[ofa-general] [PATCH] libmlx4: Re-calculate number of inline segments

[Vincent Rizza]

Yep sorry that original summary should have said 'greater than 460'. We'll work 
   up a test case to show the problem.

Roland Dreier wrote:
>  > When we set qp_attr.cap.max_inline_data between 461 and 928 we're seeing ENOMEM.
>  > Shouldn't the maximum handled be 928?
> 
> That's not what your original email said:
> 
>>> Supplying an ibv_qp_cap.max_inline_data value of 460...
> 
> But anyway, I just tried 461, 480, 511, 512, 513, 900, and 928 with the
> same test program and all work as expected (UD QPs have a limit of 900
> bytes of inline data, so asking for 928 failed, but that is correct I
> believe).
> 
> Can you give a specific QP type and data size where your version of
> num_inline_segs() returns a different value from the existing one?
> That might be the best way to explain the bug you've found.
> 
>  - R.