[ofa-general] Re: dapl attribute bug

[Steve Wise]

Hey Arlin,

Did this ever get fixed?

I think UNH is seeing this issue still.



Steve Wise wrote:
> Davis, Arlin R wrote:
>>  
>>
>>  
>>> The DAPL dat_ia_attr->max_lmr_block_size is a u32, yet the dapl code 
>>> maps this to the linux ib_device_attr->max_mr_size which is u64.
>>>
>>> This causes dapltest to fail in some cases when running over chelsio 
>>> which sets max_mr_size to 0x100000000 (4GB).  The dapl code 
>>> truncates the value to 0. See dapl/openib_cma/dapl_ib_util.c.
>>>
>>> I'm not sure what the fix should be, but maybe the dapl code should 
>>> set anything over 32 bits to 0xffffffff?
>>>
>>>     
>>
>> This attribute changed with DAT 2.0 to match the 32-bit ibv_sge
>> length field. Since there are no direct max lmr segments mappings
>> I will need add some checks when setting max_lmr_block_size from
>> max_mr_size. Thanks.
>>
>> -arlin
>
> I'll test your fix when its ready.  Lemme know.
>
>
> Steve.
>
> _______________________________________________
> general mailing list
> general at lists.openfabrics.org
> http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
>
> To unsubscribe, please visit 
> http://openib.org/mailman/listinfo/openib-general