Re: [PATCH 2/2] automation: arm64: Create a test job for testing static allocation on qemu

[Julien Grall <julien@xxxxxxx>]

On 08/07/2022 10:00, Xenia Ragiadakou wrote:
> > > > > +
> > > > > +(grep -q "Xen dom0less mode detected" qemu-staticmem.serial) || 
> > > > > exit 1
> > > > > +
> > > > > +for ((i=0; i<${#base[@]}; i++))
> > > > > +do
> > > > > +    start="$(printf "0x%016x" ${base[$i]})"
> > > > > +    end="$(printf "0x%016x" $((${base[$i]} + ${size[$i]} - 1)))"
> > > > > +    grep -q "node   0: \[mem ${start}-${end}\]" qemu-staticmem.serial
> > > > > +    if test $? -eq 1
> > > > > +    then
> > > > > +        exit 1
> > > > > +    fi
> > > > > +done
> > > > Please add a comment on top to explain what this is meant to do. 
> > > > However, I think we should avoid relying on output that we have 
> > > > written ourself. IOW, relying on Xen/Linux to always write the same 
> > > > message is risky because they can change at any time.
> > > The kernel is taken from a test-artifact container, so, IIUC, it 
> > > won't change.
> > This statement is correct today. However, we may decide to update the 
> > kernel or test multiple kernels (with different ouput).
> > In the first case, it would be a matter of updating the script. This 
> > is annoying but not too bad. In the second case, we would need to have 
> > "if version X ... else if version Y ... ".
> The particular test was relying and had a dependency on this kernel.
I think you missed my point. I don't disagree that the test today 
expects a specific version. However, there are nothing preventing us to 
change that so we long we have a matching initramfs/kernel.
> If the test is merged into the qemu-smoke-arm64.sh, the check above will 
> leave and it will be tested whether the guest makes it to the busybox, 
> based on the busybox logs, which also may change at any time.
I don't think I suggested that relying on busybox is better. Ideally we 
should run a script that prints a message.
But this is not something I am going to argue for in this patch. Relying 
on one piece of line is already far better than trying to check logs for 
each components (e.g. xen, kernel, busysbox).
Cheers,

--
Julien Grall