Re: Kernelshark v2 all-black plots area

["Yordan Karadzhov (VMware)" <y.karadz@gmail.com>]

Hi Dario,

On 9.02.21 г. 18:13 ч., Dario Faggioli wrote:
> Indeed it does, thanks a lot again.
> 
> Now, if I try "KVM Combo Plot":
> - I still see super weird results if I use the KVM clock from last
>    patch in Tzvetomir series (but I guess I'll report it there).

Can you share a screenshot of the one that looks weird. And please do 
some deeper zooming because it is very hard to judge about the quality 
of the sync from such a broad time window.

> - If I use PTP, that kind of works.
> 
> If I trace a `sleep 1` on the host, the result is like this:
> 
> http://xenbits.xen.org/people/dariof/kernelshark-v2-ptp.png

Is this a PTP synchronization?

> 
> So, there is ~ 1 sec of data from the guest, coming from trace-cmd
> agent. The host, however, went on and traced a little bit longer, even
> after closing the connection. It's not too terrible, I'm just curious
> whether this is known or it's something weird and it's only me seeing
> it.

I think this is normal.

But anyway Ceco (Tzvetomir) is the right person to answer questions 
about the timestamp synchronization. Unfortunately he is on vacation 
till the end of the week.

Thanks a lot!
Yordan

> 
> Evaluating the correctness and the accuracy of the synchronization is a
> little bit of a challenge, especially for me that I'm not used to it.
> And, at least in part, because I still don't have this change in the
> kernel I'm using right now, so some kvm_entry trace points are not
> recorded, which means it's tricky pairing VMEntries and VMExits:
> 
> https://lore.kernel.org/lkml/160873470698.11652.13483635328769030605.stgit@Wayrath/
> 
> It should come with 5.11, but I think I'll quickly apply it to the 5.10
> that I'm running, to try to understand better hat's going on.