[PATCH 01/16] KVM: arm64: Generalise VM features into a set of flags

[Steven Price]

On 28/07/2021 10:41, Marc Zyngier wrote:
> On Tue, 27 Jul 2021 19:10:27 +0100,
> Will Deacon <will at kernel.org> wrote:
>>
>> On Thu, Jul 15, 2021 at 05:31:44PM +0100, Marc Zyngier wrote:
>>> We currently deal with a set of booleans for VM features,
>>> while they could be better represented as set of flags
>>> contained in an unsigned long, similarily to what we are
>>> doing on the CPU side.
>>>
>>> Signed-off-by: Marc Zyngier <maz at kernel.org>
>>> ---
>>>  arch/arm64/include/asm/kvm_host.h | 12 +++++++-----
>>>  arch/arm64/kvm/arm.c              |  5 +++--
>>>  arch/arm64/kvm/mmio.c             |  3 ++-
>>>  3 files changed, 12 insertions(+), 8 deletions(-)
>>>
>>> diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
>>> index 41911585ae0c..4add6c27251f 100644
>>> --- a/arch/arm64/include/asm/kvm_host.h
>>> +++ b/arch/arm64/include/asm/kvm_host.h
>>> @@ -122,7 +122,10 @@ struct kvm_arch {
>>>  	 * should) opt in to this feature if KVM_CAP_ARM_NISV_TO_USER is
>>>  	 * supported.
>>>  	 */
>>> -	bool return_nisv_io_abort_to_user;
>>> +#define KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER	0
>>> +	/* Memory Tagging Extension enabled for the guest */
>>> +#define KVM_ARCH_FLAG_MTE_ENABLED			1
>>> +	unsigned long flags;
>>
>> One downside of packing all these together is that updating 'flags' now
>> requires an atomic rmw sequence (i.e. set_bit()). Then again, that's
>> probably for the best anyway given that kvm_vm_ioctl_enable_cap() looks
>> like it doesn't hold any locks.
> 
> That, and these operations are supposed to be extremely rare anyway.
> 
>>
>>>  	/*
>>>  	 * VM-wide PMU filter, implemented as a bitmap and big enough for
>>> @@ -133,9 +136,6 @@ struct kvm_arch {
>>>  
>>>  	u8 pfr0_csv2;
>>>  	u8 pfr0_csv3;
>>> -
>>> -	/* Memory Tagging Extension enabled for the guest */
>>> -	bool mte_enabled;
>>>  };
>>>  
>>>  struct kvm_vcpu_fault_info {
>>> @@ -777,7 +777,9 @@ bool kvm_arm_vcpu_is_finalized(struct kvm_vcpu *vcpu);
>>>  #define kvm_arm_vcpu_sve_finalized(vcpu) \
>>>  	((vcpu)->arch.flags & KVM_ARM64_VCPU_SVE_FINALIZED)
>>>  
>>> -#define kvm_has_mte(kvm) (system_supports_mte() && (kvm)->arch.mte_enabled)
>>> +#define kvm_has_mte(kvm)					\
>>> +	(system_supports_mte() &&				\
>>> +	 test_bit(KVM_ARCH_FLAG_MTE_ENABLED, &(kvm)->arch.flags))
>>
>> Not an issue with this patch, but I just noticed that the
>> system_supports_mte() check is redundant here as we only allow the flag to
>> be set if that's already the case.
> 
> It allows us to save a memory access if system_supports_mte() is false
> (it is eventually implemented as a static key). On the other hand,
> there is so much inlining due to it being a non-final cap that we
> probably lose on that too...

My original logic was that system_supports_mte() checks
IS_ENABLED(CONFIG_ARM64_MTE) - so this enables the code guarded with
kvm_has_mte() to be compiled out if CONFIG_ARM64_MTE is disabled.

Indeed it turns at we currently rely on this (with CONFIG_ARM64_MTE
disabled):

aarch64-linux-gnu-ld: arch/arm64/kvm/mmu.o: in function `sanitise_mte_tags':
/home/stepri01/work/linux/arch/arm64/kvm/mmu.c:887: undefined reference to `mte_clear_page_tags'
aarch64-linux-gnu-ld: arch/arm64/kvm/guest.o: in function `kvm_vm_ioctl_mte_copy_tags':
/home/stepri01/work/linux/arch/arm64/kvm/guest.c:1066: undefined reference to `mte_copy_tags_to_user'
aarch64-linux-gnu-ld: /home/stepri01/work/linux/arch/arm64/kvm/guest.c:1074: undefined reference to `mte_copy_tags_from_user'

Obviously we could pull just the IS_ENABLED() into kvm_has_mte() instead.

Steve