[ewg] iWARP uDAPL RDMA_READ(RR) test hangs on Chelsio T320

[Steve Wise]

On 6/12/2013 5:45 PM, John Jolly wrote:
> Steve,
>
> On Tue, Jun 11, 2013 at 02:26:01PM -0500, Steve Wise wrote:
>> Is this T3 or T4?
> 22:00.0 Ethernet controller: Chelsio Communications Inc T320 10GbE Dual Port Adapter
>      Subsystem: Chelsio Communications Inc Device 0001
>      Flags: bus master, fast devsel, latency 0, IRQ 11
>      Memory at e7880000 (64-bit, non-prefetchable) [size=4K]
>      Memory at e7000000 (64-bit, non-prefetchable) [size=8M]
>      Memory at e7881000 (64-bit, non-prefetchable) [size=4K]
>      [virtual] Expansion ROM at e6000000 [disabled] [size=512K]
>      Capabilities: [40] Power Management version 3
>      Capabilities: [48] Message Signalled Interrupts: Mask- 64bit+ Count=1/32 Enable-
>      Capabilities: [58] Express Endpoint, MSI 00
>      Capabilities: [94] Vital Product Data <?>
>      Capabilities: [9c] MSI-X: Enable+ Mask- TabSize=32
>      Capabilities: [100] Device Serial Number 01-00-00-00-01-00-00-00
>      Capabilities: [300] Advanced Error Reporting
>      Kernel driver in use: cxgb3
>      Kernel modules: cxgb3
>
>> If dapl calls ibv_post_send() with wr->num_sge > 1 for a read
>> request, the chelsio provider will return -EINVAL in T4 and -1 in
>> T3.
> So is this a matter of telling our customer to stop doing this because
> it's not supported?

It definitely isn't supported.  We should see some synchronous error, 
though, when posting the read work requests. IE dat_ep_post_rdma_read() 
should be failing.  It would be nice if you could verify this.


>> We shouldn't see a hang though. :)
> We are getting reports of other dapltest problems, such as
> '!(t4_rq_empty(wq))' assertion on a pair of T422s. I have the feeling
> I will become very familiar with dapltest.

There were some fixes in the flush logic of the t4 sw recently. They 
aren't upstream yet, but you can most likely get them from the latest T4 
kit from chelsio's web site.

You should definitely send email to support at chelsio.com and get a 
problem report opened for this.


Steve.