Re: [PATCH net-next] liquidio: fix kernel panic in VF driver

[David Miller]

From: Felix Manlunas 
Date: Thu, 26 Oct 2017 16:46:36 -0700

> Doing ifconfig down on VF driver in the middle of receiving line rate
> traffic causes a kernel panic:
 ...
> Reason is:  in the function assigned to net_device_ops->ndo_stop, the steps
> for bringing down the interface are done in the wrong order.  The step that
> notifies the NIC firmware to stop forwarding packets to host is done too
> late.  Fix it by moving that step to the beginning.
> 
> Signed-off-by: Felix Manlunas 
> Signed-off-by: Raghu Vatsavayi 

Applied, thanks.

[PATCH net-next] liquidio: fix kernel panic in VF driver

[Felix Manlunas]

Doing ifconfig down on VF driver in the middle of receiving line rate
traffic causes a kernel panic:

LiquidIO_VF :02:00.3: should not come here should not get rx when poll 
mode = 0 for vf
BUG: unable to handle kernel NULL pointer dereference at   (null)
.
.
.
Call Trace:
 
 ? tasklet_action+0x102/0x120
 __do_softirq+0x91/0x292
 irq_exit+0xb6/0xc0
 do_IRQ+0x4f/0xd0
 common_interrupt+0x93/0x93
 
RIP: 0010:cpuidle_enter_state+0x142/0x2f0
RSP: 0018:a6403e20 EFLAGS: 0246 ORIG_RAX: ff59
RAX:  RBX: 0003 RCX: 001f
RDX:  RSI: 2ab7519f RDI: 
RBP: a6403e58 R08: 0084 R09: 0018
R10: a6403df0 R11: 03c7 R12: 0003
R13: d27ebd806800 R14: a64d40d8 R15: 007be072823f
 cpuidle_enter+0x17/0x20
 call_cpuidle+0x23/0x40
 do_idle+0x18c/0x1f0
 cpu_startup_entry+0x64/0x70
 rest_init+0xa5/0xb0
 start_kernel+0x45e/0x46b
 x86_64_start_reservations+0x24/0x26
 x86_64_start_kernel+0x6f/0x72
 secondary_startup_64+0xa5/0xa5
Code:  Bad RIP value.
RIP:   (null) RSP: 9246ed003f28
CR2: 
---[ end trace 92731e80f31b7d7d ]---
Kernel panic - not syncing: Fatal exception in interrupt
Kernel Offset: 0x2400 from 0x8100 (relocation range: 
0x8000-0xbfff)
---[ end Kernel panic - not syncing: Fatal exception in interrupt

Reason is:  in the function assigned to net_device_ops->ndo_stop, the steps
for bringing down the interface are done in the wrong order.  The step that
notifies the NIC firmware to stop forwarding packets to host is done too
late.  Fix it by moving that step to the beginning.

Signed-off-by: Felix Manlunas 
Signed-off-by: Raghu Vatsavayi 
---
 drivers/net/ethernet/cavium/liquidio/lio_vf_main.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/cavium/liquidio/lio_vf_main.c 
b/drivers/net/ethernet/cavium/liquidio/lio_vf_main.c
index 4c3b568..ed1f073 100644
--- a/drivers/net/ethernet/cavium/liquidio/lio_vf_main.c
+++ b/drivers/net/ethernet/cavium/liquidio/lio_vf_main.c
@@ -1288,6 +1288,9 @@ static int liquidio_stop(struct net_device *netdev)
struct octeon_device *oct = lio->oct_dev;
struct napi_struct *napi, *n;
 
+   /* tell Octeon to stop forwarding packets to host */
+   send_rx_ctrl_cmd(lio, 0);
+
if (oct->props[lio->ifidx].napi_enabled) {
list_for_each_entry_safe(napi, n, >napi_list, dev_list)
napi_disable(napi);
@@ -1305,9 +1308,6 @@ static int liquidio_stop(struct net_device *netdev)
netif_carrier_off(netdev);
lio->link_changes++;
 
-   /* tell Octeon to stop forwarding packets to host */
-   send_rx_ctrl_cmd(lio, 0);
-
ifstate_reset(lio, LIO_IFSTATE_RUNNING);
 
txqs_stop(netdev);