Skip to contentSkip to conversation

igb: Do not call netif_device_detach() when PCIe link goes missing

PATCH In review Mika Westerberg 2 participants 2 messages Jan 23, 2018 – Feb 2, 2018

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 igb: Do not call netif_device_detach() when PCIe link goes missing

When the driver notices that PCIe link is gone by reading 0xffffffff from a register it clears hw->hw_addr and then calls netif_device_detach(). This happens when the PCIe device is physically unplugged for example the user disconnected the Thunderbolt cable.

However, netif_device_detach() prevents netif_unregister() from bringing the device down properly including tearing down MSI-X vectors. This triggers following crash during the driver removal:

igb 0000:0b:00.0 enp11s0f0: PCIe link lost, device now detached

  ------------[ cut here ]------------
  kernel BUG at drivers/pci/msi.c:352!
  invalid opcode: 0000 [#1] PREEMPT SMP PTI
  ...
  Call Trace:
   pci_disable_msix+0xc9/0xf0
   igb_reset_interrupt_capability+0x58/0x60 [igb]
   igb_remove+0x90/0x100 [igb]
   pci_device_remove+0x31/0xa0
   device_release_driver_internal+0x152/0x210
   pci_stop_bus_device+0x78/0xa0
   pci_stop_bus_device+0x38/0xa0
   pci_stop_bus_device+0x38/0xa0
   pci_stop_bus_device+0x26/0xa0
   pci_stop_bus_device+0x38/0xa0
   pci_stop_and_remove_bus_device+0x9/0x20
   trim_stale_devices+0xee/0x130
   ? _raw_spin_unlock_irqrestore+0xf/0x30
   trim_stale_devices+0x8f/0x130
   ? _raw_spin_unlock_irqrestore+0xf/0x30
   trim_stale_devices+0xa1/0x130
   ? get_slot_status+0x8b/0xc0
   acpiphp_check_bridge.part.7+0xf9/0x140
   acpiphp_hotplug_notify+0x170/0x1f0
   ...

To prevent the crash do not call netif_device_detach() in igb_rd32(). This should be fine because hw->hw_addr is set to NULL preventing future hardware access of the now missing device.

---
 drivers/net/ethernet/intel/igb/igb_main.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/net/ethernet/intel/igb/igb_main.c +1 -2
index c208753ff5b7..b1de866e70b2 100644
@@ -774,8 +774,7 @@ u32 igb_rd32(struct e1000_hw *hw, u32 reg)
 	if (!(~value) && (!reg || !(~readl(hw_addr)))) {
 		struct net_device *netdev = igb->netdev;
 		hw->hw_addr = NULL;
-		netif_device_detach(netdev);
-		netdev_err(netdev, "PCIe link lost, device now detached\n");
+		netdev_err(netdev, "PCIe link lost\n");
 	}
 
 	return value;
Signature · 1 line
2.15.1
Tested-by on v1 1/1
#1 lore raw
Quoted 51 lines

From: netdev-owner at vger.kernel.org [mailto:netdev-
owner at vger.kernel.org] On Behalf Of Mika Westerberg
Sent: Tuesday, January 23, 2018 2:29 AM
To: Kirsher, Jeffrey T <[email protected]>
Cc: Ferenc Boldog <[email protected]>; Nikolay Bogoychev <[email protected]>; Mika Westerberg <[email protected]>; intel-wired-lan at lists.osuosl.org; netdev at vger.kernel.org
Subject: [PATCH] igb: Do not call netif_device_detach() when PCIe link goes missing

When the driver notices that PCIe link is gone by reading 0xffffffff from a register it clears hw->hw_addr and then calls netif_device_detach(). This happens when the PCIe device is physically unplugged for example the user disconnected the Thunderbolt cable.

However, netif_device_detach() prevents netif_unregister() from bringing the device down properly including tearing down MSI-X vectors. This triggers following crash during the driver removal:

igb 0000:0b:00.0 enp11s0f0: PCIe link lost, device now detached

  ------------[ cut here ]------------
  kernel BUG at drivers/pci/msi.c:352!
  invalid opcode: 0000 [#1] PREEMPT SMP PTI
  ...
  Call Trace:
   pci_disable_msix+0xc9/0xf0
   igb_reset_interrupt_capability+0x58/0x60 [igb]
   igb_remove+0x90/0x100 [igb]
   pci_device_remove+0x31/0xa0
   device_release_driver_internal+0x152/0x210
   pci_stop_bus_device+0x78/0xa0
   pci_stop_bus_device+0x38/0xa0
   pci_stop_bus_device+0x38/0xa0
   pci_stop_bus_device+0x26/0xa0
   pci_stop_bus_device+0x38/0xa0
   pci_stop_and_remove_bus_device+0x9/0x20
   trim_stale_devices+0xee/0x130
   ? _raw_spin_unlock_irqrestore+0xf/0x30
   trim_stale_devices+0x8f/0x130
   ? _raw_spin_unlock_irqrestore+0xf/0x30
   trim_stale_devices+0xa1/0x130
   ? get_slot_status+0x8b/0xc0
   acpiphp_check_bridge.part.7+0xf9/0x140
   acpiphp_hotplug_notify+0x170/0x1f0
   ...

To prevent the crash do not call netif_device_detach() in igb_rd32(). This should be fine because hw->hw_addr is set to NULL preventing future hardware access of the now missing device.

Quoted 7 lines
---
 drivers/net/ethernet/intel/igb/igb_main.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)