[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20230901122431.GU11676@atomide.com>
Date: Fri, 1 Sep 2023 15:24:31 +0300
From: Tony Lindgren <tony@...mide.com>
To: Sean Christopherson <seanjc@...gle.com>
Cc: Marc Haber <mh+linux-kernel@...schlus.de>,
Bagas Sanjaya <bagasdotme@...il.com>,
linux-kernel@...r.kernel.org,
Linux Regressions <regressions@...ts.linux.dev>,
Linux KVM <kvm@...r.kernel.org>,
Paolo Bonzini <pbonzini@...hat.com>
Subject: Re: Linux 6.5 speed regression, boot VERY slow with anything systemd
related
* Sean Christopherson <seanjc@...gle.com> [230831 22:07]:
> On Tue, Aug 29, 2023, Marc Haber wrote:
> > Hi,
> >
> > On Tue, Aug 29, 2023 at 10:14:23AM -0700, Sean Christopherson wrote:
> > > On Tue, Aug 29, 2023, Marc Haber wrote:
> > > > Both come from virt-manager, so if the XML helps more, I'll happy to
> > > > post that as well.
> > >
> > > Those command lines are quite different, e.g. the Intel one has two
> > > serial ports versus one for the AMD VM.
> >
> > Indeed? I virt-manager, I don't see a second serial port. In either case,
> > only the one showing up in the VM as ttyS0 is being used. But thanks for
> > making me look, I discovered that the machine on the Intel host still
> > used emulated SCSI instead of VirtIO für the main disk. I changed that.
> >
> > >Unless Tony jumps in with an
> > > idea, I would try massaging either the good or bad VM's QEMU
> > > invocation, e.g. see if you can get the AMD VM to "pass" by pulling in
> > > stuff from the Intel VM, or get the Intel VM to fail by making its
> > > command line look more like the AMD VM.
> >
> > In Virt-Manager, both machines don't look THAT different tbh. I verified
> > the XML and the differences are not big at all.
> >
> > Do you want me to try different vCPU types? Currently the VM is set to
> > "Opteron_G3", would you recommend a different vCPU for the host having a
> > "AMD GX-412TC SOC" host CPU?
>
> I would be surprised if using a different vCPU type fixed anything, but it's not
> impossible that it could help. In general, unless someone from the serial driver
> side spots an issue, fixing whatever the bug is will likely require a reproducer,
> which in turn likely means narrowing down what exactly is unique about your AMD
> setup. In other words, if you have cycles to spare, anything you can do to help
> isolate the problem would be appreciated.
Yes two somewhat minimal qemu command lines for working and failing test
case sure would help to debug this.
Regards,
Tony
Powered by blists - more mailing lists