Christian, On Thu, Nov 10, 2005 at 12:27:58PM +0100, Christian Haugan Toldnes wrote: > Alain Fauconnet wrote: > >Compiling the MPT driver statically works but raises other problems > >because this machine has 2 SCSI controllers and there are detection > >order issues. >
Well... searching *harder* into Google produced an interesting hint I need to explore when I can take this box down. https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=FC4_FUSION Seems that the MPT fusion driver has been split into more than 2 modules in 2.6.13 and that I need to load the low-layer module (mptspi) first. I missed that (not that it's written anywhere in the kernel source doc...) But in the meantime: > I'm not trying to answer your real question, but might solve your problem :) > > With two (different) controllers and detection ordering issues, wouldn't > it be easy enough to compile the drivers for the _other_ controller as > modules, and include the mpt-stuff into the kernel. That way you do know > the order of detection, and the other drivers probably does not have the > same issues with unresolved symbols... :) Thanks :-) What makes the whole picture complex is that this box has a MPT Fusion controller and an IBM ServeRAID 4Lx (the infamous ips module that causes a loop during initrd processing ... see a recent posting from me: http://lists.trustix.org/pipermail/tsl-discuss/2005-October/015843.html Different machine, same symptom as for the ips. Anyway: sda (root & al.) is on the MPT, sdb (various /var subdirs and a 2nd swap) is on the ServeRAID. So I need to load the MPT drivers first. - if all is static, the kernel detects the ServeRAID first => boom - if I link mpt static and ips dynamic (but not loaded from initrd because that creates some kind of endless loop), it boots OK until it machine tries to mount stuff from sdb: this fails because it won't load a second SCSI driver module dynamically if there's already one statically in (or am I missing something?). So sdb isn't there. - OK, so I add 'ips' to /etc/modules... almost there, because 'swapon -a' is ran from /etc/rc.sysint *before* /etc/modules is processed. So I still get an error at that point (I have a swap on sdb too... yes I know I'm a pain :-) What's left? Addings insmod=ips to the LILO kernel parameters and hoping that the dreaded initrd processing stop doesn't happen when it's loaded this way? Greets, _Alain_ _______________________________________________ tsl-discuss mailing list [email protected] http://lists.trustix.org/mailman/listinfo/tsl-discuss
