Re: arm: kvm-selftest

[Christoffer Dall <christoffer.dall@linaro.org>]

On Wed, Jul 31, 2013 at 06:21:37PM -0400, Andrew Jones wrote:
> Hi Christoffer,
> 
> I'm just confirming that I found the right repo/branch for the
> most recent kvmarm unit tests. I'm looking at
> 
> github.com/virtualopensystems/linux-kvm-arm kvm-selftest
> 
> with last commit
> 
> commit b5aec35ee58c6b2d726a3e2fb56ec0e6e89976e0
> Author: Christoffer Dall <cdall@cs.columbia.edu>
> Date:   Fri Feb 15 18:37:24 2013 -0500
> 
>     WIP: These settings are useful for some more profiling
>     
>     This really needs to be cleaned up and stuff, but the whole
>     testing here has bit-rotted and should be picked up by someone.
>     
>     I know....
>     
>     Signed-off-by: Christoffer Dall <cdall@cs.columbia.edu>
> 
> If this is the most recent branch, then I guess I'll volunteer
> myself to be the someone to pick it up and see about integrating
> it into kvm-unit-tests.
> 
Hi Drew,

Yes, this is the right repo and the latest commit.  You may in fact
benefit from looking at a slightly earlier state of the repo which may
be a bit more clean.  There may also be some newer commits in some
*-selftest*-named branches in https://github.com/columbia/linux-kvm-arm.
At least, at some point I added a test that measures IPI cost similarly
to what was done on x86.

Note that the simple VMEXIT count probably doesn't work out-of-the-box,
because reading the cycle counter on ARM requires access to the PMU
registers, which are current trapped and RAZ.  I haven't looked closely
at the provisions in the ARM ARM and the TRMs to see if there's some way
we can allow the guest access to these.  Otherwise a PV interface may be
necessary.  Will Deacon can probably help us answer these questions.

-Christoffer