On Wed, Apr 13, 2016 at 8:48 PM, Jason Gunthorpe
<jguntho...@obsidianresearch.com> wrote:
> On Wed, Apr 13, 2016 at 07:11:03PM +0300, Saeed Mahameed wrote:
>
>> Fixes: b0844444590e ("net/mlx5_core: Introduce access function to read
>> internal timer ")
>> Fixes: b4ff3a36d3e4 ("net/mlx5: Use offset based reserved field names in the
>> IFC header file")
>
> Are you sure those are right? b doesn't have the
> reserved_at names.
>
Yes, as stated in the commit message, change b0844444590e replaced two
reserved fields with 3 fields with a different size than the original
two, which broke all the offsets of the fields that came after.
>
> You know, the reserved_XXX just need to have unique unchanging
> numbers, it doesn't matter what the numbers are - but you have to stop
> changing them :( That is the key to avoiding conflicts when
> backporting/merging/etc.
>
> I guess the big rename has already landed, but simply stopping the
> practice of renumbing the reserved fields would have been enough.
>
This was a mistake which had to be fixed, the offset numbering was
wrong from day one of introducing those offsets, I prefer to do it
now, once and forever.
But i totally agree, we will do our best to keep this file clean as
much as possible.
Thanks for your concern.
Saeed.
