Re: [PATCH net-next V1 00/11] net/mlx5: ConnectX-4 100G Ethernet driver

[Amir Vadai <amirv@mellanox.com>]

On Tue, Apr 14, 2015 at 5:15 AM, David Miller <davem@davemloft.net> wrote:
> From: Amir Vadai <amirv@mellanox.com>
> Date: Mon, 13 Apr 2015 14:44:50 +0300
>
>> - Removed V0 Patch 1/11 ("net/mlx5_core: Virtually extend work/completion queue
>>   buffers by one page") due to misuse of DMA API. Thanks Dave.
>
> I don't know how stupid you think I am....
Hi,

It is our interest to fix this bug here and in the existing code. We
don't like either, that our driver is mis-using the DMA API and that
it is broken on some arch's.

>
> You may have removed the patch that did the refactoring, but the
> completely broken code misusing the DMA API is still there in this
> driver.
>
> I said I want you to fix this fundamental problem with your driver
> before making major changes like this series to it.
>
> And I really meant it.
>
> Don't try to side step the issue in the way that you did here.
It is just a misunderstanding of your sentence: "I'm sorry but I'm
going to require that you fix this mis-use of the DMA API first.". I
thought that it should be fixed in this patchset first....

We were working on a fix to the existing code that is based on vmap,
and I planned to send it in a separate patch later on.
I will soon send V2 with a patch to remove the vmap()  from current
code. It might be problematic to some Infiniband users of mlx5_core,
that will now need to allocate big continuous buffers, I will send the
fix to that in a separate patch in few weeks.

Thanks,
Amir
>
> Thanks.