15:24 <dmart> arnd: ^ I think this is mainline with CONFIG_PREEMPT_RT=y. Does the above discussion still apply? (I've not been following it closely).

15:35 <ziy> I cannot boot 6.10 arm64 kernel in VMs. I tried virtconfig on QEMU, debian arm64 config on M1 VM. neither works. but v6.9 works. anything changed in 6.10? thanks.

15:36 <ziy> not booting means there is no printout at all, just hangs.

15:40 <maz> ziy: definitely boots correctly here.

15:40 <ziy> what config are you using?

15:40 <maz> mostly defconfig.

15:41 <ziy> what platform? I wonder if it is because I am running it on VM

15:42 <ziy> I am using pi4 and M1 VMs. no bare metal.

15:42 <maz> M1, KVM.

15:43 <maz> careful that there is some binutils breakage. if you have a recent binutils, the kernel fails to boot.

15:44 <ziy> that might explain. I am using gcc 13.3.0

15:44 <ziy> binutils is 2.42.50.20240710

15:44 <ziy> what version are you using?

15:46 <maz> ii binutils-aarch64-linux-gnu 2.40-2 arm64 GNU binary utilities, for aarch64-linux-gnu target

15:46 <ziy> thanks. will give that a try.

15:49 <maz> ziy: this might be relevant: https://sourceware.org/bugzilla/show_bug.cgi?id=31924

15:51 <ziy> thanks. that must be it.

16:10 <arnd> dmart: that would explain why you see problems with printk. https://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-rt-devel.git/log/?h=linux-6.10.y-rt-rebase has what remains of the rt patchset, including the ones that select ARCH_SUPPORTS_RT on arch/arm/ and arch/arm64, as well as the ones that make printk work

16:11 <arnd> dmart: it seems that the printk patches once more did not make it into mainline this merge window https://lore.kernel.org/lkml/CAHk-=whU_woFnFN-3Jv2hNCmwLg_fkrT42AWwxm-=Ha5BmNX4w@mail.gmail.com/

16:20 <dmart> arnd: Hmm, does it sound plausible that a pr_warn() from the hrtimer interrupt could trigger this situation? I'm not sure that there a recursive oops going on here, though we do seem to be in the middle of servicing a user page fault when the hrtimer interrupt comes in.

16:26 <arnd> dmart: possible, not sure about the exact code, but I can see this happening e.g. if pr_warn()/printk() tries to take any non-saw spinlocks and the hrtimer runs with preemption disabled

16:26 <arnd> dmart: do you have a serial console enabled? that is another possible source of problems

16:27 <arnd> s/you/the person reporting the issue/

16:53 <dmart> arnd: serial console> yes, I think that's part of the problem. The backtrace is roughly user page fault -> hrtimer interrupt -> printk -> pl011 driver -> lockdep splat

16:54 <dmart> arnd: We'll try the RT tree and see whether the delta seems to be responsible.