Am Montag, den 28.01.2019, 15:44 -0800 schrieb Sean Christopherson: > On Mon, Jan 28, 2019 at 10:46:41PM +0100, Jens Sauer wrote: > > Hey there, > > > > I am using a 4.14.y kernel on a host system based on Debian 9. It > > is a > > custom kernel configuration which runs since February 2018 on the > > 4.14.y series. This machine is used as qemu/kvm host server. > > > > Last week I upgraded the kernel from 4.14.76 to 4.14.96 and > > suddenly > > one of the guest runs into segmentation fault errors. > > It is a x86_64 guest running OPNsense 18.7.10_3 which is based on > > FreeBSD 11.1-RELEASE-p18. > > > > The error appears directly after booting the guest. Sometimes there > > are > > a "segmentation fault" errors in the boot log or checksum > > verification > > fails during boot. The guest appears to be slow, the web UI does > > not > > response in 90 % of the time. > > TLS connections are failing from the guest to any remote hosts, the > > errors were always caused by failing signature verification of the > > remote host. > > > > After a few minutes the FreeBSD kernel logs: "HBSD SEGVGUARD > > suspension > > expired python2.7" or "php-cgi". > > > > At no time I could see any errors on the host. > > > > I am not sure if this problem does appear in any of my linux based > > guest. I shut down the linux guests as soon as I noticed the errors > > in > > the FreeBSD guest. > > I was afraid of a bad memory stick, there were no CE/UE reported by > > the > > MC on the host. I ran memtest twice, which resulted in no error. > > > > I then made a bisect between tag v4.14.76 and v4.14.96 which > > identified > > commit 4124a4cff344abbf8187775eb643d9827830e715 > > as the first bad commit. > > > > Please let me know if you need more information or how I can help > > you > > to track down the issue. I hope this report is sufficient, it is my > > first bug report for the kernel. > > Does the attached patch resolve your issues? I'm fairly certain it's > correct, but AFAIK none of the original reporters has confirmed the > fix. Yes, it works as you expected. I can not longer reproduce the behavior described above. Thanks for your work. -- Jens Sauer <jsauer65@xxxxxxxxx>
