BUG: nvmet with cxgb4 is broken in v4.12-rc1

BUG: nvmet with cxgb4 is broken in v4.12-rc1

[Logan Gunthorpe]

Hello,

Another cxgb4 bug report for you. Testing nvmet over our T62100-LP-CR
stopped working in v4.12-rc1. We are seeing two oopses: a general
protection fault in nvmet_rdma_free_rsps and a WQ_MEM_RECLAIM warning.
These occur after a suspect cxgb4 message. Please see the full dmesg log
which is attached.

> [   41.093555] cxgb4 0000:07:00.4: AE qpid 1034 opcode 0 status 0x1 type 1 len 0x0 wrid.hi 0x0 wrid.lo 0x0
> [   41.093673] nvmet_rdma: received IB QP event: QP access error (3)

I've bisected to find this commit is the culprit:

bb58d079: cxgb4: Update IngPad and IngPack values

A bisect log is also attached. Reverting that commit also fixes the issue.

Thanks,

Logan
-------------- next part --------------
A non-text attachment was scrubbed...
Name: bisect.nvmet1.log
Type: text/x-log
Size: 2866 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-nvme/attachments/20170517/b37c36e6/attachment-0002.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: dmesg.log
Type: text/x-log
Size: 89641 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-nvme/attachments/20170517/b37c36e6/attachment-0003.bin>

nvmet with cxgb4 is broken in v4.12-rc1

[Steve Wise]

> Hello,
> 
> Another cxgb4 bug report for you. Testing nvmet over our T62100-LP-CR
> stopped working in v4.12-rc1. We are seeing two oopses: a general
> protection fault in nvmet_rdma_free_rsps and a WQ_MEM_RECLAIM warning.
> These occur after a suspect cxgb4 message. Please see the full dmesg log
> which is attached.
> 
> > [   41.093555] cxgb4 0000:07:00.4: AE qpid 1034 opcode 0 status 0x1 type 1 len
> 0x0 wrid.hi 0x0 wrid.lo 0x0
> > [   41.093673] nvmet_rdma: received IB QP event: QP access error (3)
> 
> I've bisected to find this commit is the culprit:
> 
> bb58d079: cxgb4: Update IngPad and IngPack values
> 
> A bisect log is also attached. Reverting that commit also fixes the issue.
> 

Hey Logan, this is with 1.16.43.0 firmware? 

Hey Casey/Manoj, is there some other sw or fw change missing for the ingpad change?  Or maybe something needed in the firmware config file?

Thanks,

Steve.

nvmet with cxgb4 is broken in v4.12-rc1

[Logan Gunthorpe]

On 17/05/17 01:22 PM, Steve Wise wrote:
> Hey Logan, this is with 1.16.43.0 firmware? 

Yes, that's correct.

Logan