On 24 January 2013 19:57, Peter Maydell
<address@hidden> wrote:
It appears to be one of the reasons. The other is that
the kernel is trashing the device tree because it
happens to be sitting in the same page as the last
part-page of the initrd, and free_initrd_mem() rounds
up to a page boundary when poisoning memory after
the initrd has been uncompressed.
I don't know if this is related, but in the past I remember QEMU behaving quite differently when the DTB was appended to the kernel rather than provided in the QEMU arguments (The latter would crash contrary to the first).