Re: [PATCH] KVM: s390: introduce module parameter kvm.use_gisa

[Michael Mueller <mimu@linux.ibm.com>]

On 27.02.20 13:31, David Hildenbrand wrote:
> On 27.02.20 13:04, Michael Mueller wrote:
>>
>>
>> On 27.02.20 10:26, David Hildenbrand wrote:
>>> On 27.02.20 10:10, Michael Mueller wrote:
>>>> The boolean module parameter "kvm.use_gisa" controls if newly
>>>> created guests will use the GISA facility if provided by the
>>>> host system. The default is yes.
>>>>
>>>>     # cat /sys/module/kvm/parameters/use_gisa
>>>>     Y
>>>>
>>>> The parameter can be changed on the fly.
>>>>
>>>>     # echo N > /sys/module/kvm/parameters/use_gisa
>>>>
>>>> Already running guests are not affected by this change.
>>>>
>>>> The kvm s390 debug feature shows if a guest is running with GISA.
>>>>
>>>>     # grep gisa /sys/kernel/debug/s390dbf/kvm-$pid/sprintf
>>>>     00 01582725059:843303 3 - 08 00000000e119bc01  gisa 0x00000000c9ac2642 initialized
>>>>     00 01582725059:903840 3 - 11 000000004391ee22  00[0000000000000000-0000000000000000]: AIV gisa format-1 enabled for cpu 000
>>>>     ...
>>>>     00 01582725059:916847 3 - 08 0000000094fff572  gisa 0x00000000c9ac2642 cleared
>>>>
>>>> In general, that value should not be changed as the GISA facility
>>>> enhances interruption delivery performance.
>>>>
>>>> A reason to switch the GISA facility off might be a performance
>>>> comparison run or debugging.
>>>>
>>>> Signed-off-by: Michael Mueller <mimu@linux.ibm.com>
>>>> ---
>>>>    arch/s390/kvm/kvm-s390.c | 8 +++++++-
>>>>    1 file changed, 7 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
>>>> index d7ff30e45589..5c2081488024 100644
>>>> --- a/arch/s390/kvm/kvm-s390.c
>>>> +++ b/arch/s390/kvm/kvm-s390.c
>>>> @@ -184,6 +184,11 @@ static u8 halt_poll_max_steal = 10;
>>>>    module_param(halt_poll_max_steal, byte, 0644);
>>>>    MODULE_PARM_DESC(halt_poll_max_steal, "Maximum percentage of steal time to allow polling");
>>>>    
>>>> +/* if set to true, the GISA will be initialized and used if available */
>>>> +static bool use_gisa  = true;
>>>> +module_param(use_gisa, bool, 0644);
>>>> +MODULE_PARM_DESC(use_gisa, "Use the GISA if the host supports it.");
>>>> +
>>>>    /*
>>>>     * For now we handle at most 16 double words as this is what the s390 base
>>>>     * kernel handles and stores in the prefix page. If we ever need to go beyond
>>>> @@ -2504,7 +2509,8 @@ int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
>>>>    	kvm->arch.use_skf = sclp.has_skey;
>>>>    	spin_lock_init(&kvm->arch.start_stop_lock);
>>>>    	kvm_s390_vsie_init(kvm);
>>>> -	kvm_s390_gisa_init(kvm);
>>>> +	if (use_gisa)
>>>> +		kvm_s390_gisa_init(kvm);
>>>
>>> Looks sane to me. gi->origin will remain NULL and act like
>>> css_general_characteristics.aiv wouldn't be around.
>>
>> right
>>
>>>
>>> I assume initializing the gib is fine, and having some guests use it and
>>> others not?
>>
>> Is fine as well.
>>
>>>
>>> I do wonder if it would be even clearer/cleaner to not allow to change
>>> this property on the fly, and to also not init the gib if disabled.
>>>
>>> If you want to perform performance tests, simply unload+reload KVM.
>>
>> That would work if kvm is build as module but not in-kernel, then
> 
> Right, but AFAIK that's not an usual setup (at least in distros) - and
> for testing purposes not an issue as well.
> 
> Anyhow, I'm fine with this
> 
> Reviewed-by: David Hildenbrand <david@redhat.com>
> 

Thank you!

> Thanks!
>