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

[Roland Dreier]

 > Supplying an ibv_qp_cap.max_inline_data value of 460 for mlx4_create_qp
 > was getting back ENOMEM when the max should have been 928. Tracked the bug
 > to the inline segment calculation. Here's the fix.

Any more information about what the bug really is here, or a test case?
As it stands I don't see anything wrong in theory or practice -- ie all
my tests work and I don't see why your patch makes any difference in the
value that ends up being calculated.

 - R.