On Fri, Nov 30, 2012 at 3:48 PM, Marc Zyngier <marc.zyngier@xxxxxxx> wrote:
Indeed you are right, TIDCP does trap L2CTLR on the hardware... however it doesn't trap on the Fast Models, which lead to this confusion from our side. Since the L2CTLR is not mentioned by name in the spec, we just assumed the model's behavior was canonical, it didn't occur to us at that point to also test it on the hardware.
Anyway sorry for jumping the gun. However the Fast Model definitely behaves out of spec in this case, so at least we know about that now :)
Best regards,
On 30/11/12 14:35, Peter Maydell wrote:We already have HCR_TIDCP set in HCR, which should be enough (see ARM
> On 30 November 2012 09:31, Claudio Fontana <Claudio.Fontana@xxxxxxxxxx> wrote:
>> all the handlers are already there in coproc_a15.c, but the traps
>> are not enabled at world switch in __kvm_vcpu_run.
>> In particular this was noticed because the guests were trying to
>> read the number of CPUs using l2ctlr, and the number of physical
>> host CPUs was returned instead of the number of VCPUs.
>
> L2CTLR should be trapped via HCR.TIDCP, not the HSTR.
> Traps for whole swathes of the coprocessor space via HSTR
> should be avoided unless absolutely necessary; the functionally
> grouped traps in HCR are much better.
>
> (In fact I thought we already did this? Marc?)
ARM B1.14.3).
You should definitely see L2CTLR accesses being trapped with this
configuration. What is your hardware?
Indeed you are right, TIDCP does trap L2CTLR on the hardware... however it doesn't trap on the Fast Models, which lead to this confusion from our side. Since the L2CTLR is not mentioned by name in the spec, we just assumed the model's behavior was canonical, it didn't occur to us at that point to also test it on the hardware.
Anyway sorry for jumping the gun. However the Fast Model definitely behaves out of spec in this case, so at least we know about that now :)
--
Antonios Motakis
Virtual Open Systems
Open Source KVM Virtualization Developments
Multicore Systems Virtualization Porting Services
Web: www.virtualopensystems.com
Virtual Open Systems Confidential and Proprietary Information. The
content of this e-mail message, and any documents, files or e-mail
messages attached to it, may contain confidential information. If you are not the intended recipient, or a person
entitled for delivering it to the intended
recipient, you are hereby notified that any disclosure, copying,
distribution or use of any of the information contained in or attached
to this message is not permitted in any way and prohibited.
_______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/cucslists/listinfo/kvmarm
