releases/3.18.85/e1000e-separate-signaling-for-link-check-link-up.patch

   1 From 19110cfbb34d4af0cdfe14cd243f3b09dc95b013 Mon Sep 17 00:00:00 2001
   2 From: Benjamin Poirier <bpoirier@suse.com>
   3 Date: Fri, 21 Jul 2017 11:36:26 -0700
   4 Subject: e1000e: Separate signaling for link check/link up
   5
   6 From: Benjamin Poirier <bpoirier@suse.com>
   7
   8 commit 19110cfbb34d4af0cdfe14cd243f3b09dc95b013 upstream.
   9
  10 Lennart reported the following race condition:
  11
  12 \ e1000_watchdog_task
  13     \ e1000e_has_link
  14         \ hw->mac.ops.check_for_link() === e1000e_check_for_copper_link
  15             /* link is up */
  16             mac->get_link_status = false;
  17
  18                             /* interrupt */
  19                             \ e1000_msix_other
  20                                 hw->mac.get_link_status = true;
  21
  22         link_active = !hw->mac.get_link_status
  23         /* link_active is false, wrongly */
  24
  25 This problem arises because the single flag get_link_status is used to
  26 signal two different states: link status needs checking and link status is
  27 down.
  28
  29 Avoid the problem by using the return value of .check_for_link to signal
  30 the link status to e1000e_has_link().
  31
  32 Reported-by: Lennart Sorensen <lsorense@csclub.uwaterloo.ca>
  33 Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
  34 Tested-by: Aaron Brown <aaron.f.brown@intel.com>
  35 Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
  36 Signed-off-by: Amit Pundir <amit.pundir@linaro.org>
  37 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  38
  39 ---
  40  drivers/net/ethernet/intel/e1000e/mac.c    |   11 ++++++++---
  41  drivers/net/ethernet/intel/e1000e/netdev.c |    2 +-
  42  2 files changed, 9 insertions(+), 4 deletions(-)
  43
  44 --- a/drivers/net/ethernet/intel/e1000e/mac.c
  45 +++ b/drivers/net/ethernet/intel/e1000e/mac.c
  46 @@ -410,6 +410,9 @@ void e1000e_clear_hw_cntrs_base(struct e
  47   *  Checks to see of the link status of the hardware has changed.  If a
  48   *  change in link status has been detected, then we read the PHY registers
  49   *  to get the current speed/duplex if link exists.
  50 + *
  51 + *  Returns a negative error code (-E1000_ERR_*) or 0 (link down) or 1 (link
  52 + *  up).
  53   **/
  54  s32 e1000e_check_for_copper_link(struct e1000_hw *hw)
  55  {
  56 @@ -423,7 +426,7 @@ s32 e1000e_check_for_copper_link(struct
  57          * Change or Rx Sequence Error interrupt.
  58          */
  59         if (!mac->get_link_status)
  60 -               return 0;
  61 +               return 1;
  62
  63         /* First we want to see if the MII Status Register reports
  64          * link.  If so, then we want to get the current speed/duplex
  65 @@ -461,10 +464,12 @@ s32 e1000e_check_for_copper_link(struct
  66          * different link partner.
  67          */
  68         ret_val = e1000e_config_fc_after_link_up(hw);
  69 -       if (ret_val)
  70 +       if (ret_val) {
  71                 e_dbg("Error configuring flow control\n");
  72 +               return ret_val;
  73 +       }
  74
  75 -       return ret_val;
  76 +       return 1;
  77  }
  78
  79  /**
  80 --- a/drivers/net/ethernet/intel/e1000e/netdev.c
  81 +++ b/drivers/net/ethernet/intel/e1000e/netdev.c
  82 @@ -4844,7 +4844,7 @@ static bool e1000e_has_link(struct e1000
  83         case e1000_media_type_copper:
  84                 if (hw->mac.get_link_status) {
  85                         ret_val = hw->mac.ops.check_for_link(hw);
  86 -                       link_active = !hw->mac.get_link_status;
  87 +                       link_active = ret_val > 0;
  88                 } else {
  89                         link_active = true;
  90                 }