Luigi, I've been bitten by your latest changes to ``struct ifnet'' in the following not-so-uncommon way. ;) I've been testing the polling(4) support for the vr(4) driver the whole day yesterday, and it worked pretty well. Today in the morning I've recompiled the if_vr.ko module once again, and it paniced my system right away. I've narrowed down the problem to the following: my NFS-mounted source tree gets updated regularly, and it has already picked up your changes to ``struct ifnet'' (to remove unused fields). Since I recompiled only one module, and attempted to run it, it became incompatible with the running kernel (which still has the old ``struct ifnet'' layout). As such, could you please send a HEADS UP to -current saying that the change introduces backward incompatibility, and recompile of all network modules is required to catch up with these changes. Also, an entry in src/UPDATING would be in order too. Perhaps bumping a __FreeBSD_version on behalf of this change would be a good thing too. (I'm not sure about that last one.) A real solution would be to have kernel somehow check that modules are ABI compatible with it, and refuse to load those that aren't. I'm not sure how to best implement this though. Cheers, -- Ruslan Ermilov ru_at_FreeBSD.org FreeBSD committer
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:37:50 UTC