Re: i386 Xen PV support still needed?
On Mon, 7 Aug 2017, Paolo Bonzini wrote: On 01/08/2017 23:38, Dominik 'Rathann' Mierzejewski wrote: On Tuesday, 01 August 2017 at 14:19, Florian Weimer wrote: We still build a special glibc variant for Xen which avoids certain segment-relative accesses which are difficult to emulate with paravirtualization.. Is this still needed? Can we drop it? What is the performance difference between running a regular glibc under Xen vs. this special one? I believe there may still be some value in running Fedora in a Xen i686 guest VM. The performance difference was very significant, though like others I cannot really give a figure. However, it only applied if you were running in a 32-bit hypervisor. I think this set up is pretty much dead. Even though Amazon and others are using Xen and are still running paravirtualized guests, they're using 64-bit hypervisors. CCing Vitaly for confirmation. The 32-bit (x86) hypervisor was dropped in xen-4.3.0 and as xen-4.4.x and earlier are end-of-life, the workaround presumably isn't needed now if it was just with the 32-bit hypervisor. Michael Young ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: i386 Xen PV support still needed?
On 01/08/2017 23:38, Dominik 'Rathann' Mierzejewski wrote: > On Tuesday, 01 August 2017 at 14:19, Florian Weimer wrote: >> We still build a special glibc variant for Xen which avoids certain >> segment-relative accesses which are difficult to emulate with >> paravirtualization.. >> >> Is this still needed? Can we drop it? > What is the performance difference between running a regular glibc under > Xen vs. this special one? I believe there may still be some value in > running Fedora in a Xen i686 guest VM. The performance difference was very significant, though like others I cannot really give a figure. However, it only applied if you were running in a 32-bit hypervisor. I think this set up is pretty much dead. Even though Amazon and others are using Xen and are still running paravirtualized guests, they're using 64-bit hypervisors. CCing Vitaly for confirmation. Paolo ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: i386 Xen PV support still needed?
On Wed, Aug 02, 2017 at 10:00:08PM +0200, Dominik 'Rathann' Mierzejewski wrote: > Thanks for the web archive links. Unfortunately, they don't quantify the > penalty, either. I'd feel more confident if I knew what it was (e.g. > 10x slower or 1.1x slower?) What I recall is it depended heavily on the binary and its use of TLS. I agree that we should just publicise the change and then drop the workaround. Rich. -- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones Read my programming and virtualization blog: http://rwmj.wordpress.com virt-top is 'top' for virtual machines. Tiny program with many powerful monitoring features, net stats, disk stats, logging, etc. http://people.redhat.com/~rjones/virt-top ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: i386 Xen PV support still needed?
On Wednesday, 02 August 2017 at 14:42, Richard W.M. Jones wrote: > On Tue, Aug 01, 2017 at 11:38:45PM +0200, Dominik 'Rathann' Mierzejewski > wrote: > > On Tuesday, 01 August 2017 at 14:19, Florian Weimer wrote: > > > We still build a special glibc variant for Xen which avoids certain > > > segment-relative accesses which are difficult to emulate with > > > paravirtualization.. > > > > > > Is this still needed? Can we drop it? > > This is a sort of historical feature. I had only a vague memory of > this from when I last used Xen PV & 32 bit guests (which must have > been around 2005?), but the precise problem and solution is explained > in these pages on the old XenSource wiki: > > https://web.archive.org/web/20080421123834/http://wiki.xensource.com:80/xenwiki/XenSegments > https://web.archive.org/web/20080218095938/wiki.xensource.com/xenwiki/XenSpecificGlibc > > Note the binaries for Fedora Core 3 :-) > > > What is the performance difference between running a regular glibc under > > Xen vs. this special one? I believe there may still be some value in > > running Fedora in a Xen i686 guest VM. > > I seem to recall that the performance penalty was very significant. Thanks for the web archive links. Unfortunately, they don't quantify the penalty, either. I'd feel more confident if I knew what it was (e.g. 10x slower or 1.1x slower?) > However, note that this is only for using Xen paravirtualization. > In no way would dropping this prevent you from using Xen or 32 bit > i686 guests, but you would have to use full virtualization. That > requires hardware virt, but (almost literally) every Intel and AMD > processor since 2006 has shipped with hardware virt. > > Also this doesn't apply to 64 bit guests at all (even PV) since they > didn't have to do fun things with segment registers to protect the > hypervisor from the guest. Right, so the use-case seems to be pretty limited. I'd say announce the intention to drop, publicize it widely and see what the feedback is. If almost nobody complains, then I'd say get rid of it if it makes life simpler. Regards, Dominik -- Fedora https://getfedora.org | RPMFusion http://rpmfusion.org There should be a science of discontent. People need hard times and oppression to develop psychic muscles. -- from "Collected Sayings of Muad'Dib" by the Princess Irulan ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: i386 Xen PV support still needed?
On Tue, Aug 01, 2017 at 11:38:45PM +0200, Dominik 'Rathann' Mierzejewski wrote: > On Tuesday, 01 August 2017 at 14:19, Florian Weimer wrote: > > We still build a special glibc variant for Xen which avoids certain > > segment-relative accesses which are difficult to emulate with > > paravirtualization.. > > > > Is this still needed? Can we drop it? This is a sort of historical feature. I had only a vague memory of this from when I last used Xen PV & 32 bit guests (which must have been around 2005?), but the precise problem and solution is explained in these pages on the old XenSource wiki: https://web.archive.org/web/20080421123834/http://wiki.xensource.com:80/xenwiki/XenSegments https://web.archive.org/web/20080218095938/wiki.xensource.com/xenwiki/XenSpecificGlibc Note the binaries for Fedora Core 3 :-) > What is the performance difference between running a regular glibc under > Xen vs. this special one? I believe there may still be some value in > running Fedora in a Xen i686 guest VM. I seem to recall that the performance penalty was very significant. However, note that this is only for using Xen paravirtualization. In no way would dropping this prevent you from using Xen or 32 bit i686 guests, but you would have to use full virtualization. That requires hardware virt, but (almost literally) every Intel and AMD processor since 2006 has shipped with hardware virt. Also this doesn't apply to 64 bit guests at all (even PV) since they didn't have to do fun things with segment registers to protect the hypervisor from the guest. Rich. -- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones Read my programming and virtualization blog: http://rwmj.wordpress.com Fedora Windows cross-compiler. Compile Windows programs, test, and build Windows installers. Over 100 libraries supported. http://fedoraproject.org/wiki/MinGW ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: i386 Xen PV support still needed?
On 1 August 2017 at 17:38, Dominik 'Rathann' Mierzejewski < domi...@greysector.net> wrote: > On Tuesday, 01 August 2017 at 14:19, Florian Weimer wrote: > > We still build a special glibc variant for Xen which avoids certain > > segment-relative accesses which are difficult to emulate with > > paravirtualization.. > > > > Is this still needed? Can we drop it? > > What is the performance difference between running a regular glibc under > Xen vs. this special one? I believe there may still be some value in > running Fedora in a Xen i686 guest VM. > > I would like to get an idea of who is running Fedora in an Xen i686 guest with this binary? We have a lot of "I believe" but very little data on actual usage. > Regards, > Dominik > -- > Fedora https://getfedora.org | RPMFusion http://rpmfusion.org > There should be a science of discontent. People need hard times and > oppression to develop psychic muscles. > -- from "Collected Sayings of Muad'Dib" by the Princess Irulan > ___ > devel mailing list -- devel@lists.fedoraproject.org > To unsubscribe send an email to devel-le...@lists.fedoraproject.org > -- Stephen J Smoogen. ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: i386 Xen PV support still needed?
On Tuesday, 01 August 2017 at 14:19, Florian Weimer wrote: > We still build a special glibc variant for Xen which avoids certain > segment-relative accesses which are difficult to emulate with > paravirtualization.. > > Is this still needed? Can we drop it? What is the performance difference between running a regular glibc under Xen vs. this special one? I believe there may still be some value in running Fedora in a Xen i686 guest VM. Regards, Dominik -- Fedora https://getfedora.org | RPMFusion http://rpmfusion.org There should be a science of discontent. People need hard times and oppression to develop psychic muscles. -- from "Collected Sayings of Muad'Dib" by the Princess Irulan ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org