[Qemu-devel] VersatilePB bug (or undefined behaviour)

[Qemu-devel] VersatilePB bug (or undefined behaviour)

[Marc Marí]

Hi

I was investigating a bit with a versatilepb machine, and I was trying
to boot a Linux system with a root file system through NFS. It worked
with Linux 3.2, but it didn't work with any later Linux version. This
was the error:

[    0.594471] smc91x: not found (-16).
[    0.595555] smc91x: probe of smc91x.0 failed with error -16

(Error -16 is EBUSY).

I debugged the error, and I found that the region was being marked as
"System RAM".

The problem was mine, I specified (because I did not change if from
previous boots) 1G as memory for the Versatile machine, which only has
256 MB of RAM

Should this be a bug, or is undefined behaviour? I think it is not very
difficult to check for the memory the user requests so it can be
limited to 256 MB and issue a warning if it is exceeded.

Thanks
Marc

Re: [Qemu-devel] VersatilePB bug (or undefined behaviour)

[Peter Maydell]

On 27 January 2015 at 20:04, Marc Marí <marc.mari.barcelo@gmail.com> wrote:
> The problem was mine, I specified (because I did not change if from
> previous boots) 1G as memory for the Versatile machine, which only has
> 256 MB of RAM
>
> Should this be a bug, or is undefined behaviour? I think it is not very
> difficult to check for the memory the user requests so it can be
> limited to 256 MB and issue a warning if it is exceeded.

It's user error to specify more RAM than the board can handle.
I tried some years ago to get a patch added which allowed boards
to say how much RAM they could deal with so we could report this
more helpfully, but it got shot down as insufficiently generic
(not dealing with NUMA, weird boards with memory holes, boards
that only want you to add RAM in 512K chunks, etc, etc) and I
gave up.

-- PMM