Am 27.01.22 um 23:18 schrieb Scott Bradnick:
On Thu, 2022-01-27 at 21:17 +0100, Aaron Puchert wrote:
Nice! I've done this manually so far, so I had a look at the setup:
> install -m 0644 hv_sock.conf "%{buildroot}/%{_modulesloaddir}/hv_sock.conf"
Somehow I didn't need that, but Leap 15.3 doesn't ship with the module out of the box, not even with kernel-azure. I had to install the kernel-*-extra package. For the specfile you might just add
Requires: kmod(hv_sock.ko)
to ensure that the module is there. Then in xrdp.ini:
> bulk_compression=true
Likely not needed across virtual sockets, so I turned it off. (Don't know if it makes a difference though.)
Got some responsiveness improvement by making both xrdp and Xvnc real-time processes, but I guess that might be seen as risky in general. (The former via CPUSchedulingPolicy= in a drop-in configuration, the latter by a CAP_SYS_NICE-enabled wrapper that called sched_setscheduler and then exec'ed into Xvnc. Nothing to be proud of.)
Best regards, Aaron
I'm not going to pretend I follow how the "hyper-v" package works from top to bottom, but all my "hv_sock.conf" file is doing is telling the system to load that module ~ where it comes from and why seems to be handled elsewhere. I'm guessing some intelligence baked into "hyper-v" package which sets up a few things behind the scenes.
Had another look at my VM and it turns out there is actually something in /etc/modules-load.d/, but I completely forgot about that. Sorry for the noise. As for the "Requires": it seems unneeded in Tumbleweed, but if you want to bring that package to Leap, where it's in the kernel-*-extra package which is unlikely to be installed, you should have it. And it doesn't hurt on Tumbleweed either, because the dependency is resolved by the regular kernel package. Best regards, Aaron