Re: [PATCH net-next v2 2/2] net: ipa: fix IPA validation

From: Leon Romanovsky <leon@kernel.org>
To: Alex Elder <elder@linaro.org>
Cc: davem@davemloft.net, kuba@kernel.org, andrew@lunn.ch,
	bjorn.andersson@linaro.org, evgreen@chromium.org,
	cpratapa@codeaurora.org, subashab@codeaurora.org,
	elder@kernel.org, netdev@vger.kernel.org,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH net-next v2 2/2] net: ipa: fix IPA validation
Date: Mon, 22 Mar 2021 16:17:24 +0200	[thread overview]
Message-ID: <YFim9CWuu45FcdcN@unreal> (raw)
In-Reply-To: <f152c274-6fe0-37a1-3723-330b7bfe249a@linaro.org>

On Mon, Mar 22, 2021 at 08:17:59AM -0500, Alex Elder wrote:
> On 3/22/21 1:40 AM, Leon Romanovsky wrote:
> > > I'd like to suggest a plan so I can begin to make progress,
> > > but do so in a way you/others think is satisfactory.
> > > - I would first like to fix the existing bugs, namely that
> > >    if IPA_VALIDATION is defined there are build errors, and
> > >    that IPA_VALIDATION is not consistently used.  That is
> > >    this 2-patch series.
> > The thing is that IPA_VALIDATION is not defined in the upstream kernel.
> > There is nothing to be fixed in netdev repository
> > 
> > > - I assure you that my goal is to simplify the code that
> > >    does this sort of checking.  So here are some specific
> > >    things I can implement in the coming weeks toward that:
> > >      - Anything that can be checked at build time, will
> > >        be checked with BUILD_BUG_ON().
> > +1
> > 
> > >      - Anything checked with BUILD_BUG_ON() will*not*
> > >        be conditional.  I.e. it won't be inside an
> > >        #ifdef IPA_VALIDATION block.
> > >      - I will review all remaining VALIDATION code (which
> > >        can't--or can't always--be checked at build time),
> > >        If it looks prudent to make it*always*  be checked,
> > >        I will make it always be checked (not conditional
> > >        on IPA_VALIDATION).
> > +1
> > 
> > > The result should clearly separate checks that can be done
> > > at build time from those that can't.
> > > 
> > > And with what's left (especially on that third sub-bullet)
> > > I might have some better examples with which to argue
> > > for something different.  Or I might just concede that
> > > you were right all along.
> > I hope so.
> 
> I came up with a solution last night that I'm going to try
> to implement.  I will still do the things I mention above.
> 
> The solution is to create a user space tool inside the
> drivers/net/ipa directory that will link with the kernel
> source files and will perform all the basic one-time checks
> I want to make.

We are not talking about putting this tool in upstream repo, right?
If yes, please get buy-in from David/Jakub first.

Thanks