queue-4.4/e1000e-fix-cyclic-resets-at-link-up-with-active-tx.patch

   1 From e360c5a03af43ed5858cee18981560df58e39204 Mon Sep 17 00:00:00 2001
   2 From: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
   3 Date: Mon, 14 Jan 2019 16:29:30 +0300
   4 Subject: e1000e: fix cyclic resets at link up with active tx
   5
   6 [ Upstream commit 0f9e980bf5ee1a97e2e401c846b2af989eb21c61 ]
   7
   8 I'm seeing series of e1000e resets (sometimes endless) at system boot
   9 if something generates tx traffic at this time. In my case this is
  10 netconsole who sends message "e1000e 0000:02:00.0: Some CPU C-states
  11 have been disabled in order to enable jumbo frames" from e1000e itself.
  12 As result e1000_watchdog_task sees used tx buffer while carrier is off
  13 and start this reset cycle again.
  14
  15 [   17.794359] e1000e: eth1 NIC Link is Up 1000 Mbps Full Duplex, Flow Control: None
  16 [   17.794714] IPv6: ADDRCONF(NETDEV_CHANGE): eth1: link becomes ready
  17 [   22.936455] e1000e 0000:02:00.0 eth1: changing MTU from 1500 to 9000
  18 [   23.033336] e1000e 0000:02:00.0: Some CPU C-states have been disabled in order to enable jumbo frames
  19 [   26.102364] e1000e: eth1 NIC Link is Up 1000 Mbps Full Duplex, Flow Control: None
  20 [   27.174495] 8021q: 802.1Q VLAN Support v1.8
  21 [   27.174513] 8021q: adding VLAN 0 to HW filter on device eth1
  22 [   30.671724] cgroup: cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation
  23 [   30.898564] netpoll: netconsole: local port 6666
  24 [   30.898566] netpoll: netconsole: local IPv6 address 2a02:6b8:0:80b:beae:c5ff:fe28:23f8
  25 [   30.898567] netpoll: netconsole: interface 'eth1'
  26 [   30.898568] netpoll: netconsole: remote port 6666
  27 [   30.898568] netpoll: netconsole: remote IPv6 address 2a02:6b8:b000:605c:e61d:2dff:fe03:3790
  28 [   30.898569] netpoll: netconsole: remote ethernet address b0:a8:6e:f4:ff:c0
  29 [   30.917747] console [netcon0] enabled
  30 [   30.917749] netconsole: network logging started
  31 [   31.453353] e1000e 0000:02:00.0: Some CPU C-states have been disabled in order to enable jumbo frames
  32 [   34.185730] e1000e 0000:02:00.0: Some CPU C-states have been disabled in order to enable jumbo frames
  33 [   34.321840] e1000e 0000:02:00.0: Some CPU C-states have been disabled in order to enable jumbo frames
  34 [   34.465822] e1000e 0000:02:00.0: Some CPU C-states have been disabled in order to enable jumbo frames
  35 [   34.597423] e1000e 0000:02:00.0: Some CPU C-states have been disabled in order to enable jumbo frames
  36 [   34.745417] e1000e 0000:02:00.0: Some CPU C-states have been disabled in order to enable jumbo frames
  37 [   34.877356] e1000e 0000:02:00.0: Some CPU C-states have been disabled in order to enable jumbo frames
  38 [   35.005441] e1000e 0000:02:00.0: Some CPU C-states have been disabled in order to enable jumbo frames
  39 [   35.157376] e1000e 0000:02:00.0: Some CPU C-states have been disabled in order to enable jumbo frames
  40 [   35.289362] e1000e 0000:02:00.0: Some CPU C-states have been disabled in order to enable jumbo frames
  41 [   35.417441] e1000e 0000:02:00.0: Some CPU C-states have been disabled in order to enable jumbo frames
  42 [   37.790342] e1000e: eth1 NIC Link is Up 1000 Mbps Full Duplex, Flow Control: None
  43
  44 This patch flushes tx buffers only once when carrier is off
  45 rather than at each watchdog iteration.
  46
  47 Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
  48 Tested-by: Aaron Brown <aaron.f.brown@intel.com>
  49 Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
  50 Signed-off-by: Sasha Levin <sashal@kernel.org>
  51 ---
  52  drivers/net/ethernet/intel/e1000e/netdev.c | 15 ++++++---------
  53  1 file changed, 6 insertions(+), 9 deletions(-)
  54
  55 diff --git a/drivers/net/ethernet/intel/e1000e/netdev.c b/drivers/net/ethernet/intel/e1000e/netdev.c
  56 index 6b1cacd86c6e..44312962b64e 100644
  57 --- a/drivers/net/ethernet/intel/e1000e/netdev.c
  58 +++ b/drivers/net/ethernet/intel/e1000e/netdev.c
  59 @@ -5246,8 +5246,13 @@ static void e1000_watchdog_task(struct work_struct *work)
  60                         /* 8000ES2LAN requires a Rx packet buffer work-around
  61                          * on link down event; reset the controller to flush
  62                          * the Rx packet buffer.
  63 +                        *
  64 +                        * If the link is lost the controller stops DMA, but
  65 +                        * if there is queued Tx work it cannot be done.  So
  66 +                        * reset the controller to flush the Tx packet buffers.
  67                          */
  68 -                       if (adapter->flags & FLAG_RX_NEEDS_RESTART)
  69 +                       if ((adapter->flags & FLAG_RX_NEEDS_RESTART) ||
  70 +                           e1000_desc_unused(tx_ring) + 1 < tx_ring->count)
  71                                 adapter->flags |= FLAG_RESTART_NOW;
  72                         else
  73                                 pm_schedule_suspend(netdev->dev.parent,
  74 @@ -5270,14 +5275,6 @@ link_up:
  75         adapter->gotc_old = adapter->stats.gotc;
  76         spin_unlock(&adapter->stats64_lock);
  77
  78 -       /* If the link is lost the controller stops DMA, but
  79 -        * if there is queued Tx work it cannot be done.  So
  80 -        * reset the controller to flush the Tx packet buffers.
  81 -        */
  82 -       if (!netif_carrier_ok(netdev) &&
  83 -           (e1000_desc_unused(tx_ring) + 1 < tx_ring->count))
  84 -               adapter->flags |= FLAG_RESTART_NOW;
  85 -
  86         /* If reset is necessary, do it outside of interrupt context. */
  87         if (adapter->flags & FLAG_RESTART_NOW) {
  88                 schedule_work(&adapter->reset_task);
  89 --
  90 2.19.1
  91