Description of problem: I have alot of 4gb fixup messages on console. (xm console) 4gb seg fixup, process updateuserdomai (pid 9490), cs:ip 73:001977c2 printk: 274779 messages suppressed. 4gb seg fixup, process httpd (pid 9479), cs:ip 73:001fb48e printk: 227828 messages suppressed. 4gb seg fixup, process pop3login (pid 9502), cs:ip 73:0027700f printk: 35278 messages suppressed. 4gb seg fixup, process check_perms (pid 9499), cs:ip 73:00da5ee4 printk: 28900 messages suppressed. 4gb seg fixup, process check_perms (pid 9499), cs:ip 73:00df312d printk: 7521 messages suppressed. 4gb seg fixup, process python2.4 (pid 8648), cs:ip 73:00d9c116 printk: 211596 messages suppressed. 4gb seg fixup, process syslogd (pid 26280), cs:ip 73:00b05116 printk: 473071 messages suppressed. 4gb seg fixup, process hackcheck (pid 9514), cs:ip 73:002037c2 printk: 520920 messages suppressed. 4gb seg fixup, process python2.4 (pid 8648), cs:ip 73:00d9c116 Version-Release number of selected component (if applicable): Linux vps.ispbrasil.com.br 2.6.18-1.2798.fc6xen #1 SMP Mon Oct 16 15:11:19 EDT 2006 i686 i686 i386 GNU/Linux How reproducible: xm console domU Expected results: Can you check for me if this patche disable this message and inclue it on next xen-kernel release ? http://xenbits.xensource.com/xen-unstable.hg?fd=1633f9a3a120;file=linux-2.6-xen-sparse/arch/i386/kernel/fixup.c another question: This messages can cause slow down on domU, if disable this messages my domU's will run more fast ? thanks for your help.
That patch will not help. The trouble is, mixing positive and negative segment offsets is slow under Xen. In practical terms, that means that applications linked statically against some older, pre-FC6 libraries are going to be slow. FC-6 uses a "nosegneg" library variant, autodetected at run time, to avoid this on Xen. But applications linked statically against an older libc will still use negative segment offsets, and will still take a kernel fault each time they try to do so. The log messages are simply telling you that these faults are occurring. The log is rate-limited to at most one per second, so its overhead should be minimal; but the segment fixups that it is reporting indicate a genuine slowdown that the user needs to be aware of. In your case, all these fixup messages are unexpected --- httpd, python, syslogd etc. should all be using nosegneg with FC-6. Are you perhaps using an FC-6 kernel on an older rawhide or Fedora?
yes, I am using a fc6 kernel to run fedora 4, 5, 6 and debian and other's domU's
the fc6 kernel seems to be running more slow than old version provided by xensource. xen-3.0.1-12.1_fc4.i386.rpm xen-kernel-2.6.12-xen3_12.1_fc4.i686.rpm and I have changed from "file://" to "tap:aio" in domU config files. any help to incresy performance of my domU's ?
The 4GB segmentation fixup message does not cause things to run slower. Instead, it signals when things are running slow, because userspace is using negative segment offsets. You will want to run with a nosegneg glibc variant to speed things up. Changing the kernel is not going to help at all.
Hi. I am using fc6-dom0 and fc6-domU and I am still receiving this message. this is normal ?
As Rik says in comment #4, it is not the kernel which matters here. You need to use a newer glibc which comes with a 'nosegneg' variant to fix the problem. # rpm -q glibc glibc-2.5-3 # rpm -ql glibc | grep nosegneg /lib/i686/nosegneg /lib/i686/nosegneg/libc-2.5.so /lib/i686/nosegneg/libc.so.6 ....
to use nosegneg glibc echo "hwcap 0 nosegneg" > /etc/ld.so.conf.d/nosegneg.conf this will stop messages on console
The hwcap should not be necessary: the kernel-xen package already includes an appropriate ld.so.conf file: # cat /etc/ld.so.conf.d/kernelcap-2.6.18-1.2798.fc6.conf ... hwcap 0 nosegneg If there are specific binaries still inducing this error, then please do report those.