target/linux/generic/pending-4.9/190-e100e-avoid-receiver-interrupt-bursts.patch

   1 Subject: [1/5] e1000e: Fix error path in link detection
   2 From: Benjamin Poirier <bpoirier@suse.com>
   3 X-Patchwork-Id: 792259
   4 Message-Id: <20170721183627.13373-1-bpoirier@suse.com>
   5 To: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
   6 Cc: netdev@vger.kernel.org, intel-wired-lan@lists.osuosl.org,
   7  linux-kernel@vger.kernel.org,
   8  Lennart Sorensen <lsorense@csclub.uwaterloo.ca>
   9 Date: Fri, 21 Jul 2017 11:36:23 -0700
  10
  11 In case of error from e1e_rphy(), the loop will exit early and "success"
  12 will be set to true erroneously.
  13
  14 Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
  15 ---
  16  drivers/net/ethernet/intel/e1000e/phy.c | 7 ++++---
  17  1 file changed, 4 insertions(+), 3 deletions(-)
  18
  19 diff --git a/drivers/net/ethernet/intel/e1000e/phy.c b/drivers/net/ethernet/intel/e1000e/phy.c
  20 index d78d47b41a71..86ff0969efb6 100644
  21 --- a/drivers/net/ethernet/intel/e1000e/phy.c
  22 +++ b/drivers/net/ethernet/intel/e1000e/phy.c
  23 @@ -1744,6 +1744,7 @@ s32 e1000e_phy_has_link_generic(struct e1000_hw *hw, u32 iterations,
  24         s32 ret_val = 0;
  25         u16 i, phy_status;
  26
  27 +       *success = false;
  28         for (i = 0; i < iterations; i++) {
  29                 /* Some PHYs require the MII_BMSR register to be read
  30                  * twice due to the link bit being sticky.  No harm doing
  31 @@ -1763,16 +1764,16 @@ s32 e1000e_phy_has_link_generic(struct e1000_hw *hw, u32 iterations,
  32                 ret_val = e1e_rphy(hw, MII_BMSR, &phy_status);
  33                 if (ret_val)
  34                         break;
  35 -               if (phy_status & BMSR_LSTATUS)
  36 +               if (phy_status & BMSR_LSTATUS) {
  37 +                       *success = true;
  38                         break;
  39 +               }
  40                 if (usec_interval >= 1000)
  41                         msleep(usec_interval / 1000);
  42                 else
  43                         udelay(usec_interval);
  44         }
  45
  46 -       *success = (i < iterations);
  47 -
  48         return ret_val;
  49  }
  50
  51
  52 Subject: [2/5] e1000e: Fix wrong comment related to link detection
  53 From: Benjamin Poirier <bpoirier@suse.com>
  54 X-Patchwork-Id: 792257
  55 Message-Id: <20170721183627.13373-2-bpoirier@suse.com>
  56 To: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
  57 Cc: netdev@vger.kernel.org, intel-wired-lan@lists.osuosl.org,
  58  linux-kernel@vger.kernel.org,
  59  Lennart Sorensen <lsorense@csclub.uwaterloo.ca>
  60 Date: Fri, 21 Jul 2017 11:36:24 -0700
  61
  62 Reading e1000e_check_for_copper_link() shows that get_link_status is set to
  63 false after link has been detected. Therefore, it stays TRUE until then.
  64
  65 Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
  66 ---
  67  drivers/net/ethernet/intel/e1000e/netdev.c | 4 ++--
  68  1 file changed, 2 insertions(+), 2 deletions(-)
  69
  70 diff --git a/drivers/net/ethernet/intel/e1000e/netdev.c b/drivers/net/ethernet/intel/e1000e/netdev.c
  71 index 2dcb5463d9b8..58a87134d2e5 100644
  72 --- a/drivers/net/ethernet/intel/e1000e/netdev.c
  73 +++ b/drivers/net/ethernet/intel/e1000e/netdev.c
  74 @@ -5074,7 +5074,7 @@ static bool e1000e_has_link(struct e1000_adapter *adapter)
  75
  76         /* get_link_status is set on LSC (link status) interrupt or
  77          * Rx sequence error interrupt.  get_link_status will stay
  78 -        * false until the check_for_link establishes link
  79 +        * true until the check_for_link establishes link
  80          * for copper adapters ONLY
  81          */
  82         switch (hw->phy.media_type) {
  83 @@ -5092,7 +5092,7 @@ static bool e1000e_has_link(struct e1000_adapter *adapter)
  84                 break;
  85         case e1000_media_type_internal_serdes:
  86                 ret_val = hw->mac.ops.check_for_link(hw);
  87 -               link_active = adapter->hw.mac.serdes_has_link;
  88 +               link_active = hw->mac.serdes_has_link;
  89                 break;
  90         default:
  91         case e1000_media_type_unknown:
  92
  93 Content-Transfer-Encoding: 7bit
  94 Subject: [3/5] e1000e: Fix return value test
  95 From: Benjamin Poirier <bpoirier@suse.com>
  96 X-Patchwork-Id: 792258
  97 Message-Id: <20170721183627.13373-3-bpoirier@suse.com>
  98 To: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
  99 Cc: netdev@vger.kernel.org, intel-wired-lan@lists.osuosl.org,
 100  linux-kernel@vger.kernel.org,
 101  Lennart Sorensen <lsorense@csclub.uwaterloo.ca>
 102 Date: Fri, 21 Jul 2017 11:36:25 -0700
 103
 104 All the helpers return -E1000_ERR_PHY.
 105
 106 Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
 107 ---
 108  drivers/net/ethernet/intel/e1000e/netdev.c | 2 +-
 109  1 file changed, 1 insertion(+), 1 deletion(-)
 110
 111 diff --git a/drivers/net/ethernet/intel/e1000e/netdev.c b/drivers/net/ethernet/intel/e1000e/netdev.c
 112 index 58a87134d2e5..fc6a1d9999b2 100644
 113 --- a/drivers/net/ethernet/intel/e1000e/netdev.c
 114 +++ b/drivers/net/ethernet/intel/e1000e/netdev.c
 115 @@ -5099,7 +5099,7 @@ static bool e1000e_has_link(struct e1000_adapter *adapter)
 116                 break;
 117         }
 118
 119 -       if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
 120 +       if ((ret_val == -E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
 121             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
 122                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
 123                 e_info("Gigabit has been disabled, downgrading speed\n");
 124
 125 Content-Transfer-Encoding: 7bit
 126 Subject: [4/5] e1000e: Separate signaling for link check/link up
 127 From: Benjamin Poirier <bpoirier@suse.com>
 128 X-Patchwork-Id: 792262
 129 Message-Id: <20170721183627.13373-4-bpoirier@suse.com>
 130 To: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
 131 Cc: netdev@vger.kernel.org, intel-wired-lan@lists.osuosl.org,
 132  linux-kernel@vger.kernel.org,
 133  Lennart Sorensen <lsorense@csclub.uwaterloo.ca>
 134 Date: Fri, 21 Jul 2017 11:36:26 -0700
 135
 136 Lennart reported the following race condition:
 137
 138 \ e1000_watchdog_task
 139     \ e1000e_has_link
 140         \ hw->mac.ops.check_for_link() === e1000e_check_for_copper_link
 141             /* link is up */
 142             mac->get_link_status = false;
 143
 144                             /* interrupt */
 145                             \ e1000_msix_other
 146                                 hw->mac.get_link_status = true;
 147
 148         link_active = !hw->mac.get_link_status
 149         /* link_active is false, wrongly */
 150
 151 This problem arises because the single flag get_link_status is used to
 152 signal two different states: link status needs checking and link status is
 153 down.
 154
 155 Avoid the problem by using the return value of .check_for_link to signal
 156 the link status to e1000e_has_link().
 157
 158 Reported-by: Lennart Sorensen <lsorense@csclub.uwaterloo.ca>
 159 Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
 160 ---
 161  drivers/net/ethernet/intel/e1000e/mac.c    | 11 ++++++++---
 162  drivers/net/ethernet/intel/e1000e/netdev.c |  2 +-
 163  2 files changed, 9 insertions(+), 4 deletions(-)
 164
 165 diff --git a/drivers/net/ethernet/intel/e1000e/mac.c b/drivers/net/ethernet/intel/e1000e/mac.c
 166 index b322011ec282..f457c5703d0c 100644
 167 --- a/drivers/net/ethernet/intel/e1000e/mac.c
 168 +++ b/drivers/net/ethernet/intel/e1000e/mac.c
 169 @@ -410,6 +410,9 @@ void e1000e_clear_hw_cntrs_base(struct e1000_hw *hw)
 170   *  Checks to see of the link status of the hardware has changed.  If a
 171   *  change in link status has been detected, then we read the PHY registers
 172   *  to get the current speed/duplex if link exists.
 173 + *
 174 + *  Returns a negative error code (-E1000_ERR_*) or 0 (link down) or 1 (link
 175 + *  up).
 176   **/
 177  s32 e1000e_check_for_copper_link(struct e1000_hw *hw)
 178  {
 179 @@ -423,7 +426,7 @@ s32 e1000e_check_for_copper_link(struct e1000_hw *hw)
 180          * Change or Rx Sequence Error interrupt.
 181          */
 182         if (!mac->get_link_status)
 183 -               return 0;
 184 +               return 1;
 185
 186         /* First we want to see if the MII Status Register reports
 187          * link.  If so, then we want to get the current speed/duplex
 188 @@ -461,10 +464,12 @@ s32 e1000e_check_for_copper_link(struct e1000_hw *hw)
 189          * different link partner.
 190          */
 191         ret_val = e1000e_config_fc_after_link_up(hw);
 192 -       if (ret_val)
 193 +       if (ret_val) {
 194                 e_dbg("Error configuring flow control\n");
 195 +               return ret_val;
 196 +       }
 197
 198 -       return ret_val;
 199 +       return 1;
 200  }
 201
 202  /**
 203 diff --git a/drivers/net/ethernet/intel/e1000e/netdev.c b/drivers/net/ethernet/intel/e1000e/netdev.c
 204 index fc6a1d9999b2..5a8ab1136566 100644
 205 --- a/drivers/net/ethernet/intel/e1000e/netdev.c
 206 +++ b/drivers/net/ethernet/intel/e1000e/netdev.c
 207 @@ -5081,7 +5081,7 @@ static bool e1000e_has_link(struct e1000_adapter *adapter)
 208         case e1000_media_type_copper:
 209                 if (hw->mac.get_link_status) {
 210                         ret_val = hw->mac.ops.check_for_link(hw);
 211 -                       link_active = !hw->mac.get_link_status;
 212 +                       link_active = ret_val > 0;
 213                 } else {
 214                         link_active = true;
 215                 }
 216
 217 Content-Transfer-Encoding: 7bit
 218 Subject: [5/5] e1000e: Avoid receiver overrun interrupt bursts
 219 From: Benjamin Poirier <bpoirier@suse.com>
 220 X-Patchwork-Id: 792260
 221 Message-Id: <20170721183627.13373-5-bpoirier@suse.com>
 222 To: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
 223 Cc: netdev@vger.kernel.org, intel-wired-lan@lists.osuosl.org,
 224  linux-kernel@vger.kernel.org,
 225  Lennart Sorensen <lsorense@csclub.uwaterloo.ca>
 226 Date: Fri, 21 Jul 2017 11:36:27 -0700
 227
 228 When e1000e_poll() is not fast enough to keep up with incoming traffic, the
 229 adapter (when operating in msix mode) raises the Other interrupt to signal
 230 Receiver Overrun.
 231
 232 This is a double problem because 1) at the moment e1000_msix_other()
 233 assumes that it is only called in case of Link Status Change and 2) if the
 234 condition persists, the interrupt is repeatedly raised again in quick
 235 succession.
 236
 237 Ideally we would configure the Other interrupt to not be raised in case of
 238 receiver overrun but this doesn't seem possible on this adapter. Instead,
 239 we handle the first part of the problem by reverting to the practice of
 240 reading ICR in the other interrupt handler, like before commit 16ecba59bc33
 241 ("e1000e: Do not read ICR in Other interrupt"). Thanks to commit
 242 0a8047ac68e5 ("e1000e: Fix msi-x interrupt automask") which cleared IAME
 243 from CTRL_EXT, reading ICR doesn't interfere with RxQ0, TxQ0 interrupts
 244 anymore. We handle the second part of the problem by not re-enabling the
 245 Other interrupt right away when there is overrun. Instead, we wait until
 246 traffic subsides, napi polling mode is exited and interrupts are
 247 re-enabled.
 248
 249 Reported-by: Lennart Sorensen <lsorense@csclub.uwaterloo.ca>
 250 Fixes: 16ecba59bc33 ("e1000e: Do not read ICR in Other interrupt")
 251 Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
 252 ---
 253  drivers/net/ethernet/intel/e1000e/defines.h |  1 +
 254  drivers/net/ethernet/intel/e1000e/netdev.c  | 33 +++++++++++++++++++++++------
 255  2 files changed, 27 insertions(+), 7 deletions(-)
 256
 257 diff --git a/drivers/net/ethernet/intel/e1000e/defines.h b/drivers/net/ethernet/intel/e1000e/defines.h
 258 index 0641c0098738..afb7ebe20b24 100644
 259 --- a/drivers/net/ethernet/intel/e1000e/defines.h
 260 +++ b/drivers/net/ethernet/intel/e1000e/defines.h
 261 @@ -398,6 +398,7 @@
 262  #define E1000_ICR_LSC           0x00000004 /* Link Status Change */
 263  #define E1000_ICR_RXSEQ         0x00000008 /* Rx sequence error */
 264  #define E1000_ICR_RXDMT0        0x00000010 /* Rx desc min. threshold (0) */
 265 +#define E1000_ICR_RXO           0x00000040 /* Receiver Overrun */
 266  #define E1000_ICR_RXT0          0x00000080 /* Rx timer intr (ring 0) */
 267  #define E1000_ICR_ECCER         0x00400000 /* Uncorrectable ECC Error */
 268  /* If this bit asserted, the driver should claim the interrupt */
 269 diff --git a/drivers/net/ethernet/intel/e1000e/netdev.c b/drivers/net/ethernet/intel/e1000e/netdev.c
 270 index 5a8ab1136566..803edd1a6401 100644
 271 --- a/drivers/net/ethernet/intel/e1000e/netdev.c
 272 +++ b/drivers/net/ethernet/intel/e1000e/netdev.c
 273 @@ -1910,12 +1910,30 @@ static irqreturn_t e1000_msix_other(int __always_unused irq, void *data)
 274         struct net_device *netdev = data;
 275         struct e1000_adapter *adapter = netdev_priv(netdev);
 276         struct e1000_hw *hw = &adapter->hw;
 277 +       u32 icr;
 278 +       bool enable = true;
 279 +
 280 +       icr = er32(ICR);
 281 +       if (icr & E1000_ICR_RXO) {
 282 +               ew32(ICR, E1000_ICR_RXO);
 283 +               enable = false;
 284 +               /* napi poll will re-enable Other, make sure it runs */
 285 +               if (napi_schedule_prep(&adapter->napi)) {
 286 +                       adapter->total_rx_bytes = 0;
 287 +                       adapter->total_rx_packets = 0;
 288 +                       __napi_schedule(&adapter->napi);
 289 +               }
 290 +       }
 291 +       if (icr & E1000_ICR_LSC) {
 292 +               ew32(ICR, E1000_ICR_LSC);
 293 +               hw->mac.get_link_status = true;
 294 +               /* guard against interrupt when we're going down */
 295 +               if (!test_bit(__E1000_DOWN, &adapter->state)) {
 296 +                       mod_timer(&adapter->watchdog_timer, jiffies + 1);
 297 +               }
 298 +       }
 299
 300 -       hw->mac.get_link_status = true;
 301 -
 302 -       /* guard against interrupt when we're going down */
 303 -       if (!test_bit(__E1000_DOWN, &adapter->state)) {
 304 -               mod_timer(&adapter->watchdog_timer, jiffies + 1);
 305 +       if (enable && !test_bit(__E1000_DOWN, &adapter->state)) {
 306                 ew32(IMS, E1000_IMS_OTHER);
 307         }
 308
 309 @@ -2687,7 +2705,8 @@ static int e1000e_poll(struct napi_struct *napi, int weight)
 310                 napi_complete_done(napi, work_done);
 311                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
 312                         if (adapter->msix_entries)
 313 -                               ew32(IMS, adapter->rx_ring->ims_val);
 314 +                               ew32(IMS, adapter->rx_ring->ims_val |
 315 +                                    E1000_IMS_OTHER);
 316                         else
 317                                 e1000_irq_enable(adapter);
 318                 }
 319 @@ -4204,7 +4223,7 @@ static void e1000e_trigger_lsc(struct e1000_adapter *adapter)
 320         struct e1000_hw *hw = &adapter->hw;
 321
 322         if (adapter->msix_entries)
 323 -               ew32(ICS, E1000_ICS_OTHER);
 324 +               ew32(ICS, E1000_ICS_LSC | E1000_ICS_OTHER);
 325         else
 326                 ew32(ICS, E1000_ICS_LSC);
 327  }