Re: [Xen-devel] Question about patch "svm: fix incorrect TSC scaling"

[Dongli Zhang <dongli.zhang@xxxxxxxxxx>]

On 05/29/2018 06:23 PM, Dongli Zhang wrote:
> 
> 
> On 05/29/2018 05:56 PM, Jan Beulich wrote:
>>>>> On 28.05.18 at 14:25, <dongli.zhang@xxxxxxxxxx> wrote:
>>> I am writing to ask about below patch as I am not familiar with AMD tsc
>>> scaling:
>>>
>>>  commit 11eeca65126e51f03a883907751d5ccbe4f35aa3
>>>  Author: Haozhong Zhang <haozhong.zhang@xxxxxxxxx>
>>>  Date:   Tue Dec 8 09:46:12 2015 +0100
>>>
>>>      svm: fix incorrect TSC scaling
>>>  
>>>      SVM TSC ratio is incorrectly used in the current
>>>      svm_get_tsc_offset(). This patch replaces the scaling logic in
>>>      svm_get_tsc_offset() with a correct implementation.
>>>  
>>>      Signed-off-by: Haozhong Zhang <haozhong.zhang@xxxxxxxxx>
>>>      Reviewed-by: Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx>
>>>
>>>
>>> The below equation is used in the above patch:
>>>
>>> -------------------------
>>> host_tsc * ratio * 2^-32.
>>> -------------------------

Indeed, I have an extra question that why 2^-32 is involved here?

Is it because we would like to ignore the frag part and only consider the
leading integer part? Why not just use "host_tsc * ratio"?

Thank you very much!

Dongli Zhang


>>>
>>>
>>> While the below incorrect equation (indeed code) was used before above 
>>> patch:
>>>
>>> ---------------------------------------------------------------
>>> offset = (((host_tsc >> 32U) * (ratio >> 32U)) << 32U) +
>>>          (host_tsc & 0xffffffffULL) * (ratio & 0xffffffffULL);
>>> return guest_tsc - offset;
>>> ---------------------------------------------------------------
>>>
>>> Can I summarize the above code as:
>>>
>>> tsc_h * mult * 2^32 + tsc_l * frag or
>>>
>>> tsc_h * int * 2^32 + tsc_l * frag ?
>>>
>>>
>>> Would you please help confirm why the equation is replaced? Or would you 
>>> please
>>> help and let me know the chapter/page of AMD reference manual showing the
>>> correct equation?
>>
>> This has nothing to do with AMD's doc, but only with maths: To split
>> a multiplication x * y when considering x = xh + xl (high and low parts)
>> and y = yh + yl you get x * y = xh * yh + xh * yl + xl * yh + xl * xl.
>> Width constraints make it unnecessary to calculate the xh * yh part
>> here. The original expression, however, calculated something rather
>> strange instead (only the xl * yl part was actually correct).
> 
> Thank you very much for the explanation!
> 
> Now it is clear to me.
> 
> Dongli Zhang
> 

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel