DPDK patches and discussions
 help / color / mirror / Atom feed
From: "Nélio Laranjeiro" <nelio.laranjeiro@6wind.com>
To: Moti Haimovsky <motih@mellanox.com>
Cc: adrien.mazarguil@6wind.com, yskoh@mellanox.com, dev@dpdk.org
Subject: Re: [dpdk-dev] [RFC] net/mlx5: add support 32bit build
Date: Mon, 4 Jun 2018 09:24:48 +0200	[thread overview]
Message-ID: <20180604072448.vxq5ahapnyztklyg@laranjeiro-vm.dev.6wind.com> (raw)
In-Reply-To: <1528039753-25889-1-git-send-email-motih@mellanox.com>

On Sun, Jun 03, 2018 at 06:29:13PM +0300, Moti Haimovsky wrote:
> This RFC and patch aim to add support for building mlx5 PMD on 32bit
> archs such as i686.
> The attached patch was not tested on 32bit environments and is given
> here for explanatory purposes.
> 
> Prerequisites:
> * RDMA-core with 32bit support libraries.
> * compile and build tools for 32bit systems according to DPDK documentation.
> 
> Assumptions:
> * Kernel and drivers are 64bit.
> * Only user programs and support libraries are 32Bit.
> 
> Changes proposed in the PMD for adding 32bit support:
> The attached patch provides the changes needed in the PMD in order to support
> 32bit builds with the major changes being:
> * Modifying the UAR access routine to support non-atomic 64bit writes
>   according to the mlx5 hardware reference manual.
> * Handling 64bit address fields that are written to the device WQEs.
> * Modify UAR address and size mapping to fit into the 32bit address space.
> 
> Concerns:
> * The major concern of this modification is the UAR mapping.
>   Current PMD assumes 64bit address and therefore feels free to map 4G size
>   UAR to virtual address just above the huge-page addresses which of course
>   cannot work for 32bit address space.
>   As a workaround I reduced the UAR size to half the original size but
>   this of course is not the solution.
> * Will vectorized support work on 32bit builds ?
> 
> I am interested in your inputs regarding the concerns listed above,
> especially the UAR mapping issue mentioned there.
>[...]

Hi Moti,

My biggest concern and why such support has been removed [1] is because
Mellanox OFED is not supporting 32bits compilation.  As Mellanox OFED is
still supported, we need such guarantee.

On Mellanox website there is no way to download such kind of ISO which
means such support becomes only valid for upstream RDMA-Core library
with almost newer kernels.

If you can find a way to detect the PMD is compiled in front of Mellanox
OFED to forbid such support, it would be great, otherwise I am afraid
such support cannot be added back without creating a lot of issues for
people wanting it on 32bits processors with Mellanox OFED.

Thanks,

[1] https://dpdk.org/browse/dpdk/commit/?id=ebbb81eb27daca0a89ee8f228fcf141d9eb6ef1c

-- 
Nélio Laranjeiro
6WIND

      reply	other threads:[~2018-06-04  7:24 UTC|newest]

Thread overview: 2+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2018-06-03 15:29 Moti Haimovsky
2018-06-04  7:24 ` Nélio Laranjeiro [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20180604072448.vxq5ahapnyztklyg@laranjeiro-vm.dev.6wind.com \
    --to=nelio.laranjeiro@6wind.com \
    --cc=adrien.mazarguil@6wind.com \
    --cc=dev@dpdk.org \
    --cc=motih@mellanox.com \
    --cc=yskoh@mellanox.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).