Hi Manuel, Am 03.02.2021 um 13:12 schrieb Manuel Bouyer:
On Wed, Feb 03, 2021 at 01:06:56PM +0100, Matthias Petermann wrote:Hello Manuel, Am 02.02.2021 um 19:39 schrieb Manuel Bouyer:Hello, it seems that this is due to a difference between linux and NetBSD poll implementation. This causes sockets to never be closed on the xenstored side, and instead loop ignoring the POLLIN event. The attached patch fixes it for me, but I'm not sure at this point if it's 100% correct.Thanks so much - this sounds good and I'm going to try this out right after work today!Things are still in discussion with Xen - the new version of the patch is attached. But it's possible that the first chunk will be dropped, keeping only the patch to ignore_connection() (this is the one that really matters).
Thank you for the update. It is wonderful that research is being done here. A robust Xen is a real differentiator from NetBSD and the upcoming changes (Dom0 SMP etc) sound very promising as well.
I have one more possibly relevant observation about the xenstored problem. I have another system with Xen at the following version level:
``` # uname -aNetBSD jupiter.local 9.1 NetBSD 9.1 (XEN3_DOM0) #0: Sun Oct 18 19:24:30 UTC 2020 mkrepro%mkrepro.NetBSD.org@localhost:/usr/src/sys/arch/xen/compile/XEN3_DOM0 amd64
# pkg_info|grep xen xenkernel413-4.13.1nb1 Xen 4.13.x Kernel xentools413-4.13.1nb3 Userland Tools for Xen 4.13.x ```Here the problem does not occur. However, the hardware is also very different. The problematic system is an Intel NUC5 with intel Celeron N3050, 2 cores. The system without symptoms is an AMD FX-8350 with 8 cores and lots of RAM.
Kind regards Matthias