26.07.2015 19:14, kp kirchdoerfer пишет: > Am Sonntag, 26. Juli 2015, 18:16:51 schrieb Andrew: >> 26.07.2015 18:09, kp kirchdoerfer пишет: >>> Hi Andrew; >>> >>> Am Samstag, 25. Juli 2015, 19:32:10 schrieb Andrew: >>>> Hi. >>>> >>>> I'll try to add it at this weekend. >>> I've updated config.lrp and initrd.lrp and it seems to work regarding >>> saving modules. >>> But it seems there is a problem when removing modules.sqfs. >>> While I do have saved the modules to moddb.lrp, it seems modules saved in >>> moddb.lrp are not loaded (modules.sqfs has been deleted/removed). >> Strange, fallback code should work OK. I'll test it > Please do. > >>> In addition we should make shure for updates that modules.sqfs overwrites >>> moddb.lrp - "if modules.sqfs is available, do not load moddb.lrp" >>> >>> kp >> I'm not sure that it's a good idea. At least - moddb may be leaved for >> user drivers/fresh versions of drivers. > My wording was misleading ("overwritten"). > >> Currently we: >> 0) probing drivers from initmod >> 1) trying to probe drivers from moddb >> 2) trying to probe drivers from sqfs and copy missed drivers to /lib/modules > That's what I meant - good it's in place aleady. > > kp Currently we didn't check if driver was loaded from moddb or from sqfs. So if there's already available drivers in /lib/modules (for different kernel) - they will not be overwritten.
> >>>> 25.07.2015 19:15, kp kirchdoerfer пишет: >>>>> Hi Andrew; >>>>> >>>>> Am Samstag, 4. Juli 2015, 16:00:24 schrieb Andrew: >>>>>> 04.07.2015 15:45, kp kirchdoerfer пишет: >>>>>>> Am Samstag, 4. Juli 2015, 15:31:40 schrieb Andrew: >>>>>>>> 04.07.2015 15:14, kp kirchdoerfer пишет: >>>>>>>>> Hi; >>>>>>>>> >>>>>>>>> I almost agree with Erich :) >>>>>>>>> >>>>>>>>> Am Samstag, 4. Juli 2015, 14:56:33 schrieb Andrew: >>>>>>>>>> 04.07.2015 11:34, Erich Titl пишет: >>>>>>>>>>> Hi >>>>>>>>>>> >>>>>>>>>>> Am 04.07.2015 um 09:50 schrieb Andrew: >>>>>>>>>>>> Hi. >>>>>>>>>>>> >>>>>>>>>>>> In most cases probing from squashfs is preferred over loading >>>>>>>>>>>> from >>>>>>>>>>>> moddb >>>>>>>>>>>> - at least, this will not break system after update if it'll >>>>>>>>>>>> require >>>>>>>>>>>> some modules that aren't in moddb (for ex., some ethernet >>>>>>>>>>>> driver). >>>>>>>>>>>> >>>>>>>>>>>> We may add some option to use legacy behavior with sqfs, or we >>>>>>>>>>>> may >>>>>>>>>>>> add >>>>>>>>>>>> fallback behavior with tgz package for hwdetect. >>>>>>>>>>> This would IMHO be a very ugly behaviour. I belived and still >>>>>>>>>>> believe >>>>>>>>>>> that the sqfs was a more elegant replacement for modules.tgz, and >>>>>>>>>>> it >>>>>>>>>>> is. You decided to use it also on startup with hwdetect, which >>>>>>>>>>> makes >>>>>>>>>>> IMHO most of the modules in initmod completely redundant. >>>>>>>>>> This makes moddb completely redundant, and this is leaved only for >>>>>>>>>> systems with small storage + rare cases like custom modules. >>>>>>>>>> Initmod >>>>>>>>>> is >>>>>>>>>> needed anycase - it's mounted with rootfs, and requires loaded >>>>>>>>>> storage-related modules. >>>>>>>>> Indeed; if we empty moddb, we have to make shure most of the modules >>>>>>>>> are >>>>>>>>> referenced in /etc/modules, so they'll be loaded by hwdtect during >>>>>>>>> boot >>>>>>>>> from sqfs. >>>>>>>> No, they'll be loaded like with filled moddb. >>>>>>>> >>>>>>>>> Will free about 1.8MB on the images, but I believe requires >>>>>>>>> more >>>>>>>>> >>>>>>>>> maintenance cvause we need to keep /etc/modules in sync with kernel >>>>>>>>> modules >>>>>>>>> (netfilter etc). Don't know if wildcards are suppported in >>>>>>>>> /etc/modules. >>>>>>>> netfilter/iptables/shaper modules are in .depend list of >>>>>>>> corresponding >>>>>>>> packages (<DependsOn> section, lines like 'Module = ipt_*' - yes, >>>>>>>> with >>>>>>>> wildcards; algorithm earlier was used in getdep.sh). >>>>>>> Seems I saw it, but forgot the changes you've made. So we can start to >>>>>>> work >>>>>>> with an empty moddb, but keep it for saving modules after hwdetect or >>>>>>> added by the user? >>>>>> Right. On my 5.2 boxes there's no moddb at all. >>>>>> >>>>>>>>>>> To me all this looks like we have no clear policy for module >>>>>>>>>>> handling, >>>>>>>>>>> at least in a transition phase. Just using the sqfs looks like a >>>>>>>>>>> regression to me. >>>>>>>>>>> >>>>>>>>>>> Reducing the size or even making initmod redundant would be great >>>>>>>>>>> if >>>>>>>>>>> we could achieve that. I believe with the inclusion of sqfs >>>>>>>>>>> support >>>>>>>>>>> in >>>>>>>>>>> the kernel this is possible. We should only do hwdetect if sqfs is >>>>>>>>>>> present and definitely not inhibit saving modules in moddb. >>>>>>>>>>> >>>>>>>>>>> I don't want to run hwdetect at each and every boot and there >>>>>>>>>>> should >>>>>>>>>>> be a boot time selection either just by the presence of sqfs or a >>>>>>>>>>> command line switch. >>>>>>>>>> Will it be enough good to add legacy .tgz handling to hwdetect (so >>>>>>>>>> systems with tgz instead of sqfs will have same behavior as >>>>>>>>>> earlier)? >>>>>>>>> I don't think it's necessary to revive tgz; it will just waste >>>>>>>>> disk-space >>>>>>>>> adds code etc.. >>>>>>>>> >>>>>>>>> Saving modules in modbd and get rid of modules.sqfs later and until >>>>>>>>> next >>>>>>>>> update works if I remoce /var/lib/lrpkg/sqfs.modules (the >>>>>>>>> "blacklist"). >>>>>>>>> >>>>>>>>> An idea could be to add configuration option in /etc/lrp.conf (like >>>>>>>>> lrp_SAVE_MODULES) and if set to "yes" hwdetect does not write the >>>>>>>>> modules >>>>>>>>> loaded into the blacklist - waht dou you think? >>>>>>>>> >>>>>>>>> kp >>>>>>>> Ok, we may do this. This shouldn't be too complex. Where this option >>>>>>>> should be (leaf.cfg, kernel boot line, /etc/lrp.conf or somewhere >>>>>>>> else?). >>>>>>> I think leaf.cfg would be the most obvious place. >>>>>>> >>>>>>> kp >>>>>> Ok, I'll try to add option to init/hwdetect util in some days. I think >>>>>> that by default we shouldn't save probed modules to moddb. >>>>> What about this one? >>>>> Do you have time to add the code soon? >>>>> >>>>> I'd like to have it before we are doing another rc release for 5.2. >>>>> >>>>> kp > > ------------------------------------------------------------------------------ > > _______________________________________________ > leaf-devel mailing list > leaf-devel@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/leaf-devel ------------------------------------------------------------------------------ _______________________________________________ leaf-devel mailing list leaf-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/leaf-devel