On Mon, Jan 20, 2025 at 10:12:20AM +0100, Alexandre Ferrieux wrote: > Hi Dan, > > On 20/01/2025 06:27, Dan Carpenter wrote: > > > > Thanks Alexandre for the bug report. It looks like you're CC'ing a > > bunch of networking people because you're debugging something networking > > related but the actual bug is in read_kcore_iter() so let's CC Lorenzo > > instead. > > Yes, sorry for cross-posting so widely. Of course this issue had no link with > networking, but I was in "RC7 panic", fearing a 6.13 final would be cut out in a > completely un-debuggable state, and the generic LKML is so crowded I guess it's > easy to miss a needle. > > Anyway, Lorenzo did react with record speed ! > Thanks :) > https://lore.kernel.org/lkml/3719ee8a-38ef-4aaa-aca4-b6d82df51661@xxxxxxxxxx/T/#ref533860e88ab6ed1cc632411e76178797f1135e > > TL;DR: this was a side-effect of unfinished work (not in kcore itself, rather in > the memory protections around modules AFAIU), that will shortly be reverted. The > quick fix until then is to disable ARCH_HAS_EXECMEM_ROX in Kconfig. > Yup this should have landed in 6.13, it was scheduled for an urgent fix. Therefore there will be no actual release with this problem. It's execmem ROX that's the issue, and how it interacts with removing ranges from the direct mapping with the kcore code not seeming to be aware of this and breaking. Appreciate the report and attention guys! :) Cheers, Lorenzo
