[PATCH] KVM: arm/arm64: Access CNTHCTL_EL2 bit fields correctly

Marc Zyngier marc.zyngier at arm.com
Tue Nov 29 01:36:58 PST 2016


On 29/11/16 03:28, Jintack Lim wrote:
> On Mon, Nov 28, 2016 at 1:39 PM, Marc Zyngier <marc.zyngier at arm.com> wrote:
>> On 28/11/16 17:43, Marc Zyngier wrote:
>>> Hi Jintack,
> 
> Hi Marc,
> 
>>>
>>> On 28/11/16 16:46, Jintack Lim wrote:
>>>> Bit positions of CNTHCTL_EL2 are changing depending on HCR_EL2.E2H bit.
>>>> EL1PCEN and EL1PCTEN are 1st and 0th bits when E2H is not set, but they
>>>> are 11th and 10th bits respectively when E2H is set.  Current code is
>>>> unintentionally setting wrong bits to CNTHCTL_EL2 with E2H set, which
>>>> may allow guest OS to access physical timer. So, fix it.
>>>>
>>>> Signed-off-by: Jintack Lim <jintack at cs.columbia.edu>
>>>> ---
>>>>  arch/arm/include/asm/kvm_timer.h     | 33 +++++++++++++++++++
>>>>  arch/arm64/include/asm/kvm_timer.h   | 62 ++++++++++++++++++++++++++++++++++++
>>>>  include/clocksource/arm_arch_timer.h |  6 ++--
>>>>  virt/kvm/arm/hyp/timer-sr.c          |  8 ++---
>>>>  4 files changed, 103 insertions(+), 6 deletions(-)
>>>>  create mode 100644 arch/arm/include/asm/kvm_timer.h
>>>>  create mode 100644 arch/arm64/include/asm/kvm_timer.h
>>>>
>>
>> [...]
>>
>>> We could make it nicer (read "faster") by introducing a
>>> hyp_alternate_select construct that only returns a value instead
>>> of calling a function. I remember writing something like that
>>> at some point, and dropping it...
>>
>> So here's what this could look like (warning, wacky code ahead,
>> though I fixed a stupid bug that was present in the previous patch).
>> The generated code is quite nice (no branch, only an extra mov
>> instruction on the default path)... Of course, completely untested!
> 
> This looks much cleaner than my patch.
> While we are at it, is it worth to consider that we just need to set
> those bits once for VHE case, not for every world switch as an
> optimization?

Ah! That's a much better idea indeed! And we could stop messing with
cntvoff_el2 as well, as it doesn't need to be restored to zero on exit.
Could you try and respin something along those lines?

Thanks,

	M.
-- 
Jazz is not dead. It just smells funny...



More information about the linux-arm-kernel mailing list