Re: Mellanox ConnectX-3 VF driver can't work with 16 CPUs?

On Thu, 9 Feb 2017 10:03:20 +0000
Dexuan Cui [off-list ref] wrote:

Hi, 
While trying SR-IOV with a Linux guest running on Hyper-V, I found
this issue: the VF driver can't work if the guest has 16 virtual CPUs
(less vCPUs, e.g. 8,  can work fine):

[    9.927820] mlx4_core: Mellanox ConnectX core driver v2.2-1 (Feb,
2014) [    9.927882] mlx4_core: Initializing b961:00:02.0
[    9.970994] mlx4_core b961:00:02.0: Detected virtual function -
running in slave mode [    9.976783] mlx4_core b961:00:02.0: Sending
reset [    9.985858] mlx4_core b961:00:02.0: Sending vhcr0
[   10.004855] mlx4_core b961:00:02.0: HCA minimum page size:512
[   10.010465] mlx4_core b961:00:02.0: Timestamping is not supported
in slave mode [   10.203065] mlx4_core b961:00:02.0: Failed to
initialize event queue table, aborting [   10.226728] mlx4_core:
probe of b961:00:02.0 failed with error -12

I'm using the mainline kernel (4.10.0-rc4).

Any idea?

Thanks for reporting this, Dexuan.  We indeed found a bug in the Linux
driver, which requested too many EQEs per EQ created in SRIOV mode.
This resulted in the VF exhausting its supply of MTT entries.

We will submit an upstream fix for this soon (after thorough testing).

-Jack

Thanks,
-- Dexuan