Re: [dpdk-dev] [PATCH 1/2] ethdev: make flow API primary/secondary process safe

[Suanming Mou <suanmingm@nvidia.com>]

> -----Original Message-----
> From: Ferruh Yigit <ferruh.yigit@intel.com>
> Sent: Thursday, April 15, 2021 3:43 PM
> To: Suanming Mou <suanmingm@nvidia.com>; Ori Kam <orika@nvidia.com>;
> Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>; NBU-Contact-Thomas
> Monjalon <thomas@monjalon.net>
> Cc: dev@dpdk.org; Stephen Hemminger <stephen@networkplumber.org>;
> Matan Azrad <matan@nvidia.com>; Shahaf Shuler <shahafs@nvidia.com>; Ajit
> Khaparde <ajit.khaparde@broadcom.com>; Somnath Kotur
> <somnath.kotur@broadcom.com>
> Subject: Re: [dpdk-dev] [PATCH 1/2] ethdev: make flow API primary/secondary
> process safe
> 
> On 4/15/2021 3:55 AM, Suanming Mou wrote:
> > Hi,
> >
> >> -----Original Message-----
> >> From: Ferruh Yigit <ferruh.yigit@intel.com>
> >> Sent: Wednesday, April 14, 2021 9:07 PM
> >> To: Suanming Mou <suanmingm@nvidia.com>; Ori Kam <orika@nvidia.com>;
> >> Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>; NBU-Contact-
> Thomas
> >> Monjalon <thomas@monjalon.net>
> >> Cc: dev@dpdk.org; Stephen Hemminger <stephen@networkplumber.org>
> >> Subject: Re: [dpdk-dev] [PATCH 1/2] ethdev: make flow API
> >> primary/secondary process safe
> >>
> >> On 3/16/2021 11:48 PM, Suanming Mou wrote:
> >>> Hi Stephen,
> >>>
> >>>> -----Original Message-----
> >>>> From: Stephen Hemminger <stephen@networkplumber.org>
> >>>> Sent: Tuesday, March 16, 2021 3:27 AM
> >>>> To: dev@dpdk.org
> >>>> Cc: Stephen Hemminger <stephen@networkplumber.org>; Suanming Mou
> >>>> <suanmingm@nvidia.com>
> >>>> Subject: [PATCH 1/2] ethdev: make flow API primary/secondary
> >>>> process safe
> >>>>
> >>>> Posix mutex are not by default safe for protecting for usage from
> >>>> multiple processes. The flow ops mutex could be used by both
> >>>> primary and secondary processes.
> >>>
> >>> Process safe is something more widely scope. I assume it should be
> >>> another
> >> feature but not a bugfix for thread-safe?
> >>> And the fag RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE we have added is just
> >> thread safe.
> >>>
> >>
> >> Hi Suanming,
> >>
> >> I think 'RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE' flag and what this patch
> >> address are different issues.
> >>
> >> 'RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE' is to add/remove synchronization
> >> support for flow APIs, that is for thread safety as flag name suggests.
> >>
> >> This patch is to solve the problem for multi process, where commit
> >> log describes as posix mutex is not safe for multiple process.
> >
> > So for PMDs which not set the RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE
> capability bit, they will have the process level protection in multi-process.
> > For PMDs which set the RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE capability
> bit, this change does not help with these PMDs. If the PMD with
> RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE capability bit  internally does not
> support multi-process, they may still suffer crash etc.
> 
> Correct
> 
> > (If I understand correctly, mlx PMD level now should support multi-process,
> but better to have the confirmation from maintainers with much deeper level).
> > I assume this patch solves the posix mutex for multi-process only, hard to say
> the flow API primary/secondary process safe after that patch.
> >
> 
> I am also not quite sure how PMDs that doesn't require mutex at all, (mlx5, bnxt,
> sfc) behave on multi process. Is calling flow APIs from both primary/secondary
> safe?

This depends on the vendor driver. For mlx, it should be safe, not sure others.

> 
> >>
> >>
> >> Stephen,
> >> Are you aware of any downside setting 'PTHREAD_PROCESS_SHARED'
> >> attribute to the mutex? Any possible performance implications?
> >>
> >> Ori,
> >> Since mlx is heavily using the flow API, is it possible to test this
> >> patch? If there is no negative impact, I think we can get this patch, what do
> you think?
> >>
> >>>>