdev_warn(dev, "Transmit timeout, status %2x %4x %4x %4x\n", cpr8(Cmd), cpr16(CpCmd), @@ -1257,13 +1243,26 @@ static void cp_tx_timeout(struct net_dev spin_lock_irqsave(&cp->lock, flags); + netif_dbg(cp, tx_err, cp->dev, "TX ring head %d tail %d desc %x\n", + cp->tx_head, cp->tx_tail, cpr16(TxDmaOkLowDesc)); + for (i = 0; i < CP_TX_RING_SIZE; i++) { + netif_dbg(cp, tx_err, cp->dev, + "TX slot %d @%p: %08x (%08x) %08x %llx %p\n", + i, &cp->tx_ring[i], le32_to_cpu(cp->tx_ring[i].opts1), + cp->tx_opts[i], le32_to_cpu(cp->tx_ring[i].opts2), + le64_to_cpu(cp->tx_ring[i].addr), + cp->tx_skb[i]); + } + cp_stop_hw(cp); cp_clean_rings(cp); rc = cp_init_rings(cp); cp_start_hw(cp); - cp_enable_irq(cp); + __cp_set_rx_mode(dev); + cpw16_f(IntrMask, cp_norx_intr_mask); netif_wake_queue(dev); + napi_schedule_irqoff(&cp->napi); spin_unlock_irqrestore(&cp->lock, flags); }