Re: [PULL 59/68] aspeed: Add an AST2600 eval board

[Joel Stanley]

On Tue, 15 Oct 2019 at 17:55, Peter Maydell <address@hidden> wrote:
>
> On Tue, 15 Oct 2019 at 18:43, Cédric Le Goater <address@hidden> wrote:
> > On 15/10/2019 19:03, Peter Maydell wrote:
> > > On Mon, 14 Oct 2019 at 17:05, Peter Maydell <address@hidden> wrote:
> > > (armhf)pmaydell@mustang-maydell:~/qemu$
> > > ./build/all-a32/arm-softmmu/qemu-system-arm -M ast2600-evb
> > > qemu-system-arm: at most 2047 MB RAM can be simulated
> > >
> > > It's also a pretty rudely large amount of RAM to allocate
> > > by default: it caused 'make check' to fail on my OSX
> > > box, which is 64-bits but doesn't have huge swathes
> > > of free RAM.
> > >
> > > I'm going to drop this patch from my queue and redo
> > > the pullreq.
> >
> > ok. We do have such a board.
> >
> > What do you suggest ? We can lower the RAM to 1G for QEMU.
>
> 1GB is OK -- we have several machines that set default_ram_size to that.
>
> If we want to handle more generally boards which have a
> larger ram size by default then we probably need to
> work on the 'make check' infrastructure -- right now we
> have a generic test that just checks "can we instantiate
> every machine model", which is what's falling over.

I hit this when bumping the RAM size for the powernv machine too. In
that case it was a convenient size given the memory layout we have, so
changing to be less than 2GB was not making the model less accurate.

For the ast2600evb machine, the board actually has 2GB of RAM, and
it's useful for guests to test in this environment. I'd not had any
reports of it failing to launch due to a failed memory allocation,
which I guess is indicative of the systems people who use the model
have.

Peter, did you have any thoughts on how to exclude such guests from
testing in environments where memory is limited?

Cheers,

Joel