On 13.04.14 07:56, Benjamin Herrenschmidt wrote:
On Sun, 2014-04-13 at 00:38 +0100, Peter Maydell wrote:
From a design standpoint I find that totally retarded btw :-)
I agree it's bonkers; it's just that fixing it requires a big pile
of design and implementation work from somebody; and in
the meantime "both ends must be configured identically" is
at least a consistent and coherent position...
Sure I'm fine with that. I was just advocating for putting the TB
value in the stream so we can barf if it differs but I don't care
*that* much if it's not there.
I think on some chips it might be configurable so it won't hurt....
So for chips that might be configurable, configuration has to happen via
some SPR, right? Wouldn't it make sense to transfer that configuration
SPR on these and thus the configuration as well?
Really, the main problem I see with adding the tb frequency into the
migration stream is that we'd change the migration format. Worst case it
would be a few useless bytes on migration - not a big deal. But I don't
think we support any case where it would actually be required - it'd
only be a safety net.
Alex
