Previous INDEX Next
Improved Web Publishing Google - Where did my pages go?

Yet Another FreeBSD 4.x to 5.x upgrade

Following the upgrade of the hydrus web/mail server to FreeBSD 5.4, the backup server, chrome, also needed upgrading. This operation was made slightly more complicated by the fact that extremely important users (i.e. my family) relied on chrome as a file and print server.

Since /home was used as a backup for a couple of the other machines on the network, I wanted to preserve this partition across the upgrade. I therefore re-used the existing partition table, but ensured that newfs was invoked for all partitions except /home.

One slightly worrying aspect of the disk setup was that sysinstall told me that the geometry of 119150/16/63 was wrong, and it was using 7476/255/63. The former geometry was exactly the same as that determined by 4.11. I tried using the G option to set the geometry back, but sysinstall always reverted to its preferred figures. I pressed on regardless.

The upgrade continued without problem, and I was able to re-boot into FreeBSD 5.4, with my original /home partition intact. However, I did notice the following message on boot (if the printer was turned on):

  Interrupt storm detected on "irq7: lpt0"; throttling interrupt source

Hmm, that never happened in 4.11. I ignored it for now, but I was forced to deal with it sooner rather than later.

I pkg_add'd cvsup-without-gui, and updated the src and ports trees. I then added the following packages (directly or indirectly):

apache                      
autoconf                    
cmucl                       
compat4x-i386               
cvsup-without-gui           
emacs                       
expat                       
ezm3                        
gettext                     
gmake                       
libiconv                    
libtool                     
libxml2                     
m4                          
perl                        
pkgconfig                   
popt                        
portupgrade                 
procmail                    
py24-libxml2                
python                      
rsync                       
ruby                        
ruby18-bdb1                 
samba                       

To minimise the number of applications on chrome, I chose to use BSD printing for Samba, rather than cups as before. By default, Samba requires cups-base, so I set the following flags in /etc/make.conf

WITHOUT_CUPS=true
WITHOUT_LDAP=true

On testing Samba, I found that printing on the HP LaserJet5N attached to chrome was weird. The printer's message window showed "Processing job", then "Ready", then "Receiving Data", then ready, until, after an inordinate length of time, the document actually printed. Initially, I wasn't sure if this was caused by the BSD printing, or related to the interrupt storm message, or some odd 5.4 problem. First, I rechecked the /etc/printcap file; was I sending a banner or something by mistake?

HPLaserJet5N|HP LaserJet5N:\
        :sh:sf:mx#0:\
        :lp=/dev/lpt0:sd=/var/spool/output/HP:lf=/var/log/lpd-errs:

No, that looked OK - suppressing header and trailing formfeed. I then added a filter, which tee'd the print data to a file. I then realised the printing was very slow for large files; the interrupt storm must be to blame.

Armed with this information, a quick web search turned up the solution, from both these now defunct links: http://www.xs4all.nl/~rsmith/freebsd/ and http://www.n0dez.com/notes.html.

Add the following to /boot/device.hints:

  hint.ppc.0.flags="0x28"

This forces the parallel port to operate in polled rather than interrupt mode. You can achieve the same effect with lptcontrol -p. Printing speed was now normal.

Previous INDEX Next
Improved Web Publishing Google - Where did my pages go?