Re: [dpdk-dev] DPDK Release Status Meeting 16/01/2020

* Re: [dpdk-dev] DPDK Release Status Meeting 16/01/2020
@ 2020-01-17 11:39 Ananyev, Konstantin
  0 siblings, 0 replies; 10+ messages in thread
From: Ananyev, Konstantin @ 2020-01-17 11:39 UTC (permalink / raw)
  To: Yigit, Ferruh, Akhil Goyal, dpdk-dev; +Cc: Thomas Monjalon

Hi lads,

> >>>>> * next-net-crypto
> >>>>>   * Pull request sent
> >>>>>   * There is a performance concern on some ipsec-gw patches,
> >>>>>     they can go in -rc2 if the issue is solved
> >>>>>   * CPU crypto from last release may be breaking ABI, need to confirm
> >>>>
> >>>> AFAIK, there is no ABI breakage.
> >>>
> >>> This is the output of validate-abi.sh.
> >>>
> >>> 	Change 							Effect
> >>> 1 Field sym_cpu_process has been added to this type. 	              1) This field will
> >> not be initialized by old clients.
> >>>                                                                                                                    2) Size of the
> >> inclusive type has been changed.
> >>>
> >>> 								NOTE: this
> >> field should be accessed only from the new library
> >>> functions, otherwise it may result in crash or incorrect behavior of applications.
> >>> 2 Size of this type has been changed from 128 bytes to 136 bytes. 	The
> >> fields or parameters of such data type may be incorrectly
> >>> initialized or accessed by old client applications.
> >>
> >> This is struct rte_cryptodev_ops, which is AFAIK, not part of public API.
> >> So not sure, why do you consider it as ABI breakage?
> >>
> >
> > If this is not an issue, than I am fine with it.
> 
> The ABI change between cryptodev and PMDs are allowed, that is contained within
> DPDK and not a user interface [1].
> 
> [1] Unless some inline functions are directly accessing the dev_ops, as
> (unfortunately) done in the ethdev.

Thanks Ferruh for confirmation.
For cryptodev we don't have such inline functions, plus the we add
new entry at the bottom of struct rte_cryptodev_ops, so I believe we
are safe here.

> 
> >
> >>>
> >>> Apart from that, IPSEC also has breakage, but that is experimental, so not an
> >> issue.
> >>>
> >>>>
> >>>>>     and discussed dummy variable is missing, may be postponed to next
> >> release
> >>>>
> >>>> Not sure I understand last sentence, could the author explain
> >>>> what dummy variables we are talking about.
> >>>
> >>> In one of the techboard meeting around 19.11 timeframe, during the
> >> discussion for approving methodology for CPU-crypto, it was
> >>> proposed that in order to avoid delay, a dummy variable can be introduced in
> >> cryptodev API/ABI to avoid any ABI breakage in
> >>> upcoming releases. But this was not done.
> >>
> >> Dummy variable for what?
> >> If you are talking about sym_cpu_process - we just added it into
> >> rte_cryptodev_ops, instead of
> >> ' struct rte_cryptodev' instead.
> >> That way we avoid any ABI breakage without introducing any churn in
> >> rte_cryptodev itself , see above.
> >
> > Then why was there so much resistance on this approach when there is no ABI breakage.
> > I thought there was ABI breakage because of this change.
> >
> > BTW this patchset is a bit late and it came after merge deadline 15 Jan.
> > Ideally all library related patches should go in RC1.
> > I would check if I could make it to the RC2.
> > I have 3 IPSec series to work on before RC2.
> >

^ permalink raw reply	[flat|nested] 10+ messages in thread