[dpdk-dev] [PATCH 1/1] net/enic: enic driver doesn't work with 2nd process

[Hyong Youb Kim (hyonkim)]

> -----Original Message-----
> From: dev <dev-bounces at dpdk.org> On Behalf Of dirk
> Sent: Tuesday, August 27, 2019 6:03 PM
> To: dev at dpdk.org
> Subject: [dpdk-dev] [PATCH 1/1] net/enic: enic driver doesn't work with 2nd
> process
> 
> net/enic: enic driver doesn't work with 2nd process. the problem is that the
> private data structure enic uses pointers to rte_eth_dev and rte_pci_device
> which are overwritten when a 2nd process starts using this device.
> given patch removes these two pointers from the private data and uses the
> correct pointers. For us the given patch works fine on a cisco rack.
> (we used three processes working concurrently)
> 
> Signed-off-by: dirk <dirk.lenz at ng4t.com>
> ---
>  drivers/net/enic/enic.h               |  36 ++++---
>  drivers/net/enic/enic_clsf.c          |   5 +-
>  drivers/net/enic/enic_ethdev.c        |  79 ++++++---------
>  drivers/net/enic/enic_main.c          | 177 +++++++++++++++++++--------------
> -
>  drivers/net/enic/enic_res.c           |   7 +-
>  drivers/net/enic/enic_res.h           |   6 +-
>  drivers/net/enic/enic_rxtx_vec_avx2.c |   5 +-
>  7 files changed, 156 insertions(+), 159 deletions(-)
> 

Hi,

Thanks for the patch and reporting the bug!

Yes, enic->rte_dev and enic->pdev are supposed to be set and used only
by the primary. I can reproduce the crash by running two different
executables as primary and secondary. Using different executables so
that primary and secondary processes load rte_eth_devices at different
addresses.

I see other related issues including...
- eth_enicpmd_dev_init overwriting primary's fields in struct enic,
  other than rte_dev and pdev.
- Secondary not setting optimal rx/tx handlers.
- RSS update handlers (e.g. rss_hash_update) not checking for primary

I will send a patch series with Reported-by: Dirk Lenz. And you can
try it out and add Tested-by on that series. Are you okay with this
plan? Did I get your name right?

Thanks a lot.
-Hyong