Re: [PATCH v6 0/2] lib: checksum: Fix issues with checksum tests

[Palmer Dabbelt]

On Wed, 07 Feb 2024 18:44:42 PST (-0800), linux@xxxxxxxxxxxx wrote:
> On 2/7/24 18:09, Charlie Jenkins wrote:
> > On Wed, Feb 07, 2024 at 05:45:22PM -0800, Andrew Morton wrote:
> > > On Wed, 07 Feb 2024 16:22:49 -0800 Charlie Jenkins <charlie@xxxxxxxxxxxx> wrote:
> > >
> > > > The ip_fast_csum and csum_ipv6_magic tests did not have the data
> > > > types properly casted, and improperly misaligned data.
> > >
> > > Neither this nor the two patch's changelogs describe *why* these changes
> > > are needed.  They merely assert that we need to do this.
> > >
> > > IOW, when fixing a bug please always describe the user-visible effects
> > > of that bug.
> >
> > I can add a description that says that the tests are being fixed because
> > they caused failures on systems without misaligned access support. As
> > for the casting patch it's a bit less pertinent but I can add that it
> > allows the code to pass sparse checks.
> >
> > - Charlie
> I don't know exactly what Andrew is asking for, but maybe including the
> error log from the failed selftests and/or the sparse errors would be
> sufficient ?
>
> Not sure though if any of those count as "user visible".

Ya, for compiler warning/error workarounds I usually just include 
something like "without this, I get $ERROR".  Something like 
28ea54bade76 ("RISC-V: Don't rely on positional structure 
initialization").

For the aligned access on there was a fairly interesting discussion on 
why this hadn't tripped up before, I forget if it was on LKML or IRC (or 
Slack or just in the office).  That's worth including...