[OpenPOWER-HDL-Cores] https://bugzilla.kernel.org/show_bug.cgi?id=213257

Benjamin Herrenschmidt benh at kernel.crashing.org
Sat May 29 00:21:25 UTC 2021


On Fri, 2021-05-28 at 12:51 +0100, Luke Kenneth Casson Leighton wrote:
> not sure where this might best else be sent.
> 
> Lauri has written a mini kvm runner for ppc64 with the intention to
> allow single-step execution where a memory dump and full register
> dump
> may be carried out after every single instruction.  it manages around
> 35k instructions per second this way.
> 
> this to be able to compare against hardware, qemu, simulators,
> anything and everything, running benchmarks unit tests full programs
> full virtualised OSes, anything.
> 
> unfortunately, all FP operations after single-step through the entire
> program are zero contents.
> 
> we suspect that insufficient context / state is saved/restored on the
> step.
> 
> is this a known issue or something that was overlooked?

You man in qemu ? You have MSR:FP set ?

Cheers,
Ben.




More information about the OpenPOWER-HDL-Cores mailing list