RE: [PATCH v4] net/ice: retry sending adminQ command after failure

["Zhang, Peng1X" <peng1x.zhang@intel.com>]

Because under the following given situation, it still have some issue. To this given situation, it is sure that retry mechanism is necessary. 
While is it confirm that under other situation retry mechanism is always better than without retry? 
If modify 'ice_dcf_send_aq_cmd', it will affect the handling to AdminQ under DCF. It looks like it will affect much more than following given situation. 

> -----Original Message-----
> From: Zhang, Qi Z <qi.z.zhang@intel.com>
> Sent: Wednesday, June 1, 2022 11:23 AM
> To: Zhang, Peng1X <peng1x.zhang@intel.com>; Yang, Qiming
> <qiming.yang@intel.com>; dev@dpdk.org
> Cc: stable@dpdk.org
> Subject: RE: [PATCH v4] net/ice: retry sending adminQ command after failure
> 
> 
> 
> > -----Original Message-----
> > From: Zhang, Peng1X <peng1x.zhang@intel.com>
> > Sent: Wednesday, June 1, 2022 10:46 AM
> > To: Zhang, Qi Z <qi.z.zhang@intel.com>; Yang, Qiming
> > <qiming.yang@intel.com>; dev@dpdk.org
> > Cc: stable@dpdk.org
> > Subject: RE: [PATCH v4] net/ice: retry sending adminQ command after
> > failure
> >
> > Because currently just focus on the following given situation, and use
> > 'hw-
> > >dcf_enabled' it should only affect on DCF.
> > After having checked code 'ice_dcf_send_aq_cmd', it already have retry
> > mechanism.
> 
> 
> If it is already retried, then why we need this patch?
> Btw, we are talking about "retry virtual channel command from DCF to kernel
> PF" here, not the case "retry polling reply status of the a request".
> 
> >
> > > -----Original Message-----
> > > From: Zhang, Qi Z <qi.z.zhang@intel.com>
> > > Sent: Wednesday, June 1, 2022 9:54 AM
> > > To: Zhang, Peng1X <peng1x.zhang@intel.com>; Yang, Qiming
> > > <qiming.yang@intel.com>; dev@dpdk.org
> > > Cc: stable@dpdk.org
> > > Subject: RE: [PATCH v4] net/ice: retry sending adminQ command after
> > > failure
> > >
> > >
> > >
> > > > -----Original Message-----
> > > > From: Zhang, Peng1X <peng1x.zhang@intel.com>
> > > > Sent: Wednesday, June 1, 2022 9:49 AM
> > > > To: Zhang, Qi Z <qi.z.zhang@intel.com>; Yang, Qiming
> > > > <qiming.yang@intel.com>; dev@dpdk.org
> > > > Cc: stable@dpdk.org
> > > > Subject: RE: [PATCH v4] net/ice: retry sending adminQ command
> > > > after failure
> > > >
> > > > I see, so ice_aq_retry_send_cmd  function will  conduct whether
> > > > DCF is enabled or not with 'hw->dcf_enabled'.
> > >
> > > Why not just enable retry in ice_dcf_send_aq_cmd?
> > >
> > >
> > > > > -----Original Message-----
> > > > > From: Zhang, Qi Z <qi.z.zhang@intel.com>
> > > > > Sent: Tuesday, May 31, 2022 7:52 PM
> > > > > To: Zhang, Peng1X <peng1x.zhang@intel.com>; Yang, Qiming
> > > > > <qiming.yang@intel.com>; dev@dpdk.org
> > > > > Cc: stable@dpdk.org
> > > > > Subject: RE: [PATCH v4] net/ice: retry sending adminQ command
> > > > > after failure
> > > > >
> > > > >
> > > > >
> > > > > > -----Original Message-----
> > > > > > From: Zhang, Peng1X <peng1x.zhang@intel.com>
> > > > > > Sent: Wednesday, June 1, 2022 1:48 AM
> > > > > > To: Yang, Qiming <qiming.yang@intel.com>; Zhang, Qi Z
> > > > > > <qi.z.zhang@intel.com>; dev@dpdk.org
> > > > > > Cc: Zhang, Peng1X <peng1x.zhang@intel.com>; stable@dpdk.org
> > > > > > Subject: [PATCH v4] net/ice: retry sending adminQ command
> > > > > > after failure
> > > > > >
> > > > > > From: Peng Zhang <peng1x.zhang@intel.com>
> > > > > >
> > > > > > The origin design is if error happen during the step 3 of
> > > > > > following given situation, it will return error directly
> > > > > > without retry. While in current patch, it will retry again and
> > > > > > again during certain
> > > time.
> > > > > > If retry succeed, rule can be continuously created. It will
> > > > > > improve success rate of creating rule under following given situation.
> > > > > >
> > > > > > The given situation as following steps show:
> > > > > > step 1. Kernel PF and DCF are ready at the beginning.
> > > > > > step 2. A VF reset happen, kernel send an event to DPDK DCF
> > > > > > and set STATE to pause.
> > > > > > step 3. Before DPDK DCF receive the event, it is possible a
> > > > > > rule creation is ongoing and switch rules, recipe, or vsi list
> > > > > > related adminQ operation is executing.
> > > > > > step 4. Then result of operation is failure, it will lead to
> > > > > > error return to DPDK DCF. DPDK DCF error code will be set as
> > > > > > EINVAL, not
> > > > EAGAIN.
> > > > > >
> > > > > > Fixes: 6bad5047be24 ("net/ice/base: return correct error
> > > > > > code")
> > > > > > Fixes: 453d087ccaff ("net/ice/base: add common functions")
> > > > > > Cc: stable@dpdk.org
> > > > > >
> > > > > > Signed-off-by: Peng Zhang <peng1x.zhang@intel.com>
> > > > > > ---
> > > > > >  v4 changes:
> > > > > >  - Add retry mechanism if fail to send adminQ command under
> > > > > > given
> > > > > situation.
> > > > > >  v3 Changes:
> > > > > >  - Add the situation description, expected error code and
> > > > > > incorrect error code
> > > > > >  - in commit log.
> > > > > >  v2 Changes:
> > > > > >  - Modify DCF state checking mechanism.
> > > > > >
> > > > > >  drivers/net/ice/base/ice_common.c |  2 +-
> > > > > > drivers/net/ice/base/ice_switch.c |
> > > > > > 46 +++++++++++++++++++++++++++----
> > > > > > drivers/net/ice/base/ice_switch.h
> > > > > > |  5
> > > > > > ++++
> > > > > >  3 files changed, 46 insertions(+), 7 deletions(-)
> > > > > >
> > > > > > diff --git a/drivers/net/ice/base/ice_common.c
> > > > > > b/drivers/net/ice/base/ice_common.c
> > > > > > index db87bacd97..013c255371 100644
> > > > > > --- a/drivers/net/ice/base/ice_common.c
> > > > > > +++ b/drivers/net/ice/base/ice_common.c
> > > > > > @@ -2127,7 +2127,7 @@ ice_aq_alloc_free_res(struct ice_hw *hw,
> > > > > > u16 num_entries,
> > > > > >
> > > > > >  	cmd->num_entries = CPU_TO_LE16(num_entries);
> > > > > >
> > > > > > -	return ice_aq_send_cmd(hw, &desc, buf, buf_size, cd);
> > > > > > +	return ice_aq_retry_send_cmd(hw, &desc, buf, buf_size, cd);
> > > > >
> > > > > this is the fix only for DCF, we don't need to retry in a PF driver context.
> > > > > Better keep the same function name, but implements the retry
> > > > > mechanism inside the function, and should only be triggered
> > > > > when it is in a DCF context