[Question] About scaling factor of Enhanced Counter Virtualization

[Question] About scaling factor of Enhanced Counter Virtualization

[zhukeqian]

Hi all,

We are studying Enhanced Counter Virtualization (introduced by ARMv8.6 extension), and here is a question
raised by Biaoxiang Ye <yebiaoxiang@huawei.com>:

Described in the ARMv8.6 Extension Specification:
Note: the scaling factor CNTSCALE is designed as a 2.62 bit fixed point number, so permitting a scaling up by
(nearly) a factor 4. The scaling factor CNTISCALE is signed as an 8.56 number for the scaling of the values
written into the timers for comparison with the actual count. This implies that the greatest scaling down of the
counter supported in (nearly) a factor of *512*.

We think the number "512" should be "256" (2^8), or do we miss something?

Thanks,
Keqian
_______________________________________________
kvmarm mailing list
kvmarm@lists.cs.columbia.edu
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm

Re: [Question] About scaling factor of Enhanced Counter Virtualization

[Marc Zyngier]

Hi Keqian,

On 2020-07-14 03:20, zhukeqian wrote:
> Hi all,
> 
> We are studying Enhanced Counter Virtualization (introduced by ARMv8.6
> extension), and here is a question
> raised by Biaoxiang Ye <yebiaoxiang@huawei.com>:
> 
> Described in the ARMv8.6 Extension Specification:
> Note: the scaling factor CNTSCALE is designed as a 2.62 bit fixed
> point number, so permitting a scaling up by
> (nearly) a factor 4. The scaling factor CNTISCALE is signed as an 8.56
> number for the scaling of the values
> written into the timers for comparison with the actual count. This
> implies that the greatest scaling down of the
> counter supported in (nearly) a factor of *512*.
> 
> We think the number "512" should be "256" (2^8), or do we miss 
> something?

This register doesn't seem to be described in ARM DDI 0487F.b,
which is the official documentation and does contain the ARMv8.6
material, including ARMv8.6-ECV.

Either you are looking at confidential information (and nobody
can answer you in public), or obsolete information (and nobody
knows what this is about).

In any case, this is most probably a question for ARM's support
team, and not the kernel mailing lists.

Thanks,

         M.
-- 
Jazz is not dead. It just smells funny...
_______________________________________________
kvmarm mailing list
kvmarm@lists.cs.columbia.edu
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm

Re: [Question] About scaling factor of Enhanced Counter Virtualization

[zhukeqian]

Hi Marc,

On 2020/7/14 19:34, Marc Zyngier wrote:
> Hi Keqian,
> 
> On 2020-07-14 03:20, zhukeqian wrote:
>> Hi all,
>>
>> We are studying Enhanced Counter Virtualization (introduced by ARMv8.6
>> extension), and here is a question
>> raised by Biaoxiang Ye <yebiaoxiang@huawei.com>:
>>
>> Described in the ARMv8.6 Extension Specification:
>> Note: the scaling factor CNTSCALE is designed as a 2.62 bit fixed
>> point number, so permitting a scaling up by
>> (nearly) a factor 4. The scaling factor CNTISCALE is signed as an 8.56
>> number for the scaling of the values
>> written into the timers for comparison with the actual count. This
>> implies that the greatest scaling down of the
>> counter supported in (nearly) a factor of *512*.
>>
>> We think the number "512" should be "256" (2^8), or do we miss something?
> 
> This register doesn't seem to be described in ARM DDI 0487F.b,
> which is the official documentation and does contain the ARMv8.6
> material, including ARMv8.6-ECV.
> 
> Either you are looking at confidential information (and nobody
> can answer you in public), or obsolete information (and nobody
> knows what this is about).
> 
Well, it's alpha release version :( . Many thanks!

Thanks,
Keqian
> In any case, this is most probably a question for ARM's support
> team, and not the kernel mailing lists.
> 
> Thanks,
> 
>         M.
_______________________________________________
kvmarm mailing list
kvmarm@lists.cs.columbia.edu
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm

Re: [Question] About scaling factor of Enhanced Counter Virtualization

[Marc Zyngier]

On 2020-07-14 13:15, zhukeqian wrote:
> Hi Marc,
> 
> On 2020/7/14 19:34, Marc Zyngier wrote:
>> Hi Keqian,
>> 
>> On 2020-07-14 03:20, zhukeqian wrote:
>>> Hi all,
>>> 
>>> We are studying Enhanced Counter Virtualization (introduced by 
>>> ARMv8.6
>>> extension), and here is a question
>>> raised by Biaoxiang Ye <yebiaoxiang@huawei.com>:
>>> 
>>> Described in the ARMv8.6 Extension Specification:
>>> Note: the scaling factor CNTSCALE is designed as a 2.62 bit fixed
>>> point number, so permitting a scaling up by
>>> (nearly) a factor 4. The scaling factor CNTISCALE is signed as an 
>>> 8.56
>>> number for the scaling of the values
>>> written into the timers for comparison with the actual count. This
>>> implies that the greatest scaling down of the
>>> counter supported in (nearly) a factor of *512*.
>>> 
>>> We think the number "512" should be "256" (2^8), or do we miss 
>>> something?
>> 
>> This register doesn't seem to be described in ARM DDI 0487F.b,
>> which is the official documentation and does contain the ARMv8.6
>> material, including ARMv8.6-ECV.
>> 
>> Either you are looking at confidential information (and nobody
>> can answer you in public), or obsolete information (and nobody
>> knows what this is about).
>> 
> Well, it's alpha release version :( . Many thanks!

Wow. I suggest you discard this document, as it is obsolete.
ARMv8.6 is a released version of the architecture, and
everything should be in the ARM ARM.

         M.
-- 
Jazz is not dead. It just smells funny...
_______________________________________________
kvmarm mailing list
kvmarm@lists.cs.columbia.edu
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm

Re: [Question] About scaling factor of Enhanced Counter Virtualization

[zhukeqian]

On 2020/7/14 20:23, Marc Zyngier wrote:
> On 2020-07-14 13:15, zhukeqian wrote:
>> Hi Marc,
>>
>> On 2020/7/14 19:34, Marc Zyngier wrote:
>>> Hi Keqian,
>>>
>>> On 2020-07-14 03:20, zhukeqian wrote:
>>>> Hi all,
>>>>
>>>> We are studying Enhanced Counter Virtualization (introduced by ARMv8.6
>>>> extension), and here is a question
>>>> raised by Biaoxiang Ye <yebiaoxiang@huawei.com>:
>>>>
>>>> Described in the ARMv8.6 Extension Specification:
>>>> Note: the scaling factor CNTSCALE is designed as a 2.62 bit fixed
>>>> point number, so permitting a scaling up by
>>>> (nearly) a factor 4. The scaling factor CNTISCALE is signed as an 8.56
>>>> number for the scaling of the values
>>>> written into the timers for comparison with the actual count. This
>>>> implies that the greatest scaling down of the
>>>> counter supported in (nearly) a factor of *512*.
>>>>
>>>> We think the number "512" should be "256" (2^8), or do we miss something?
>>>
>>> This register doesn't seem to be described in ARM DDI 0487F.b,
>>> which is the official documentation and does contain the ARMv8.6
>>> material, including ARMv8.6-ECV.
>>>
>>> Either you are looking at confidential information (and nobody
>>> can answer you in public), or obsolete information (and nobody
>>> knows what this is about).
>>>
>> Well, it's alpha release version :( . Many thanks!
> 
> Wow. I suggest you discard this document, as it is obsolete.
> ARMv8.6 is a released version of the architecture, and
> everything should be in the ARM ARM.
OK, there are many difference :) .

Thanks,
Keqian
> 
>         M.
_______________________________________________
kvmarm mailing list
kvmarm@lists.cs.columbia.edu
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm