On Thu, Aug 2, 2012 at 1:37 PM, Christoffer Dall <c.dall@xxxxxxxxxxxxxxxxxxxxxx> wrote:
On Thu, Aug 2, 2012 at 12:25 PM, Rusty Russell <rusty.russell@xxxxxxxxxx> wrote:you probably also have Kernel Debugging turned on...
> On Thu, 2 Aug 2012 01:14:21 +0200, Antonios Motakis <a.motakis@xxxxxxxxxxxxxxxxxxxxxx> wrote:
>> On Thu, Aug 2, 2012 at 12:12 AM, Christoffer Dall <
>> c.dall@xxxxxxxxxxxxxxxxxxxxxx> wrote:
>> > When compiling the v10-stage kernel in ARM mode with this patch applied, I
>> > get:
>> >
>> > arch/arm/kvm/interrupts.S: Assembler messages:
>> > arch/arm/kvm/interrupts.S:691: Error: invalid constant (438) after fixup
>> > arch/arm/kvm/interrupts.S:697: Error: invalid constant (424) after fixup
>> > make[1]: *** [arch/arm/kvm/interrupts.o] Error 1
>> >
>>
>> Ok, I will try to reproduce this on v10-stage. Thanks.
>
> I got a similar problem when I applied your patch on top of the vgic
> stuff.
On an unrelated note, Antionios, how was this tested? Did you run
>
> Moving the cpumask to the end solved it for my case, but that was for
> offset 0x4c4 IIRC: this looks different. We may have to rewrite that
> part of the asm to use multiple instructions...
>
SoftFloat or something equivalent in the guest? If you have a script
and some binaries, can you provide this?
I have some floating point apps to run in the guest, however it is noteworthy that most VFP enabled apps will work even without VFP support. They will just silently mess with the host's registers. When I first wrote the vfp patch, I used some code in the host to detect when the host's registers would change values; what I do today, is mostly add a lot of breakpoints in the modeldebugger and watch how the code behaves, compare before and after switching register values etc.
I'm definitely interested in more ways to test this, but I do feel this is stable enough (I've had reports from Alexander that he can run Android guests that would hopelessly crash before).
Some code paths will not be testable though on the fast model though, just because we won't be able to test different versions and configurations of VFP, etc.
-Christoffer
_______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/cucslists/listinfo/kvmarm
