Thank you for your help, Thibaut. I think we've found the right way of initializing routerboot. I've sent a v2 of the patch.
Denis вт, 16 нояб. 2021 г. в 13:20, Thibaut <ha...@slashdirt.org>: > > Hi, > > > Le 16 nov. 2021 à 10:46, Denis K <denis281...@gmail.com> a écrit : > > > > I got it! _exit macro of rb_hardconfig_init seems to cause this kernel > > panic. Here is a patch. > > Thanks; indeed that’s logical: functions marked as __exit are discarded when > the module is built into the kernel: they’re no longer available when the > mtd_notifier executes the remove() callback. I missed that. Since these are > very small, I agree it’s ok to remove the __exit qualifier. > > I see the same rationale for removing the __init qualifier from the init > functions since the mtd notifier could be called at any point during kernel > execution, but that bothers me a bit more because these are somewhat large > functions. Then again it’s a tradeoff between doing the right thing (using a > notifier) vs using a workaround (late_initcall()). I think it’s acceptable. > > Anyway, I’ve refactored the patch with your changes and I added a note and > edited the commit message about the point above; see attachment below. > Hopefully this one is good to go :) > > You are the author of this patch and you did all the heavy lifting (thanks > again for that), so if you have no further changes/comments I’d suggest you > submit this as a v2. I’ll ACK it then (please CC-me - this should be > automatic with my SoB-line added). > > Thanks, > Thibaut > _______________________________________________ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
