On 28.10.2011, at 03:56, David Gibson wrote: > kvmppc_read_int_cpu_dt() returns -1 on errors. However > kvmppc_host_cpu_def() assumes that it can use the return value > directly for setting the corresponding instruction support bits. This > will give invalid results when kvmppc_read_int_cpu_dt() returns -1 > (i.e. 0xffffffff). > > In fact, it gets worse: the common case for -1 return values is when > the properties are simply not present. On some machines this means > the features are not available but in other cases it just means that > the firmware on this platform does not provide this information. So, > when there's an error reading the properties we really can't tell if > vmx or dfp is supported. > > Therefore, this patch checks for -1 return values and falls back to > qemu's built in table of CPU capabilities in that case. > > Reported-by: Nishanth Aravamudan <n...@us.ibm.com> > Signed-off-by: David Gibson <da...@gibson.dropbear.id.au>
I already have this patch in my tree, though with different author. I've added Nishanth as Reported-by on it now and you as Acked-by, if you're ok with that. Alex
