On Fri, Aug 09, 2024 at 12:40:39PM +0800, Herbert Xu wrote: > On Thu, Aug 08, 2024 at 12:54:10PM -0700, Linus Torvalds wrote: > > > > I don't know the crypto registration API enough to even guess at what > > the fix to break the recursion is. > > > > Herbert? > > Yes my plan is to fix this in the Crypto API and not do any recursive > loads as we used to do. > > The immediate cause of the recursive load is the self-test system > (if it is not disabled through Kconfig). The algorithm registration > does not return until after the self-test has successfully executed. > For the algorithm in question, that involves loading a fallback > algorithm which is what triggered the recursive module load. > > We had an issue when algorithms were built into the kernel, where > due to the random of registration calls, a self-test may invoke > an algorithm which is built into the kernel but not yet registered. > There it was resolved by postponing all self-tests until all > algorithms had been registered (or when an algorithm was first used, > which would trigger the self-test for that algorithm there and then). > > I will extend this to modules and let the registration return > as soon as the new algorithm can be looked up. The self-test > can then complete asynchronously. > > Russell, is it OK with you if we only resolve this in the mainline > kernel or do you want a solution that can be backported as well? That's fine - I've blacklisted the arm-aes-bs module in modprobe.conf on the affected machines. Thanks! -- RMK's Patch system: https://www.armlinux.org.uk/developer/patches/ FTTP is here! 80Mbps down 10Mbps up. Decent connectivity at last!
