Re: WARNING in kvm_age_gfn

[Oliver Upton]

Hi,

On Wed, Mar 13, 2024 at 01:02:10PM +0800, cheung wall wrote:
> Hello,
> 
> when using Syzkaller to fuzz the latest Linux Kernel arm64 version,
> the following crash

First, thank you. Glad to see folks are fuzzing KVM/arm64 now.

> was triggered on:
> 
> HEAD commit: 0dd3ee31125508cd67f7e7172247f05b7fd1753a  (tag: v6.7)

In order for these sorts of bug reports to be actionable, they really
need to be against a close-to-upstream tree. 6.8 is out now, and 6.9-rc1
is right around the corner.

Having said that, I think there might be an issue or two with your bot
because...

> git tree: upstream
> 
> console output: https://pastebin.com/raw/3NCyqPUn
> 
> kernel config: https://pastebin.com/raw/mXEg4abU
> 
> C reproducer: https://pastebin.com/raw/N0gEqk5x

This reproducer doesn't work on upstream (tested kvmarm-6.9 tag), and

> WARNING: CPU: 0 PID: 711 at arch/arm64/kvm/mmu.c:1592
> kvm_age_gfn+0x64/0xdc
> data/embfuzz/emblinux/linux-4a61839152cc3e9e00ac059d73a28d148d622b30/arch/arm64/kvm/mmu.c:1592

The SHA1 in this path is suspiciously the same as 6.1.61. Are you
certain the bot is working on the correct tree? Nevertheless, I think
the WARN and associated bug were addressed in commit df6556adf27b
("KVM: arm64: Correctly handle page aging notifiers for unaligned
memslot").

Last thing -- I'm guessing your syzkaller runs will dredge up quite a bit
of low-hanging bugs given how little exposure this code has had. Could
you please aggregate the first batch of reports and send as a single
series of email? Makes it much easier to keep track of.

-- 
Thanks,
Oliver