Re: [dpdk-stable] [dpdk-dev] [PATCH] devtools: skip the symbol check when map file under drivers

[Thomas Monjalon <thomas@monjalon.net>]

22/05/2019 15:40, Neil Horman:
> On Wed, May 22, 2019 at 01:12:34PM +0000, Jerin Jacob Kollanukkaran wrote:
> > From: Neil Horman <nhorman@tuxdriver.com>
> > > On Wed, May 22, 2019 at 03:05:54AM +0000, Jerin Jacob Kollanukkaran
> > > wrote:
> > > > From: Neil Horman <nhorman@tuxdriver.com>
> > > > > On Wed, May 22, 2019 at 01:26:28AM +0530, jerinj@marvell.com wrote:
> > > > > > From: Jerin Jacob <jerinj@marvell.com>
> > > > > >
> > > > > > Drivers do not have ABI.
> > > > > > Skip the symbol check if map file under drivers directory.
> > > > > >
> > > > > > Fixes: 4bec48184e33 ("devtools: add checks for ABI symbol
> > > > > > addition")
> > > > > >
> > > > > > Cc: stable@dpdk.org
> > > > > > Cc: Neil Horman <nhorman@tuxdriver.com>
> > > > > >
> > > > > Sorry, but I'm not ok with this, because many of our DPDK PMDs have
> > > > > functions that get exported which are meant to be called by
> > > > > applications directly.  The
> > > >
> > > > OK. Just to update my knowledge, Should those API needs to go through
> > > > ABI/API depreciation process?
> > > >
> > > Yes, they definately should, they are API's just as any other in the core DPDK
> > > library.
> > 
> > OK
> > 
> > 
> > > 
> > > > Actually, I am concerned about the APIs, which is called between
> > > > drviers not the application. For example,
> > > > drivers/common/dpaax/rte_common_dpaax_version.map
> > > >
> > > > it is not interface to application rather it is for intra driver case.
> > > > I think, I can change my logic to Skip the symbols which NOT starting with
> > > rte_.
> > > > Agree?
> > > >
> > > No, Thats just one case, and if those calls are between drivers, so be it, but
> > > those still need to be stable, and we have other examples (like the bonding
> > > or dummy driver), which have additional APIs that are explicitly meant to be
> > > used by an application.
> > 
> > There is no disagreement on the API that exposed to application.
> > I am concerned with internal driver APIs. For example, I am getting following warning
> > 
> > ERROR: symbol otx2_mbox_alloc_msg_rsp is added in the DPDK_19.05 section, but is expected to be added in the EXPERIMENTAL section of the version map
> > 
> Thats a warning about the fact that you added an API call in a versioned section
> of a library instead of the Experimental section, thats part of our policy.  New
> APIs need to go through the experimental tag first.
> 
> I understand what you are saying about driver only apis, and I mentioned that in
> my other email farther down the thread.  The problem is that "driver only apis"
> are currently just a conceptual thing for us to discuss.  They're still,
> practially speaking, API's that any downstream user can access and become
> dependent on, which we need to manage, either by keeping the API stable, so it
> stays usable for all callers, or by developing a way to mark driver only API's
> as such.  I proposed a method that you might use to do the latter in my other email.
> 
> > This API suppose to be called only a octeontx2 network driver from octeontx2 common driver
> > i.e application should not expect any stability on intra driver functions or it does not meant to
> > be used by application.
> > 
> Ok, but again, your assertion is that its driver to driver only, but in
> practicaility, that assertion is irrelevant.  Those symbols are still exposed
> for general use, and weather or not you say they aren't part of the ABI, the
> fact of the matter is, there is no way to tell the difference from a linked
> object standpoint.  Instead of hobbling the tool to just not scan anything, you
> need to find a way to differentiate these symbols, so that you can enforce your
> assertion that there are restrictions on where these APIs are called from.
> 
> Neil
> 
> > Thomas,
> > Any thought on this?

As Neil said, we need to differentiate the internal APIs.
We already have this issue in a number of places like EAL, or ethdev,
and it was poorly addressed with some comments like "@internal".

Practically I don't care about stability of these internal functions,
but I agree that it creates a mess in the tooling and confuse users.