Hey folks,
Seems that CX4 (Ethernet) sriov ports probe with link down in the latest stable 4.14 kernel. I upgrated firmware but with no luck :(
Has anyone seen this issue as well?
Kernel: stable 4.14.47 Firmware: 12.22.1002 psid: MT_2140110033
sysfs port state is DOWN and phys port state is Disabled.
I tried to add as much debug as I could to dump for the experts: -- [ 3243.373948] mlx5_core 0000:03:00.0: mlx5_core_sriov_configure:210:(pid 1616): requested num_vfs 1 [ 3243.374321] mlx5_core 0000:03:00.0: mlx5_device_enable_sriov:115:(pid 1616): successfully enabled VF* 0 [ 3243.482126] pci 0000:03:00.1: [15b3:1014] type 00 class 0x020000 [ 3243.482490] pci 0000:03:00.1: Max Payload Size set to 256 (was 128, max 512) [ 3243.482510] pci 0000:03:00.1: enabling Extended Tags [ 3243.487075] mlx5_core 0000:03:00.1: enabling device (0000 -> 0002) [ 3243.487220] mlx5_core 0000:03:00.1: firmware version: 12.22.1002 [ 3243.570233] mlx5_core 0000:03:00.1: handle_hca_cap:517:(pid 5010): Current Pkey table size 128 Setting new size 128 [ 3244.302809] mlx5_core 0000:03:00.1: alloc_comp_eqs:776:(pid 5010): allocated completion EQN 20 [ 3244.303052] mlx5_core 0000:03:00.1: alloc_comp_eqs:776:(pid 5010): allocated completion EQN 21 [ 3244.303296] mlx5_core 0000:03:00.1: alloc_comp_eqs:776:(pid 5010): allocated completion EQN 22 [ 3244.303590] mlx5_core 0000:03:00.1: alloc_comp_eqs:776:(pid 5010): allocated completion EQN 23 [ 3244.303825] mlx5_core 0000:03:00.1: alloc_comp_eqs:776:(pid 5010): allocated completion EQN 24 [ 3244.304049] mlx5_core 0000:03:00.1: alloc_comp_eqs:776:(pid 5010): allocated completion EQN 25 [ 3244.304276] mlx5_core 0000:03:00.1: alloc_comp_eqs:776:(pid 5010): allocated completion EQN 26 [ 3244.304499] mlx5_core 0000:03:00.1: alloc_comp_eqs:776:(pid 5010): allocated completion EQN 27 [ 3244.307241] mlx5_core 0000:03:00.1: mlx5_nic_vport_update_local_lb:939:(pid 5010): disable local_lb [ 3244.307613] mlx5_core 0000:03:00.1: MLX5E: StrdRq(0) RqSz(1024) StrdSz(1) RxCqeCmprss(0) [ 3244.308008] mlx5_core 0000:03:00.1: Assigned random MAC address 42:7b:57:5a:f2:06 [ 3244.460140] mlx5_core 0000:03:00.1 ens1f1: renamed from eth0 [ 3244.627854] mlx5_core 0000:03:00.1 ens1f1: Link down [ 3244.629033] IPv6: ADDRCONF(NETDEV_UP): ens1f1: link is not ready [ 3244.629480] IPv6: ADDRCONF(NETDEV_UP): ens1f1: link is not ready [ 3244.639340] pkey = 0xffff [ 3244.644081] pkey = 0xffff [ 3244.645809] pkey = 0xffff [ 3244.646510] pkey = 0x0 [ 3244.646557] pkey = 0x0 [ 3244.646600] pkey = 0x0 [ 3244.646649] pkey = 0x0 [ 3244.646692] pkey = 0x0 [ 3244.646751] pkey = 0x0 [ 3244.646793] pkey = 0x0 [ 3244.646835] pkey = 0x0 [ 3244.646882] pkey = 0x0 [ 3244.646933] pkey = 0x0 [ 3244.646986] pkey = 0x0 [ 3244.647044] pkey = 0x0 [ 3244.647086] pkey = 0x0 [ 3244.647136] pkey = 0x0 [ 3244.647182] pkey = 0x0 [ 3244.647240] pkey = 0x0 [ 3244.647283] pkey = 0x0 [ 3244.647333] pkey = 0x0 [ 3244.647384] pkey = 0x0 [ 3244.647434] pkey = 0x0 [ 3244.647497] pkey = 0x0 [ 3244.647554] pkey = 0x0 [ 3244.647598] pkey = 0x0 [ 3244.647652] pkey = 0x0 [ 3244.647695] pkey = 0x0 [ 3244.647737] pkey = 0x0 [ 3244.647786] pkey = 0x0 [ 3244.647839] pkey = 0x0 [ 3244.647884] pkey = 0x0 [ 3244.647934] pkey = 0x0 [ 3244.647986] pkey = 0x0 [ 3244.648049] pkey = 0x0 [ 3244.648092] pkey = 0x0 [ 3244.648134] pkey = 0x0 [ 3244.648184] pkey = 0x0 [ 3244.648241] pkey = 0x0 [ 3244.648286] pkey = 0x0 [ 3244.648334] pkey = 0x0 [ 3244.648385] pkey = 0x0 [ 3244.648435] pkey = 0x0 [ 3244.648495] pkey = 0x0 [ 3244.648539] pkey = 0x0 [ 3244.648584] pkey = 0x0 [ 3244.648634] pkey = 0x0 [ 3244.648685] pkey = 0x0 [ 3244.648748] pkey = 0x0 [ 3244.648792] pkey = 0x0 [ 3244.648836] pkey = 0x0 [ 3244.648885] pkey = 0x0 [ 3244.648935] pkey = 0x0 [ 3244.648998] pkey = 0x0 [ 3244.649042] pkey = 0x0 [ 3244.649086] pkey = 0x0 [ 3244.649138] pkey = 0x0 [ 3244.649185] pkey = 0x0 [ 3244.649235] pkey = 0x0 [ 3244.649285] pkey = 0x0 [ 3244.649344] pkey = 0x0 [ 3244.649388] pkey = 0x0 [ 3244.649435] pkey = 0x0 [ 3244.649484] pkey = 0x0 [ 3244.649549] pkey = 0x0 [ 3244.649593] pkey = 0x0 [ 3244.649638] pkey = 0x0 [ 3244.649686] pkey = 0x0 [ 3244.649741] pkey = 0x0 [ 3244.649785] pkey = 0x0 [ 3244.649836] pkey = 0x0 [ 3244.649886] pkey = 0x0 [ 3244.649936] pkey = 0x0 [ 3244.649988] pkey = 0x0 [ 3244.651604] pkey = 0x0 [ 3244.651652] pkey = 0x0 [ 3244.651697] pkey = 0x0 [ 3244.651744] pkey = 0x0 [ 3244.651788] pkey = 0x0 [ 3244.651839] pkey = 0x0 [ 3244.651889] pkey = 0x0 [ 3244.651939] pkey = 0x0 [ 3244.651989] pkey = 0x0 [ 3244.652048] pkey = 0x0 [ 3244.652090] pkey = 0x0 [ 3244.652140] pkey = 0x0 [ 3244.652195] pkey = 0x0 [ 3244.652241] pkey = 0x0 [ 3244.652300] pkey = 0x0 [ 3244.652343] pkey = 0x0 [ 3244.652403] pkey = 0x0 [ 3244.652450] pkey = 0x0 [ 3244.652493] pkey = 0x0 [ 3244.652540] pkey = 0x0 [ 3244.652593] pkey = 0x0 [ 3244.652640] pkey = 0x0 [ 3244.652697] pkey = 0x0 [ 3244.652755] pkey = 0x0 [ 3244.652800] pkey = 0x0 [ 3244.652848] pkey = 0x0 [ 3244.652892] pkey = 0x0 [ 3244.652950] pkey = 0x0 [ 3244.652993] pkey = 0x0 [ 3244.653041] pkey = 0x0 [ 3244.653091] pkey = 0x0 [ 3244.653141] pkey = 0x0 [ 3244.653191] pkey = 0x0 [ 3244.653241] pkey = 0x0 [ 3244.653294] pkey = 0x0 [ 3244.653341] pkey = 0x0 [ 3244.653391] pkey = 0x0 [ 3244.653531] pkey = 0x0 [ 3244.653593] pkey = 0x0 [ 3244.653935] pkey = 0x0 [ 3244.653996] pkey = 0x0 [ 3244.654253] pkey = 0x0 [ 3244.654974] pkey = 0x0 [ 3244.655038] pkey = 0x0 [ 3244.655097] pkey = 0x0 [ 3244.655143] pkey = 0x0 [ 3244.655194] pkey = 0x0 [ 3244.655253] pkey = 0x0 [ 3244.655308] pkey = 0x0 [ 3244.655350] pkey = 0x0 [ 3244.655393] pkey = 0x0 [ 3244.655448] pkey = 0x0 [ 3244.655494] pkey = 0x0 [ 3244.655544] pkey = 0x0 [ 3244.655595] pkey = 0x0 [ 3244.655648] pkey = 0x0 --
On Wed, Jun 20, 2018 at 01:40:16PM +0300, Sagi Grimberg wrote:
Hey folks,
Seems that CX4 (Ethernet) sriov ports probe with link down in the latest stable 4.14 kernel. I upgrated firmware but with no luck :(
Has anyone seen this issue as well?
I'm shooting to the dark, isn't this related to commit e3ca34880652250f524022ad89e516f8ba9a805b ?
Thanks
Seems that CX4 (Ethernet) sriov ports probe with link down in the latest stable 4.14 kernel. I upgrated firmware but with no luck :(
Has anyone seen this issue as well?
I'm shooting to the dark, isn't this related to commit e3ca34880652250f524022ad89e516f8ba9a805b ?
Probably not, this only effects rdma ulps what want to map to irq affinities. This is an ethernet vf probed with link down.
On Wed, Jun 20, 2018 at 1:40 PM, Sagi Grimberg sagi@grimberg.me wrote:
Hey folks,
Seems that CX4 (Ethernet) sriov ports probe with link down in the latest stable 4.14 kernel. I upgrated firmware but with no luck :(
Has anyone seen this issue as well?
Kernel: stable 4.14.47 Firmware: 12.22.1002 psid: MT_2140110033
sysfs port state is DOWN and
do you mean ip link on the pf shows link state down for the vf?
phys port state is Disabled.
where do you see it is disabled and what happens if you take the PF netdev link up?
We had a bug fix there recently [1] -- but it was for the switchdev mode not the legacy mode, are you using the switchdev mode? (if not, I recommend going there, the legacy mode is not going to last long)
Also, FWIW the whole ethernet sriov upstreaming is done in netdev, not rdma
Or.
[1] 84c9c8f net/mlx5e: Don't override vport admin link state in switchdev mode
Hey folks,
Seems that CX4 (Ethernet) sriov ports probe with link down in the latest stable 4.14 kernel. I upgrated firmware but with no luck :(
Has anyone seen this issue as well?
Kernel: stable 4.14.47 Firmware: 12.22.1002 psid: MT_2140110033
sysfs port state is DOWN and
do you mean ip link on the pf shows link state down for the vf?
the PF is fine, the VF is down.
phys port state is Disabled.
where do you see it is disabled and what happens if you take the PF netdev link up?
Nothing happens if I take it up, the carrier is off. I see the port phys state via infiniband sysfs (which exposes it from the hca vport context).
We had a bug fix there recently [1] -- but it was for the switchdev mode not the legacy mode,
It was backported to 4,14 stable...
are you using the switchdev mode? (if not, I recommend going there, the legacy mode is not going to last long)
No, I wasn't sure if it was mature enough in 4.14. And stable backports don't always propagate immediately...
I'll try switchdev mode.
Also, FWIW the whole ethernet sriov upstreaming is done in netdev, not rdma
I know, it should have went to netdev... Next time (which I hope won't come too soon ;))
On Wed, Jun 20, 2018 at 8:06 PM, Sagi Grimberg sagi@grimberg.me wrote:
I'll try switchdev mode.
in switchdev mode you have the take the host side VF representor netdev up such that the VF vport link will be enabled. Also in switchdev mode you need some host side controller SW to program the e-switch forwarding rule. You can make the experiment just to see if the link problem persists.
On Wed, Jun 20, 2018 at 01:40:16PM +0300, Sagi Grimberg wrote:
Hey folks,
Seems that CX4 (Ethernet) sriov ports probe with link down in the latest stable 4.14 kernel. I upgrated firmware but with no luck :(
Did this work on older kernels? If so, can you use 'git bisect' to track down the offending commit?
thanks,
greg k-h
On Wed, Jun 20, 2018 at 9:44 PM, Greg KH gregkh@linuxfoundation.org wrote:
On Wed, Jun 20, 2018 at 01:40:16PM +0300, Sagi Grimberg wrote:
Hey folks,
Seems that CX4 (Ethernet) sriov ports probe with link down in the latest stable 4.14 kernel. I upgrated firmware but with no luck :(
Did this work on older kernels? If so, can you use 'git bisect' to track down the offending commit?
so... what was/is the resolution here?
Or.
linux-stable-mirror@lists.linaro.org
-
Greg KH -
Leon Romanovsky -
Or Gerlitz -
Sagi Grimberg