Re: Ryzen 4000 (Mobile) Softlocks/Micro-stutters

[Dylanger Daly <dylangerdaly@xxxxxxxxxxxxxx>]

Hi Jan, thank you for responding.

Indeed this is for dom0, I only recently tried limiting a domU to 1 core and 
observed absolutely no softlocks, UI animations are smooth as butter with 1 
core only.

Indeed I believe this is a CPU Scheduling issue, I've tried both the older 
credit and RTDS however both don't boot correctly.
The number of cores on this CPU is 8, 16 threads however Qubes by default 
disables SMT, sched_credit2_max_cpus_runqueue is 16 by default, I've tried 
testing with setting this to 7 or 8 however it'll either not boot, or nothing 
will change.

There are a number of credit2 tweak-ables so I'm hoping to play around and drop 
the `dom0_max_vcpus=1`, I suspect `sched_credit2_max_cpus_runqueue` is the main 
thing to play with.

I did manage to get it booting with sched_credit2_max_cpus_runqueue=7 but it 
ended up locking up shortly after X launched on dom0

‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐

On Thursday, October 15th, 2020 at 7:18 PM, Jan Beulich <jbeulich@xxxxxxxx> 
wrote:

> On 15.10.2020 02:38, Dylanger Daly wrote:
>
> > I'm currently using Xen 4.14 (Qubes 4.1 OS) on a Ryzen 7 4750U PRO, by 
> > default I'll experience softlocks where the mouse for example will jolt 
> > from time to time, in this state it's not usable.
>
> From what you say below I imply this is in Dom0?
>
> > Adding `dom0_max_vcpus=1 dom0_vcpus_pin` to Xen's CMDLINE results in no 
> > more jolting however performance isn't what it should be on an 8 core CPU, 
> > softlocks are still a problem within domU's, any sort of UI animation for 
> > example.
> >
> > Reverting this commit (8e2aa76dc1670e82eaa15683353853bc66bf54fc) results in 
> > even worse performance with or without the above changes to CMDLINE, and 
> > it's not usable at all.
>
> You saying this surely has a reason, but making the connection would
>
> help. I don't consider it surprising that a revert of an improvement
>
> makes things worse. You having bothered to find a certain code change
>
> also makes me suspect you've experimented with other scheduler
>
> related settings - if so, please share all data you've got. (FAOD -
>
> with the information provided I have no idea what to suggest, sorry.)
>
> Jan